gergely/gergelypolonkai-web-jekyll
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Convert the whole site to use Pelican instead of Jekyll

Browse Source
This commit is contained in:
Gergely Polonkai
2019-11-05 06:21:56 +01:00
parent 49961a3007
commit d5c1c942f0
534 changed files with 7315 additions and 6642 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
content/blog/2011-05-12-ethical-hacking-2011.rst Normal file
View File

@@ -0,0 +1,25 @@
Ethical Hacking 2012
####################
:date: 2011-05-12T20:54:42Z
:category: blog
:tags: conference
:url: blog/2011/5/12/ethical-hacking-2011.html
:save_as: blog/2011/5/12/ethical-hacking-2011.html
:status: published
:author: Gergely Polonkai
Today I went to the Ethical Hacking conference with my boss. It was my first appearance at such
conferences, but I hope there will be more. Although we just started to redesign our IT security
infrastructure with a 90% clear goal, it was nice to hear that everything is vulnerable. I was
thinking if we should sell all our IT equipments, fire all our colleagues (you know, to prevent
social engineering), and move to the South Americas to herd llamas or sheep, so the only danger
would be some lurking pumas or jaguars. Or I simply leave my old background image on my desktop,
from the well-known game, which says: Trust is a weakness.
Anyways, the conference was really nice. We heard about the weaknesses of Android, Oracle, and
even FireWire. They showed some demos about everything, exploited some free and commercial
software with no problem at all. We have seen how much power the virtualisation admin has
(although I think it can be prevented, but I’m not sure yet). However, in the end, we could see
that the Cloud is secure (or at least it can be, in a few months or so), so I’m not totally
pessimistic. See you next time at Hacktivity!

75
content/blog/2011-05-12-gentoo-hardened-desktop-with-gnome-3-round-one.rst Normal file
View File

@@ -0,0 +1,75 @@
Gentoo hardened desktop with GNOME 3 – Round one
################################################
:date: 2011-05-12T20:32:41Z
:url: blog/2011/5/12/gentoo-hardened-desktop-with-gnome-3-round-one.html
:save_as: blog/2011/5/12/gentoo-hardened-desktop-with-gnome-3-round-one.html
:category: blog
:tags: gentoo,gnome3,selinux
:author: Gergely Polonkai
:status: published
After having some hard times with Ubuntu (upgrading from 10.10 to 11.04), I decided to switch back
to my old friend, Gentoo. As I’m currently learning about Linux hardening, I decided to use the
new SELinux profile, which supports the v2 reference policy.
Installation was pretty easy, using the `Gentoo x86 Handbook
<http://www.gentoo.org/doc/hu/handbook/handbook-x86.xml>`_. This profile automatically turns on
the ``USE=selinux`` flag (so does the old SELinux profile), but deprecated ``FEATURE=loadpolicy``
(which is turned on by the profile, so portage will complain about it until you disable it in
``/etc/make.conf``).
For the kernel, I chose ``hardened-sources-2.6.37-r7``. This seems to be recent enough for my
security testing needs. I turned on both SELinux, PaX and grsecurity. So far, I have no problem
with it, but I don’t have X installed yet, which will screw up things for sure.
After having those hard times with Ubuntu mentioned before, I decided not to install Grub2 yet, as
it renders things unusable (eg. my Windows 7 installation, which I sometimes need at the office).
So I installed Grub 0.97 (this is the only version marked as stable, as I remember), touched
``/.autorelabel``, and reboot.
My first mistake was using an UUID as the root device on the kernel parameter list (I don’t want
to list all the small mistakes like forgetting to include to correct SATA driver from my kernel
and such). Maybe I was lame, but after including ``/dev/sda5`` instead of the UUID thing, it
worked like…
Well, charm would not be the good word. For example, I forgot to install the ``lvm2`` package, so
nothing was mounted except my root partition. After I installed it with the install CD, I assumed
everything will be all right, but I was wrong.
``udev`` and LVM is a critical point in a hardened environment. ``udev`` itself doesn’t want to
work without the ``CONFIG_DEVFS_TEMPFS=y`` kernel option, so I also had to change that. It seemed
that it can be done without the install CD, as it compiled the kernel with no problems. However,
when it reached the point when it compresses the kernel with gzip, it stopped with a ``Permission
denied`` message (although it was running with root privileges).
The most beautiful thing in the hardened environment with Mandatory Access Control enabled is that
root is not a real power user any more by default. You can get this kind of messages many times.
There are many tools to debug these, I will talk about these later.
So, my ``gzip`` needed a fix. After digging a bit on the Internet, I found that the guilty thing
is text relocation, which can be corrected if ``gzip`` is compiled with PIC enabled. Thus, I
turned on ``USE=pic`` flag globally, and tried to remerge gzip. Of course it failed, as it had to
use gzip to unpack the gzip sources. So it did when I tried to install the PaX tools and
``gradm`` to turn these checks off. The install CD came to the rescue again, with which I
successfully recompiled gzip, and with this new gzip, I compressed my new kernel, with which udev
started successfully. So far, so good, let’s try to reboot!
Damn, LVM is still not working. So I decided to finally consult the Gentoo hardened guide. It
says that the LVM startup scripts under ``/lib/rcscripts/…`` must be modified, so LVM will put its
lock files under ``/etc/lvm/lock`` instead of ``/dev/.lvm``. After this step and a reboot, LVM
worked fine (finally).
The next thing was the file system labelling. SELinux should automatically relabel the entire
file system at boot time whenever it finds the ``/.autorelabel`` file. Well, in my case it didn’t
happen. After checking the `Gentoo Hardening <http://wiki.gentoo.org/wiki/Hardened_Gentoo>`_
docs, I realised that the ``rlpkg`` program does exactly the same (as far as I know, it is
designed specifically for Gentoo). So I ran ``rlpkg``, and was kind of shocked. It says it will
relabel ext2, ext3, xfs and JFS partitions. Oh great, no ext4 support? Well, consulting the
forums and adding some extra lines to ``/etc/portage/package.keywords`` solved the problem
(``rlpkg`` and some dependencies had to have the ``~x86`` keyword set). Thus, ``rlpkg``
relabelled my file systems (I checked some directories with ``ls -lZ``, it seemed good for me).
Now it seems that everything is working fine, except the tons of audit messages. Tomorrow I will
check them with ``audit2why`` or ``audit2allow`` to see if it is related with my SELinux lameness,
or with a bug in the policy included with Gentoo.

32
content/blog/2011-05-13-zabbix-performance-tip.rst Normal file
View File

@@ -0,0 +1,32 @@
Zabbix performance tip
######################
:date: 2011-05-13T19:03:31Z
:category: blog
:tags: zabbix,monitoring
:url: blog/2011/5/13/zabbix-performance-tip.html
:save_as: blog/2011/5/13/zabbix-performance-tip.html
:status: published
:author: Gergely Polonkai
Recently I have switched from `MRTG <http://oss.oetiker.ch/mrtg/>`_ + `Cacti
<http://www.cacti.net/>`_ + `Nagios <http://www.nagios.org/>`_ + `Gnokii
<http://www.gnokii.org/>`_ to `Zabbix <http://www.zabbix.com/>`_, and I must say I’m more than
satisfied with it. It can do anything the former tools did, and much more. First of all, it can
do the same monitoring as Nagios did, but it does much more fine. It can check several parameters
within one request, so network traffic is kept down. Also, its web front-end can generate any
kinds of graphs from the collected data, which took Cacti away. Also, it can do SNMP queries
(v1-v3), so querying my switches’ port states and traffic made easy, taking MRTG out of the
picture (I know Cacti can do it either, it had historical reasons we had both tools installed).
And the best part: it can send SMS messages via a GSM modem natively, while Nagios had to use
Gnokii. The trade-off is, I had to install Zabbix agent on all my monitored machines, but I think
it worths the price. I even have had to install NRPE to monitor some parameters, which can be a
pain on Windows hosts, while Zabbix natively supports Windows, Linux and Mac OS/X.
So I only had to create a MySQL database (which I already had for NOD32 central management), and
install Zabbix server. Everything went fine, until I reached about 1300 monitored parameters.
MySQL seemed to be a bit slow on disk writes, so my Zabbix “queue” filled up in no time. After
reading some forums, I decided to switch to PostgreSQL instead. Now it works like charm, even
with the default Debian settings. However, I will have to add several more parameters, and my
boss wants as many graphs as you can imagine, so I’m more than sure that I will have to fine tune
my database later.

26
content/blog/2011-05-18-gentoo-hardened-desktop-with-gnome-3-round-two.rst Normal file
View File

@@ -0,0 +1,26 @@
Gentoo hardened desktop with GNOME 3 – Round two
################################################
:date: 2011-05-18T10:28:14Z
:category: blog
:tags: gentoo,gnome3,selinux
:url: blog/2011/5/18/gentoo-hardened-desktop-with-gnome-3-round-two.html
:save_as: blog/2011/5/18/gentoo-hardened-desktop-with-gnome-3-round-two.html
:status: published
:author: Gergely Polonkai
After several hours of ``package.keywords``/``package.use`` editing and package compiling, I
managed to install GNOME 3 on my notebook. Well, I mean, the GNOME 3 packages. Unfortunately the
``fglrx`` driver didn’t seem to recognise my ATI Mobility M56P card, and the open source driver
didn’t want to give me GLX support. When I finally found some clues on what should I do, I had to
use my notebook for work, so I installed Fedora 14 on it. Then I realised that GNOME 3 is already
included in Rawhide (Fedora 15), so I quickly downloaded and installed that instead. Now I have
to keep this machine in a working state for a few days, so I will learn SELinux stuff in its
native environment.
When I installed Fedora 14, the first AVC message popped up after about ten minutes. That was a
good thing, as I wanted to see ``setroubleshoot`` in action. However, in Fedora 15, the AVC
bubbles didn’t show up even after a day. I raised my left eyebrow and said that’s impossible,
SELinux must be disabled. And it’s not! It’s even in enforcing mode! And it works just fine. I
like it, and I hope I will be able to get the same results with Gentoo if I can get back to
testing…

34
content/blog/2011-05-27-citrix-xenserver-vs-debian-5-0-upgrade-to-6-0.rst Normal file
View File

@@ -0,0 +1,34 @@
Citrix XenServer 5.5 vs. Debian 5.0 upgrade to 6.0
##################################################
:date: 2011-05-27T17:33:41Z
:category: blog
:tags: citrix-xenserver,debian
:url: blog/2011/5/27/citrix-xenserver-vs-debian-5-0-upgrade-to-6-0.html
:save_as: blog/2011/5/27/citrix-xenserver-vs-debian-5-0-upgrade-to-6-0.html
:status: published
:author: Gergely Polonkai
Few weeks ago I’ve upgraded two of our Debian based application servers from 5.0 to 6.0.
Everything went fine, as the upgraded packages worked well with the 4.2 JBoss instances. For the
new kernel we needed a reboot, but as the network had to be rebuilt, I postponed this reboot until
the network changes. With the network, everything went fine again, we successfully migrated our
mail servers behind a firewall. Also the Xen server (5.5.0, upgrade to 5.6 still has to wait for
a week or so) revolted well with some storage disks added. But the application servers remained
silent…
After checking the console, I realised that they don’t have an active console. And when I tried
to manually start them, XenServer refused with a message regarding ``pygrub``.
To understand the problem, I had to understand how XenServer boots Debian. It reads the
``grub.conf`` on the first partition’s root or ``/boot`` directory, and starts the first option,
without asking (correct me, if I’m mistaken somewhere). However, this ``pygrub`` thing can not
parse the new, grub2 config. This is kinda frustrating.
For the first step, I quickly installed a new Debian 5.0 system from my template. Then I attached
the disks of the faulty virtual machine, and mounted all its partitions. This way I could reach
my faulty 6.0 system with a chroot shell, from which I could install the ``grub-legacy`` package
instead of grub, install the necessary kernel and XenServer tools (which were missing from both
machines somehow), then halt the rescue system, and start up the original instance.
Next week I will do an upgrade on the XenServer to 5.6.1. I hope no such problems will occur.

23
content/blog/2011-05-27-oracle-database-incompatible-with-oracle-linux.rst Normal file
View File

@@ -0,0 +1,23 @@
Oracle Database “incompatible” with Oracle Linux?
#################################################
:date: 2011-05-27T17:53:31Z
:category: blog
:tags: linux,oracle
:url: blog/2011/5/27/oracle-database-incompatible-with-oracle-linux.html
:save_as: blog/2011/5/27/oracle-database-incompatible-with-oracle-linux.html
:status: published
:author: Gergely Polonkai
Today I gave a shot to install `Oracle Linux
<http://www.oracle.com/us/technologies/linux/overview/index.html>`_. I thought I could easily
install an Oracle DBA on it. Well, I was naive.
As only the 5.2 version is supported by XenServer 5.5, I downloaded that version of Oracle Linux.
Installing it was surprisingly fast and easy, it asked almost nothing, and booted without any
problems.
After this came the DBA, 10.2, which bloated an error message in my face saying that this is an
unsupported version of Linux. Bah.
Is it only me, or is it really strange that Oracle doesn’t support their own distro?

29
content/blog/2011-06-10-proxy-only-non-existing-files-with-mod-proxy-and-mod-rewrite.rst Normal file
View File

@@ -0,0 +1,29 @@
Proxy only non-existing files with mod_proxy and mod_rewrite
############################################################
:date: 2011-06-10T14:20:43Z
:category: blog
:tags: apache
:url: blog/2011/6/10/proxy-only-non-existing-files-with-mod-proxy-and-mod-rewrite.html
:save_as: blog/2011/6/10/proxy-only-non-existing-files-with-mod-proxy-and-mod-rewrite.html
:status: published
:author: Gergely Polonkai
Today I got an interesting task. I had to upload some pdf documents to a site. The domain is
ours, but we don’t have access to the application server that is hosting the page yet. Until we
get it in our hands, I did a trick.
I enabled `mod_rewrite`, `mod_proxy` and `mod_proxy_http`, then added the following lines to my
apache config:
.. code-block:: apache
RewriteEngine on
RewriteRule ^/$ http://172.16.72.131:8080/ [QSA,L,P]
RewriteCond %{REQUEST_FILENAME} !-f
RewriteRule ^/(.*) http://172.16.72.131:8080/$1 [QSA,L,P]
Order allow,deny
Allow from all
I’m not totally sure it’s actually secure, but it works for now.

26
content/blog/2011-09-18-inverse-of-sort.rst Normal file
View File

@@ -0,0 +1,26 @@
Inverse of `sort`
#################
:date: 2011-09-18T14:57:31Z
:category: blog
:tags: linux,command-line
:url: blog/2011/9/18/inverse-of-sort.html
:save_as: blog/2011/9/18/inverse-of-sort.html
:status: published
:author: Gergely Polonkai
I’m using \*NIX systems for about 14 years now, but it can still show me new things. Today I had
to generate a bunch of random names. I’ve create a small perl script which generates permutations
of some usual Hungarian first and last names, occasionally prefixing it with a ‘Dr.’ title or
using double first names. For some reasons I forgot to include uniqueness check in the script.
When I ran it in the command line, I realized the mistake, so I appended ``| sort | uniq`` to the
command line. So I had around 200 unique names, but in alphabetical order, which was awful for my
final goal. Thus, I tried shell commands like ``rand`` to create a random order, and when many of
my tries failed, the idea popped in my mind (not being a native English speaker): “I don’t have to
create »random order«, but »shuffle the list«. So I started typing ``shu``, pressed Tab in the
Bash shell, and voilà! ``shuf`` is the winner, it does just exactly what I need:
**NAME**
shuf - generate random permutations
Thank you, Linux Core Utils! :)

14
content/blog/2011-12-11-why-you-should-always-test-your-software-with-production-data.rst Normal file
View File

@@ -0,0 +1,14 @@
Why you should always test your software with production data
#############################################################
:date: 2011-12-11T12:14:51Z
:category: blog
:tags: development,testing,ranting
:url: blog/2011/12/11/why-you-should-always-test-your-software-with-production-data.html
:save_as: blog/2011/12/11/why-you-should-always-test-your-software-with-production-data.html
:status: published
:author: Gergely Polonkai
I’m writing a software for my company in PHP, using the Symfony 2 framework. I’ve finished all
the work, created some sample data, it loaded perfectly. Now I put the whole thing into
production and tried to upload the production data into it. Guess what… it didn’t load.

25
content/blog/2012-03-20-php-5-4-released.rst Normal file
View File

@@ -0,0 +1,25 @@
PHP 5.4 released
################
:date: 2012-03-20T13:31:12Z
:category: blog
:tags: php
:url: blog/2012/3/20/php-5-4-released.html
:save_as: blog/2012/3/20/php-5-4-released.html
:status: published
:author: Gergely Polonkai
After a long time of waiting, PHP announced 5.4 release on 1 March (also, today they announced
that they finally migrate to Git, which is sweet from my point of view, but it doesn’t really
matter).
About a year ago we became very agressive towards a developer who created our internal e-learning
system. Their database was very insecure, and they didn’t really follow industry standards in
many ways. Thus, we forced them to move from Windows + Apache 2.0 + PHP 5.2 + MySQL 4.0 to Debian
Linux 6.0 + Apache 2.2 + PHP 5.3 + MySQL 5.1. It was fun (well, from our point of view), as their
coders… well… they are not so good. The code that ran “smoothly” on the old system failed at many
points on the new one. So they code and code, and write more code. And they still didn’t finish.
And now 5.4 is here. Okay, I know it will take some time to get into the Debian repositories, but
it’s here. And they removed ``register_globals``, which will kill that funny code again at so
many points that they will soon get to rewrite the whole code to make it work. And I just sit
here in my so-much-comfortable chair, and laugh. Am I evil?

29
content/blog/2012-03-27-fast-world-fast-updates.rst Normal file
View File

@@ -0,0 +1,29 @@
Fast world, fast updates
########################
:date: 2012-03-27T06:18:43Z
:category: blog
:tags: linux
:url: blog/2012/3/27/fast-world-fast-updates.html
:save_as: blog/2012/3/27/fast-world-fast-updates.html
:status: published
:author: Gergely Polonkai
We live in a fast world, that’s for sure. When I first heard about Ubuntu Linux and their goals,
I was happy: they gave a Debian to everyone, but in different clothes. It had fresh software in
it, and even they gave support of a kind. It was easy to install and use, even if one had no
Linux experience before. So people liked it. I’ve even installed it on some of my servers
because of the new package versions that came more often. Thus I got an up to date system.
However, it had a price. After a while, security updates came more and more often, and when I had
a new critical update every two or three days, I’ve decided to move back to Debian. Fortunately I
did this at the time of a new release, so I didn’t really loose any features.
After a few years passed, even Debian is heading this very same way. But as I see, the cause is
not the same. It seems that upstream software is hitting these bugs, and even the Debian guys
don’t have the time to check for them. At the time of a GNOME version bump (yes, GNOME 3 is a
really big one for the UN\*X-like OSes), when hundreds of packages need to be checked, security
bugs show off more often. On the other hand however, Debian is releasing a new security update
every day (I had one on each of the last three days). This, of course, is good from one point of
view as we get a system that is more secure, but most administrators don’t have maintenance
windows this often. I can think of some alternatives like Fedora, but do I really have to change?
Dear fellow developers, please code more carefully instead!

23
content/blog/2012-06-14-wordpress-madness.rst Normal file
View File

@@ -0,0 +1,23 @@
Wordpress madness
#################
:date: 2012-06-14T06:40:12Z
:category: blog
:tags: wordpress,ranting
:url: blog/2012/6/14/wordpress-madness.html
:save_as: blog/2012/6/14/wordpress-madness.html
:status: published
:author: Gergely Polonkai
I’m a bit fed up that I had to install `MySQL <http://www.mysql.com/>`_ on my server to have
`Wordpress <http://wordpress.org/>`_ working, so I’ve Googled a bit to find a solution for my
pain. I found `this <http://codex.wordpress.org/Using_Alternative_Databases>`_. I don’t know
when this post was written, but I think it’s a bit out of date. I mean come on, PDO is the part
of PHP for ages now, and they say adding a DBAL to the dependencies would be a project as large as
(or larger than) WP itself. Well, yes, but PHP is already a dependency, isn’t it? Remove it
guys, it’s too large!
Okay, to be serious… Having a heavily MySQL dependent codebase is a bad thing in my opinion, and
changing it is no easy task. But once it is done, it would be a child’s play to keep it up to
date, and to port WP to other database backends. And it would be more than enough to call it 4.0,
and raising version numbers fast is a must nowadays (right, Firefox and Linux Kernel guys?)

25
content/blog/2012-06-18-ssh-login-failed-on-red-hat-enterprise-linux-6-2.rst Normal file
View File

@@ -0,0 +1,25 @@
SSH login FAILed on Red Had Enterprise Linux 6.2
################################################
:date: 2012-06-18T18:28:45Z
:category: blog
:tags: linux,selinux,ssh,red-hat
:url: blog/2012/6/18/ssh-login-failed-on-red-hat-enterprise-linux-6-2.html
:save_as: blog/2012/6/18/ssh-login-failed-on-red-hat-enterprise-linux-6-2.html
:status: published
:author: Gergely Polonkai
Now this was a mistake I should not have done…
About a month ago I have moved my AWS EC2 machine from Amazon Linux to RHEL 6.2. This was good.
I have moved all my files and stuff, recreated my own user, everything was just fine. Then I
copied my `gitosis <https://github.com/tv42/gitosis>`_ account (user ``git`` and its home
directory). Then I tried to log in. It failed. I was blaming OpenSSH for a week or so, changed
the config file in several ways, tried to change the permissions on ``~git/.ssh/*``, but still
nothing. Permission were denied, I was unable to push any of my development changes. Now after a
long time of trying, I coincidently ``tail -f``-ed ``/var/log/audit/audit.log`` (wanted to open
``auth.log`` instead) and that was my first good point. It told me that ``sshd`` was unable to
read ``~git/.ssh/authorized_keys``, which gave me the idea to run ``restorecon`` on ``/home/git``.
It solved the problem.
All hail SELinux and RBAC!

31
content/blog/2012-06-22-upgrades-requiring-a-reboot-on-linux-at-last.rst Normal file
View File

@@ -0,0 +1,31 @@
Upgrades requiring a reboot on Linux? At last!
##############################################
:date: 2012-06-22T20:04:51Z
:category: blog
:tags: linux
:url: blog/2012/6/22/upgrades-requiring-a-reboot-on-linux-at-last.html
:save_as: blog/2012/6/22/upgrades-requiring-a-reboot-on-linux-at-last.html
:status: published
:author: Gergely Polonkai
I’ve recently received an article on Google+ about Fedora’s new idea: package upgrades that
require a reboot. The article said that Linux guys have lost their primary adoo: “Haha! I don’t
have to reboot my system to install system upgrades!” My answer was always this: “Well, actually
you should…”
I think this can be a great idea if distros implement it well. PackageKit was a good first step
on this road. That software could easily solve such an issue. However, it is sooo easy to do it
wrong. The kernel, of course, can not be upgraded online (or could it be? I have some theories on
this subject, wonder if it can be implemented…), but other packages are much different. From the
users’ point of view the best would be if the packages would be upgraded in the background
seemlessly. E.g. PackageKit should check if the given executable is running. If not, it should
upgrade it, while notifying the user like “Hey dude, don’t start Anjuta now, I’m upgrading it!”,
or simply denying to start it. Libraries are a bit different, as PackageKit should check if any
running executables are using the library. Meanwhile, PK should also keep a notification
somewhere telling the users that some packages could be upgraded, but without stopping
this-and-that, it can not be done.
I know these things are easier said than done. But I think (a) users should tell such ideas to
the developers and (b) developers (mostly large companies, like Microsoft or Apple) should listen
to them, and at least think of these ideas. Some users are not as stupid as they think…

67
content/blog/2012-09-05-some-thoughts-about-that-dead-linux-desktop.rst Normal file
View File

@@ -0,0 +1,67 @@
Some thoughts about that dead Linux Desktop
###########################################
:date: 2012-09-05T09:01:31Z
:category: blog
:tags: linux,ranting
:url: blog/2012/9/5/some-thoughts-about-that-dead-linux-desktop.html
:save_as: blog/2012/9/5/some-thoughts-about-that-dead-linux-desktop.html
:status: published
:author: Gergely Polonkai
There were some arguments in the near past on `What Killed the Linux Desktop
<http://tirania.org/blog/archive/2012/Aug-29.html>`_. After reading many replies, like `Linus
Torvalds’s
<http://www.zdnet.com/linus-torvalds-on-the-linux-desktops-popularity-problems-7000003641/>`_, I
have my own thoughts, too.
I know my place in the world, especially in the online community. I’m a Linux user for about 15
years and a Linux administrator for 10 years now, beginning with WindowMaker and something that I
remember as GNOME without a version number. I have committed some minor code chunks and
translations in some minor projects, so I’m not really into it from the “write” side (well, until
now, since I have began to write this blog, and much more, but don’t give a penny for my words
until you see it).
I’m using Linux since 2.2 and GNOME since 1.whatever. It’s nice that a program compiled years ago
still runs on today’s Linux kernel, especially if you see old DOS/Windows software failing to
start on a new Windows 7 machine. I understand Linus’ point that breaking external APIs is bad,
and I think it can work well on the kernel’s level. But the desktop level is much different. As
the Linux Desktop has such competitors (like OS/X and Windows’ Aero and Metro), they have to give
something new to the users almost every year to keep up with them. Eye candies are a must (yes,
of course my techy fellows, they are worthless, but users *need* it), and they can not be created
without extending APIs. And the old API… well, it fades away fast. I don’t really understand
however, why they have to totally disappear, like `GTK_DIALOG_NO_SEPARATOR
<http://developer.gnome.org/gtk/stable/GtkDialog.html#GtkDialogFlags>`_ in Gtk+3. It could be
replaced with a 0 value (e.g: it won’t do anything). This way my old Gtk+2 program could compile
with Gtk+3 nicely. Also, there could be a small software that goes through your source code and
warn you about such deprecated (and no-doer but still working) things. Porting applications
between Gtk+ (and thus, GNOME) versions became a real pain, which makes less enthusiast
programmers stop developing for Linux. Since I’m a GNOME guy for years, I can tell nothing about
Qt and KDE, but for the GNOME guys, this is a bad thing. As of alternatives, there is Java. No,
wait… it turned out recently that `it has several security bugs
<http://www.theregister.co.uk/2012/08/31/critical_flaw_found_in_patched_java>`_. Also it’s not
that multiplatform as they say (I can’t find the article on that at the moment, but I have proof).
Also, the JVMs out there eat up so much resources, which makes it a bit hard and expensive to use.
Also, I see another problem: those blasted package managers. RPM, DPKG, Portage, whatever. What
the hell? Why are there so many? Why do developers reinvent the wheel? The nave is too small or
there are to few spokes? Come on… we live in an open source world! Contribute to the one and
only package manager (which one is that I don’t actually care)! I’m sure the two (three, many)
bunches of develoeprs could make a deal. Thus, it could become better and “outsider” companies
would be happier to distribute their software for Linux platforms.
And now that we get to the big companies. I don’t really understand them. nVidia and ATI made
their own closed source drivers for Linux. Some other hardware vendors also write Linux drivers,
and as the kernel API doesn’t really change, they will work for a long time. But what about
desktop application vendors? Well, they try to stick to a desktop environment or two, and if they
change too frequently, they stop developing for Linux, like Skype did (OK, maybe Skype has other
reasons, but you see my point). But why? The main part for Linux programs is the Linux kernel
and the basic userland like libc/stdlib++. If you write graphical software, it will have to use
X-Windows. Yes, it’s much different in many ways, mostly because they have a… well… pretty ugly
design by default. But still, it’s the same on every Linux distributions, as it became somewhat
an industry standard, as it was already on the market back in the old UN\*X days. The protocol
itself changed just like the Linux kernel: almost no change at all, just some new features.
So what kills the Linux desktop in my opinion is these constant wars inside, and the lack of
support from the outside. Open Source is good, but until these (mostly the first) problems are
not resolved, Linux Desktop can do nothing on the market. It’s a downward spiral hard to escape.

73
content/blog/2012-09-07-how-to-start-becoming-a-web-developer.rst Normal file
View File

@@ -0,0 +1,73 @@
How to start becoming a web developer
#####################################
:date: 2012-09-07T18:12:12Z
:category: blog
:tags: development,technology
:url: blog/2012/9/7/how-to-start-becoming-a-web-developer.html
:save_as: blog/2012/9/7/how-to-start-becoming-a-web-developer.html
:status: published
:author: Gergely Polonkai
A friend of mine asked me today how to become a web developer. It took me a while, but I made up
a checklist. It’s short, but it’s enough for the first steps.
First of all, learn English
===========================
Well, if you read this, maybe this was a bad first point…
Choose a language and stick to it!
==================================
For the UN\*X/Linux line, there is PHP. It’s free, easy to learn, and has many free tools and
documentations available. It can be used in a functional or an object-oriented way.
C# is another good way to start, but for the Windows line. It’s fully object oriented, and the
web is full of tutorials, how-tos and other resources.
Learn the basics of the system you are working on
=================================================
To become a good developer, learn at least the basics of the system you are working on. Basic
commands can always come in handy. Debugging (yes, you will do tons of bugs for sure) can become
much easier if you know the huge set of tools provided by your OS. You should also try to develop
in the chosen environment. Chose PHP? Get a Linux desktop! ASP.NET? Get a Windows. Everything
will be much easier!
Learn the basics of the web server you are using
================================================
PHP can run on `Apache <http://httpd.apache.org/>`_ (as a module), or any CGI-capable webserver,
like `lighttpd <http://www.lighttpd.net/>`_ or `nginx <http://nginx.org/>`_ (well, it can also run
on IIS, but trust me: you don’t want that). ASP.NET is designed for IIS, and although some
scripts can be run under a mono-capable server, it should still stay there.
Whichever you choose, learn the basics! How to start and stop the service,
basic configuration methods, modules/extensions, and so on. It’s more than sure
that you will face some issues while developing, so it can never hurt.
Keep your versions under control
================================
Version control is critical nowadays. It gives you a basic backup solution, can come in handy
with debugging, and if you ever want to work in a team, you will badly need it.
Subversion is a bit out of date now, and it’s kind of hard to set up.
Git is no easy. You will have to learn a lot of stuff, but basicly it’s just another version
control system. Just choose if you want to stick to merge-then-commit or rebase-then-commit, get
a client, and get on the run.
Microsoft’s Team Foundation is another good way if you are working in a team. It provides several
other features besides version controlling, and is well integrated into Visual Studio, which is
highly recommended for Windows based development.
Choose an environment to work in
================================
There are so many good tools out there. You should choose according to the language and OS on
what you are working on. `Zend Studio <http://www.zend.com/en/products/studio>`_ or `Netbeans
<https://netbeans.org/>`_ are both good tools for PHP development, while `Visual Studio
<http://www.visualstudio.com/>`_ is a best buy for Windows development. Both of these have many
ups and downs, but once you get in touch with their deeper parts, you will like them.

17
content/blog/2012-09-10-do-not-track-in-ie10-vs-apache.rst Normal file
View File

@@ -0,0 +1,17 @@
Do-Not-Track in IE10 vs. Apache
###############################
:date: 2012-09-10 20:22:32Z
:category: blog
:tags: apache,technology
:url: blog/2012/9/10/do-not-track-in-ie10-vs-apache.html
:save_as: blog/2012/9/10/do-not-track-in-ie10-vs-apache.html
:status: published
:author: Gergely Polonkai
`Apache developer decided not to accept Do-Not-Track headers from IE10 users
<http://arstechnica.com/security/2012/09/apache-webserver-updated-to-ignore-do-not-track-settings-in-ie-10/>`_,
because it’s enabled by default. So… if I install a plugin that hides the fact from the web
server that I’m using IE10, I become eligible of using it. But if I do this, I simply became
eligible because I consciously installed that addon, so I could actually use it without hiding the
fact. Sorry if I’m a bit Philosoraptorish…

74
content/blog/2012-09-16-symfony-2-create-role-and-class-based-acls-with-your-roles-coming-from-the-orm.rst Normal file
View File

@@ -0,0 +1,74 @@
Symfony 2 – Create role- and class-based ACLs with your roles coming from the ORM
#################################################################################
:date: 2012-09-16T18:39:25Z
:category: blog
:tags: php,symfony
:url: blog/2012/9/16/symfony-2-create-role-and-class-based-acls-with-your-roles-coming-from-the-orm.html
:save_as: blog/2012/9/16/symfony-2-create-role-and-class-based-acls-with-your-roles-coming-from-the-orm.html
:status: published
:author: Gergely Polonkai
During the last weeks I had some serious issues with one of my private Symfony 2 projects. One of
my goals was to create a dynamic security system, e.g my administrators wanted to create roles,
and grant these roles access to different object types (classes) and/or objects.
So I have created a ``User`` entity, which implements ``UserInterface`` and
``AdvancedUserInterface``, the latter for the possibility to enable/disable accounts and such. It
had a ``$roles`` property, which was a ``ManyToMany`` relation to the ``Role`` entity, which
implemented ``RoleInterface``. Also I have created my own role hierarchy service that implements
``RoleHierarchyInterface``.
So far so good, first tests. It soon turned out that if ``User::getRoles()`` returns a
``DoctrineCollection`` as it does by default, then the standard
.. code-block:: php
<?php
$this->get('security.context')->isGranted('ROLE_ADMIN');
doesn’t work. I know, it should not be hard coded, as my roles and permission tables are dynamic,
I have just tested. So I fixed my ``User`` entity so ``getRoles()`` returns an array of ``Role``
objects instead of the ``DoctrineCollection``. Also I implemented a ``getRolesCollection()``
method to return the original collection, but I think it will never be used.
After that, I had to implement some more features so I put this task away. Then, I tried to
create my first ACL.
.. code-block:: php
<?php
$securityIdentity = new RoleSecurityIdentity('ROLE_ADMIN');
$objectIdentity = new ObjectIdentity('newsClass', 'Acme\\DemoBundle\\Entity\\News');
$acl = $aclProvider->createAcl($objectIdentity);
$acl->insertClassAce($securityIdentity, MaskBuilder::MASK_OWNER);
$aclProvider->updateAcl($acl);
I was about to check if the user who is logged in has an ``OWNER`` permission on the ``User``
class.
.. code-block:: php
<?php
$this->objectIdentity = new ObjectIdentity(self::OBJECT_ID, self::OBJECT_FQCN);
if ($this->securityContext->isGranted('OWNER', $this->objectIdentity) === false) {
throw new AccessDeniedException('You don’t have the required permissions!');
}
The ACL was defined based on a role, so everyone who had the ``ROLE_ADMIN`` role should gain
access to the user listing page. But they didn’t. It took several weeks to find the cause, I
have put it on `stackoverflow
<http://stackoverflow.com/questions/12057795/symfony-2-1-this-getsecurity-context-isgrantedrole-admin-returns-fa>`_
and the Symfony Google Group, but no usable answers.
Then I went off for debugging. Setting up NetBeans for xdebug-based PHP debugging was real fun
under Fedora, but that’s another story. After a while I have found that Symfony’s basic access
decision manager checks for ``$role->getRole()`` only if ``$role`` is an instance of
``Symfony\Component\Security\Core\Role\Role``, instead of checking if the object implements
``Symfony\Component\Security\Core\Role\RoleInterface``. So I’ve checked if the bug is already
reported. It turned out that it was, and my solution was available in a specific commit about a
year ago, but as `Johannes Schmitt commented, it would introduce a security issue
<https://github.com/symfony/symfony/commit/af70ac8d777873c49347ac828a817a400006cbea>`_, so it was
reverted. Unfortunately neither Johannes Schmitt, nor Fabien Potencier (nor anyone else) could (or
wanted) to tell about this issue. So the final (and somewhat hack-like) solution was to extend
``Symfony\Component\Security\Core\Role\Role``. And boom! It worked.

36
content/blog/2012-10-07-smsgateway-and-smssender.rst Normal file
View File

@@ -0,0 +1,36 @@
SmsGateway and SmsSender
########################
:date: 2012-10-07T00:10:26Z
:category: blog
:tags: development,php,symfony
:url: blog/2012/10/7/smsgateway-and-smssender.html
:save_as: blog/2012/10/7/smsgateway-and-smssender.html
:status: published
:author: Gergely Polonkai
I have just uploaded my SmsGateway, SmsSender and SmsSenderBundle packages to `GitHub
<http://github.com/gergelypolonkai>`_ and `Packagist <http://packagist.org>`_. I hope some of you
will find it useful.
* SmsGateway
* `SmsGateway on GitHub`_
* `SmsGateway on Packagist`_
* SmsSender
* `SmsSender on GitHub`_
* `SmsSender on Packagist`_
* SmsSenderBundle
* `SmsSenderBundle on GitHub`_
* `SmsSenderBundle on Packagist`_
.. _SmsGateway on GitHub: https://github.com/gergelypolonkai/smsgateway
.. _SmsGateway on Packagist: https://packagist.org/packages/gergelypolonkai/smsgateway
.. _SmsSender on GitHub: https://github.com/gergelypolonkai/smssender
.. _SmsSender on Packagist: https://packagist.org/packages/gergelypolonkai/smssender
.. _SmsSenderBundle on GitHub: https://github.com/gergelypolonkai/smssender-bundle
.. _SmsSenderBundle on Packagist: https://packagist.org/packages/gergelypolonkai/smssender-bundle

22
content/blog/2012-10-13-changing-the-session-cookie-s-name-in-symfony-2.rst Normal file
View File

@@ -0,0 +1,22 @@
Changing the session cookie’s name in Symfony 2
###############################################
:date: 2012-10-13T12:49:28Z
:category: blog
:tags: symfony,development
:url: blog/2012/10/13/changing-the-session-cookie-s-name-in-symfony-2.html
:save_as: blog/2012/10/13/changing-the-session-cookie-s-name-in-symfony-2.html
:status: published
:author: Gergely Polonkai
I have a development server, on which I have several Symfony 2.x projects under the same hostname
in different directories. Now I’m facing a funny problem which is caused by that the cookies
Symfony places for each of my projects have the same name.
To change this, you will have to modify the ``config.yml`` file like this:
.. code-block:: yaml
session:
name: SiteSpecificSessionName
lifetime: 3600

52
content/blog/2012-12-20-symfony-2-configuration-array-of-associative-arrays.rst Normal file
View File

@@ -0,0 +1,52 @@
Symfony 2 Configuration – Array of associative arrays
#####################################################
:date: 2012-12-20T12:03:23Z
:category: blog
:tags: php,symfony
:url: blog/2012/12/20/symfony-2-configuration-array-of-associative-arrays.html
:save_as: blog/2012/12/20/symfony-2-configuration-array-of-associative-arrays.html
:status: published
:author: Gergely Polonkai
Few days ago I have struggled with a problem using Symfony2 configuration. I
wanted to add the following kind of configuration to ``config.yml``:
.. code-block:: yaml
acme_demo:
transitions:
- { hc_cba: 180 }
- { cba_hc: -1 }
The problem was that the stuff under ``transitions`` is dynamic, so those ``hc_cba`` and
``cba_hc`` tags can be pretty much anything. After hitting many errors, I came to the solution:
.. code-block:: php
<?php
$rootNode
->children()
->arrayNode('state_machine')
->requiresAtLeastOneElement()
->beforeNormalization()
->ifArray()
->then(function($values) {
$ret = array();
foreach ($values as $value) {
foreach ($value as $transition => $time) {
$ret[] = array('transition' => $transition, 'time' => e);
}
}
return $ret;
})
->end()
->prototype('array')
->children()
->scalarNode('transition')->end()
->scalarNode('time')->end()
->end()
->end()
->end()
->end()
;

13
content/blog/2013-01-05-development-man-pages-on-fedora.rst Normal file
View File

@@ -0,0 +1,13 @@
Development man pages on Fedora
###############################
:date: 2013-01-05T18:20:41Z
:category: blog
:tags: development,fedora
:url: blog/2013/1/5/development-man-pages-on-fedora.html
:save_as: blog/2013/1/5/development-man-pages-on-fedora.html
:status: published
:author: Gergely Polonkai
If you use Fedora (like me), and can’t find the development manual pages for e.g. ``printf(3)``
(like me), just ``yum install man-pages`` (like me).

99
content/blog/2013-01-06-registering-an-enum-type-in-glib-s-type-system.rst Normal file
View File

@@ -0,0 +1,99 @@
Registering an enum type in GLib’s type system
##############################################
:date: 2013-01-06T02:34:03Z
:category: blog
:tags: c,development,glib
:url: blog/2013/1/6/registering-an-enum-type-in-glib-s-type-system.html
:save_as: blog/2013/1/6/registering-an-enum-type-in-glib-s-type-system.html
:status: published
:author: Gergely Polonkai
I faced a problem in my `GLib <https://developer.gnome.org/glib/>`_ self-teaching project, `wMUD
<https://github.com/gergelypolonkai/wmud>`_ today. I wanted to register a signal for a
``GObject``, whose handler should accept two ``enum`` parameters for which I had to register a new
``GEnum`` type in the ``GObject`` type system. However, the `documentation on this feature
<https://developer.gnome.org/gobject/unstable/gtype-non-instantiable.html>`_ (thanks for pointing
out goes to `hashem` on ``#gnome-hackers``) is not… uhm… obvious. Making the long story short, I
have checked with the ``GIO`` sources for an example, and using that, I have created this small,
working chunk:
.. code-block:: c
#ifndef __WMUD_CLIENT_STATE_H__
#define __WMUD_CLIENT_STATE_H__
#include <glib-object.h>
/**
* WmudClientState:
* @WMUD_CLIENT_STATE_FRESH: Client is newly connected. Waiting for a login
* player name
* @WMUD_CLIENT_STATE_PASSWAIT: Login player name is entered, waiting for a
* login password
* @WMUD_CLIENT_STATE_MENU: Authentication was successful, player is now in the
* main game menu
* @WMUD_CLIENT_STATE_INGAME: Character login was successful, player is now
* in-game
* @WMUD_CLIENT_STATE_YESNO: Player was asked a yes/no question, and we are
* waiting for the answer. client.yesNoCallback MUST be set at this point!
* TODO: if wmudClient had a prevState field, and there would be some hooks
* that are called before and after the client enters a new state, this
* could be a three-state stuff, in which the player can enter e.g ? as
* the answer, so they would be presented with the question again.
* @WMUD_CLIENT_STATE_REGISTERING: Registering a new player. Waiting for the
* e-mail address to be given
* @WMUD_CLIENT_STATE_REGEMAIL_CONFIRM: E-mail address entered séms valid,
* waiting for confirmation
*
* Game client states.
*/
typedef enum {
WMUD_CLIENT_STATE_FRESH,
WMUD_CLIENT_STATE_PASSWAIT,
WMUD_CLIENT_STATE_MENU,
WMUD_CLIENT_STATE_INGAME,
WMUD_CLIENT_STATE_YESNO,
WMUD_CLIENT_STATE_REGISTERING,
WMUD_CLIENT_STATE_REGEMAIL_CONFIRM
} WmudClientState;
GType wmud_client_state_get_type (void) G_GNUC_CONST;
#define WMUD_TYPE_CLIENT_STATE (wmud_client_state_get_type())
#endif /* __WMUD_CLIENT_STATE_H__ */
.. code-block:: c
#include "wmudclientstate.h"
GType
wmud_client_state_get_type (void)
{
static volatile gsize g_define_type_id__volatile = 0;
if (g_once_init_enter(&g_define_type_id__volatile)) {
static const GEnumValue values[] = {
{ WMUD_CLIENT_STATE_FRESH, "WMUD_CLIENT_STATE_FRESH", "fresh" },
{ WMUD_CLIENT_STATE_PASSWAIT, "WMUD_CLIENT_STATE_PASSWAIT", "passwait" },
{ WMUD_CLIENT_STATE_MENU, "WMUD_CLIENT_STATE_MENU", "menu" },
{ WMUD_CLIENT_STATE_INGAME, "WMUD_CLIENT_STATE_INGAME", "ingame" },
{ WMUD_CLIENT_STATE_YESNO, "WMUD_CLIENT_STATE_YESNO", "yesno" },
{ WMUD_CLIENT_STATE_REGISTERING, "WMUD_CLIENT_STATE_REGISTERING", "registering" },
{ WMUD_CLIENT_STATE_REGEMAIL_CONFIRM, "WMUD_CLIENT_STATE_REGEMAIL_CONFIRM", "regemail-confirm" },
{ 0, NULL, NULL }
};
GType g_define_type_id = g_enum_register_static(g_intern_static_string("WmudClientState"), values);
g_once_init_leave(&g_define_type_id__volatile, g_define_type_id);
}
return g_define_type_id__volatile;
}
Still, it can be made more perfect by using the `glib-mkenums
<http://developer.gnome.org/gobject/stable/glib-mkenums.html>`_ tool. I will read through the
GLib Makefiles tomorrow for some hints on this.
Edit: you can find the glib-mkenums solution `here
<{filename}2014-08-16-registering-an-enum-type-in-glib-glib-mkenums-magic.rst>`_.

16
content/blog/2013-01-14-git-rm-cached-madness.rst Normal file
View File

@@ -0,0 +1,16 @@
git rm --cached madness
#######################
:date: 2013-01-14T21:38:00Z
:category: blog
:tags: development,git
:url: blog/2013/1/14/git-rm-cached-madness.html
:save_as: blog/2013/1/14/git-rm-cached-madness.html
:status: published
:author: Gergely Polonkai
I have recently learned about ``git rm --cached``. It’s a very good tool, as it removes a file
from tracking, without removing your local copy of it. However, be warned that if you use ``git
pull`` in another working copy, the file will be removed from there! If you accidentally put the
configuration of a production project, and remove it on your dev machine, it can cause a lot of
trouble ;)

53
content/blog/2013-01-17-jms-diextrabundle-s-greppatternfinder-grep-exits-with-status-code-2-on-fedora-18.rst Normal file
View File

@@ -0,0 +1,53 @@
JMS\\DiExtraBundle’s GrepPatternFinder – grep exits with status code 2 on Fedora 18
###################################################################################
:date: 2013-01-17T00:32:12Z
:category: blog
:tags: fedora,selinux,symfony
:url: blog/2013/1/17/jms-diextrabundle-s-greppatternfinder-grep-exits-with-status-code-2-on-fedora-18.html
:save_as: blog/2013/1/17/jms-diextrabundle-s-greppatternfinder-grep-exits-with-status-code-2-on-fedora-18.html
:status: published
:author: Gergely Polonkai
Yesterday I’ve upgraded my development machines from Fedora 17 to Fedora
18. Although it went well, my `Symfony <http://symfony.com/>`_ projects stopped working with a
message like this:
.. code-block:: log
RuntimeException: Command "/usr/bin/grep --fixed-strings --directories=recurse --devices=skip --files-with-matches --with-filename --color=never --include=*.php 'JMS\DiExtraBundle\Annotation'
'/var/www/html/gergelypolonkaiweb/app/../src'
'/var/www/html/gergelypolonkaiweb/vendor/symfony/symfony/src/Symfony/Bundle/FrameworkBundle'
'/var/www/html/gergelypolonkaiweb/vendor/symfony/symfony/src/Symfony/Bundle/SecurityBundle'
'/var/www/html/gergelypolonkaiweb/vendor/symfony/symfony/src/Symfony/Bundle/TwigBundle'
'/var/www/html/gergelypolonkaiweb/vendor/symfony/monolog-bundle/Symfony/Bundle/MonologBundle'
'/var/www/html/gergelypolonkaiweb/vendor/symfony/swiftmailer-bundle/Symfony/Bundle/SwiftmailerBundle'
'/var/www/html/gergelypolonkaiweb/vendor/symfony/assetic-bundle/Symfony/Bundle/AsseticBundle'
'/var/www/html/gergelypolonkaiweb/vendor/doctrine/doctrine-bundle/Doctrine/Bundle/DoctrineBundle'
'/var/www/html/gergelypolonkaiweb/vendor/sensio/framework-extra-bundle/Sensio/Bundle/FrameworkExtraBundle'
'/var/www/html/gergelypolonkaiweb/vendor/jms/aop-bundle/JMS/AopBundle'
'/var/www/html/gergelypolonkaiweb/vendor/jms/security-extra-bundle/JMS/SecurityExtraBundle'
'/var/www/html/gergelypolonkaiweb/vendor/doctrine/doctrine-migrations-bundle/Doctrine/Bundle/MigrationsBundle'
'/var/www/html/gergelypolonkaiweb/vendor/friendsofsymfony/jsrouting-bundle/FOS/JsRoutingBundle'
'/var/www/html/gergelypolonkaiweb/vendor/avalanche123/imagine-bundle/Avalanche/Bundle/ImagineBundle'
'/var/www/html/gergelypolonkaiweb/vendor/genemu/form-bundle/Genemu/Bundle/FormBundle'
'/var/www/html/gergelypolonkaiweb/src/GergelyPolonkai/FrontBundle'
'/var/www/html/gergelypolonkaiweb/src/GergelyPolonkai/GeshiBundle'
'/var/www/html/gergelypolonkaiweb/vendor/symfony/symfony/src/Symfony/Bundle/WebProfilerBundle'
'/var/www/html/gergelypolonkaiweb/vendor/sensio/distribution-bundle/Sensio/Bundle/DistributionBundle'
'/var/www/html/gergelypolonkaiweb/vendor/sensio/generator-bundle/Sensio/Bundle/GeneratorBundle'" exited with non-successful status code "2".
After getting through my logs and such, I’ve finally found out that the new SELinux policy is
causing the trouble together with git. Eventually, my ``.git/logs`` directory is tagged as
``unconfined_u:object_r:httpd_log_t:s0``. ``httpd_log_t`` type is not readable by the
``system_u:system_r:httpd_t:s0`` user, which makes ``/usr/bin/grep`` throw an access denied error.
To fix this, I needed to do
.. code-block:: shell
semanage fcontext -a -t httpd_sys_content_t '/var/www(/.*)?/\.git/logs(/.*)?'
as root. This makes ``.git`` directories readable for the httpd process, thus, for ``grep``. The
optimal solution would be to tell ``GrepPatternFinder`` to ignore version control stuff, so the
``httpd`` process would have no access to them at all. Also, in production, removing the ``.git``
or ``.svn`` directories could be a good idea.

33
content/blog/2013-02-19-mount-device-or-resource-busy-after-enabling-multipath.rst Normal file
View File

@@ -0,0 +1,33 @@
mount: device or resource busy after enabling multipath
#######################################################
:date: 2013-02-19T23:09:05Z
:category: blog
:tags: linux,heartbeat-cluster
:url: blog/2013/2/19/mount-device-or-resource-busy-after-enabling-multipath.html
:save_as: blog/2013/2/19/mount-device-or-resource-busy-after-enabling-multipath.html
:status: published
:author: Gergely Polonkai
We have a heartbeat cluster with two nodes. It has been running for several months without
problems. The shared storage is on an IBM DS3400, on which we have a large volume formatted with
ext4.
Today I decided to reboot the active node for security reasons. So I’ve switched to the passive
node, which failed at the first step: it was unable to mount the storage (``/dev/sda1``). After
whining for a few moments, I tried to mount it by hand, which told me
.. code-block:: log
/dev/sda1 already mounted or /data is busy
I’ve quickly made sure that none of that was true. After checking this-and-that, it turned out
that the passive node had ``multipathd`` running, so I looked under ``/dev/mapper``, and found two
symlinks there, ``<long-long WWN>`` and ``<long-long WWN>-part1``. As the partition table and the
disk size was the same as on ``/dev/sda``, I tried to
.. code-block:: shell
mount /dev/<long-long WWN>-part1 /data
and voilà! It worked like charm!

23
content/blog/2013-02-27-why-i-stopped-using-annotation-based-routing-in-symfony-today.rst Normal file
View File

@@ -0,0 +1,23 @@
Why I stopped using annotation based routing in Symfony today
#############################################################
:date: 2013-02-27T23:10:24Z
:category: blog
:tags: development,symfony
:url: blog/2013/2/27/why-i-stopped-using-annotation-based-routing-in-symfony-today.html
:save_as: blog/2013/2/27/why-i-stopped-using-annotation-based-routing-in-symfony-today.html
:status: published
:author: Gergely Polonkai
I have read several opinions about routing configuration in Symfony. I stayed with annotation
based routing as it was convinient for me to see the URL right above the controller action. This
was because by just checking the URL, I remembered the controlling code, as they always were fresh
ones. Well, until today.
I had to take a look into an old (Sf 2.0, last commit was about 3 months ago) project of mine. In
the same run I’ve upgraded the whole project to 2.2 (it was a fast one, thanks for `JMikola@GitHub
<https://github.com/jmikola>`_ for the quick reply on my issue with `JmikolaJsAssetsHelperBundle
<https://github.com/jmikola/JmikolaJsAssetsHelperBundle>`_ again!). After that I went on to the
requested change. Now, finding a route in about 40 controller files spread between 3 bundles can
really be a pain! So I’ve finished with annotation based routing. It’s still a nice feature,
it’s simply not for me.

41
content/blog/2013-03-05-fedora-can-t-change-active-directory-password-via-kpasswd.rst Normal file
View File

@@ -0,0 +1,41 @@
Fedora can’t change Active Directory password via kpasswd
#########################################################
:date: 2013-03-05T08:55:04Z
:category: blog
:tags: fedora,kerberos,active-directory
:url: blog/2013/3/5/fedora-can-t-change-active-directory-password-via-kpasswd.html
:save_as: blog/2013/3/5/fedora-can-t-change-active-directory-password-via-kpasswd.html
:status: published
:author: Gergely Polonkai
I wanted to change my AD password today. As the AD is actually a Kerberos server, I was pretty
sure that ``kpasswd`` will do the trick. However, ``kpasswd`` output looked like this:
.. code-block:: output
$ kpasswd
Password for polonkai.gergely@EXAMPLE.LOCAL:
Enter new password:
Enter it again:
kpasswd: Cannot find KDC for requested realm changing password
I’ve checked ``kinit`` and ``klist``, everything looked fine. After a while it came
to my mind that password changing is done through the kadmin server, not
through the KDC. It seems that when I set up the Active Directory membership,
the ``admin_server`` directive is not get written to ``krb5.conf``. So all I had to
do was to put
.. code-block:: conf
admin_server = ad.example.local
in that file, and voilà!
.. code-block:: output
$ kpasswd
Password for polonkai.gergely@EXAMPLE.LOCAL:
Enter new password:
Enter it again:
Password changed.

25
content/blog/2013-03-05-haversine-in-mysql.rst Normal file
View File

@@ -0,0 +1,25 @@
Haversine in MySQL
##################
:date: 2013-03-05T12:49:28Z
:category: blog
:tags: mysql,development
:url: blog/2013/3/5/haversine-in-mysql.html
:save_as: blog/2013/3/5/haversine-in-mysql.html
:status: published
:author: Gergely Polonkai
Just insert it in your database, feed them two Google coordinates, and you get the distance in
kilometres. If you happen to need it in miles, change the constant ``12756.200`` in the
``RETURN`` row to ``7922.6`` instead.
.. code-block:: sql
DELIMITER $$
CREATE FUNCTION `haversine` (lng1 FLOAT, lat1 FLOAT, lng2 FLOAT, lat2 FLOAT)
RETURNS float NO SQL DETERMINISTIC
BEGIN
SET @a = ABS(POWER(SIN(RADIANS(lat1 - lat2)) / 2, 2) + COS(RADIANS(lat1)) * COS(RADIANS(lat2)) * POWER(SIN(RADIANS(lng1 - lng2)) / 2, 2));
RETURN 12756.200 * ATAN2(SQRT(@a), SQRT(1 - @a));
END$$

25
content/blog/2013-03-13-dvorak-and-me.rst Normal file
View File

@@ -0,0 +1,25 @@
Dvorak and me
#############
:date: 2013-03-13T21:20:13Z
:category: blog
:tags: linux
:url: blog/2013/3/13/dvorak-and-me.html
:save_as: blog/2013/3/13/dvorak-and-me.html
:status: published
:author: Gergely Polonkai
A few months ago I have decided to switch to the Dvorak layout. After using QWERTY (well, QWERTZ,
to be precise) for almost 17 years, it was a hard decision, but now I think it worthed the try. I
started with the UK (Dvorak with UK punctuation) layout, and in about four weeks, I’ve almost
reached my original typing speed. Today I have modified the Hungarian xkb definitions file to add
the Hungarian accended letters like ű to the layout, so I don’t have to use dead keys anymore
(which apparently turned out to be a problem, as the Linux version of Java doesn’t support dead
keys at all).
Best thing is, as I never learned proper 10-finger typing, but learned Dvorak that way, I can
switch between QWERTY and Dvorak more or less painlessly (about 10 minutes of confusion, so to
say).
Conclusion: I don’t know yet if this was actually a good decision, but it wasn’t bad, after all.
But seeing people’s faces when they try to type on my machine totally worths it.

25
content/blog/2013-04-09-renaming-a-symfony-2-bundle.rst Normal file
View File

@@ -0,0 +1,25 @@
Renaming a Symfony 2 bundle
###########################
:date: 2013-04-09T22:29:48Z
:category: blog
:tags: development,symfony
:url: blog/2013/4/9/renaming-a-symfony-2-bundle.html
:save_as: blog/2013/4/9/renaming-a-symfony-2-bundle.html
:status: published
:author: Gergely Polonkai
Today I’ve realised that the name I gave to one of my Symfony 2 bundles should be something else.
To rename a bundle, one must do four things (at least).
1. Change the namespace from ``Vendor\OldBundle`` to ``Vendor\NewBundle`` in every PHP class
(sounds like pain? It is…)
2. Change the name of files and classes. Some files under ``src/Vendor/OldBundle`` (and the
classes in them) contain the name of the bundle, like
``OldBundle/DependencyInjection/VendorOldBundleExtension.php`` and
``OldBundle/VendorOldBundle.php``. You should rename them, or Symfony won’t find the classes
defined in them! When done, rename the whole bundle directory either.
3. Change the configuration files accordingly, including ``AppKernel.php``. These config files
are usually ``routing.yml``, ``services.yml``, and in some cases, ``config.yml``.
4. Change the references in other parts of your code. A ``grep -r OldBundle .`` will usually
help…

107
content/blog/2013-05-06-installing-otrs-in-fedora-18-with-selinux-enabled.rst Normal file
View File

@@ -0,0 +1,107 @@
Installing OTRS in Fedora 18 with SELinux enabled
#################################################
:date: 2013-05-06T06:01:52Z
:category: blog
:tags: fedora,selinux,otrs
:url: blog/2013/5/6/installing-otrs-in-fedora-18-with-selinux-enabled.html
:save_as: blog/2013/5/6/installing-otrs-in-fedora-18-with-selinux-enabled.html
:status: published
:author: Gergely Polonkai
I’ve read somewhere in an OTRS installation howto that if you want to install OTRS, you will have
to disable SELinux. Well, I won’t.
During the last few months, I have been using Fedora 18 with SELinux on all of my desktop machines
and on my notebook, and I had no problems at all. Meanwhile I got familiar with SELinux itself,
and got used to solving problems caused by it. So I started ``tail -f /var/log/httpd/error_log``
in one terminal (to see if something Apache related thing appears), ``tail -f
/var/log/audit/audit.log`` in another (to see errors caused by SELinux), opened the admin manual
at the installation chapter, took a deep breath, and went on.
Throughout this article, I will refer to OTRS 3.2.6 as OTRS and Fedora 18 (with only “stock”
repositories) as Fedora. I assume that you have already installed OTRS in a non-SELinux
environment before, and that you have at least some basic knowledge about SELinux, MAC, RBAC, and
all the like. I’m installing OTRS in ``/opt/otrs``, so if you install it somewhere else, you will
have to modify the paths below. Also, if you happen to install under ``/var/www`` (I wouldn’t
recommend it), that directory already has the ``httpd_sys_content_t`` type, so you won’t have to
set it explicitly.
As the first step I have unpacked the archive to ``/opt/otrs``. This directory is the OTRS
default, many config files have it hardcoded, and changing it is no easy task.
Running ``otrs.CheckModules.pl`` gave me a list of missing perl modules. Red Hat and Fedora makes
it easy to install these, as you don’t have to know the RPM package name, just the perl module
name:
.. code-block:: shell
yum install 'perl(Crypt::SSLeay)' \
'perl(DBD::Pg)' \
'perl(GD)' \
'perl(JSON::XS)' \
'perl(GD::Text)' \
'perl(GD::Graph)' \
'perl(Mail::IMAPClient)' \
'perl(Net::DNS)' \
'perl(PDF::API2)' \
'perl(Text::CSV_XS)' \
'perl(YAML::XS)'
I also needed to install ``mod_perl``. Although ``otrs.CheckModules.pl`` didn’t mention it, the
default settings use syslog as the logging module, so unless you change it in ``Config.pm``, you
will also need to install ``'perl(Unix::Syslog)'``, either.
The default SELinux policy doesn’t permit any network connection to be initiated by Apache httpd.
As OTRS needs to connect to its database, you need to enable it explicitly. In older
distributions, the ``httpd_can_network_connect`` was the SELinux boolean for this, but recent
installations also have a ``httpd_can_network_connect_db`` flag. As far as I know, this enables
all network connections to the well-known database servers’ default port, but I will have to check
for it. For me, with a MySQL listening on its standard port, the ``setsebool
httpd_can_network_connect_db=1`` command just did it.
With SELinux enabled, Apache won’t be able to read anything that’s not marked with the
``httpd_sys_content_t`` type, nor write anywhere without the ``httpd_sys_rw_content_t`` type. The
trivial, quick and dirty solution is to label all the files as ``httpd_sys_rw_content_t``, and let
everything go. However, the goal of SELinux is just the opposite of this: grant access only to
what is really needed. After many trial-and-error steps, it finally turned out that for OTRS to
work correctly, you must set
* ``httpd_sys_content_t``
* on ``/opt/otrs/var/httpd/htdocs``
* ``httpd_script_exec_t``
* on ``/opt/otrs/bin/cgi-bin``
* ``httpd_sys_rw_content_t``
* on ``/opt/otrs/Kernel``
* on ``/opt/otrs/var/sessions``
* on ``/opt/otrs/var/log`` (unless you use syslog for logging)
* on ``/opt/otrs/var/packages`` (this is used only when you download an .opm package)
* on ``/opt/otrs/var/stats``
* on ``/opt/otrs/var/tmp``
* on ``/opt/otrs/bin`` (I wonder why this is required, though)
To do this, use the following command:
.. code-block:: sh
semanage fcontext -a -t <context> <directory regex>
Where ``<directory regex>`` is something like ``/opt/otrs/Kernel(/.*)?``. When this is done, all
you have to do is running ``restorecon -vR /opt/otrs`` so it will relabel everything with the
correct types (you can omit ``-v``, I just like to see what my software do).
The last thing I faced is that Fedora is more restrictive on reading directories other than
``/var/www``. It has a ``Require all denied`` on ``<Directory />``, and a ``Require all granted``
on ``<Directory /var/www>``, so ``/opt/otrs/var/httpd/htdocs`` will throw a ``403 Forbidden
(client denied by server configuration)`` error. To get rid of this, I had to modify
``scripts/apache2-httpd.include.conf`` and add ``Require all granted`` to both the ``cgi-bin`` and
``htdocs`` directories.
As I will have to use OTRS in a production environment soon with SELinux enabled, it is more than
sure that this list will change in the near future. As there are no official documentation on
this (I haven’t find one yet), I have to do it with the trial-and-error way, so be patient!

25
content/blog/2013-09-16-swe-glib-final-release.rst Normal file
View File

@@ -0,0 +1,25 @@
SWE-GLib final release
######################
:date: 2013-09-16T21:37:17Z
:category: blog
:tags: development,astrology
:url: blog/2013/9/16/swe-glib-final-release.html
:save_as: blog/2013/9/16/swe-glib-final-release.html
:status: published
:author: Gergely Polonkai
Few of you may know that I’m interested in astrology. About two months ago I have decided to
create an astrologers’ software for the GNOME desktop. Since then, I have contacted Jean-André
Santoni, who created a software called `Astrognome <https://code.google.com/p/astrognome/>`_ some
years ago. We exchanged some e-mails, and after several weeks of coding, I’m proud to present
`SWE-GLib <https://github.com/gergelypolonkai/swe-glib>`_ 1.0.1. This is “just” a library which
wraps `Swiss Ephemeris <http://www.astro.com/swisseph/>`_, creating a nice GLib-ish interface
around it. See the project page and the built-in GTK-Doc document for more information.
The astrologer’s software I’m writing will be Astrognome (you can check the `GitHub repository
<https://github.com/gergelypolonkai/astrognome>`_ already, thanks for Jean-André for letting me
use the name). It is currently in pre-alpha status, but already utilizes SWE-GLib (it just can’t
display the results yet). If you happen to be interested in astrology and/or Astrognome, fork the
repository and contribute! You can also contact me (or open an enhancement issue on GitHub) if
you have any ideas.

22
content/blog/2013-09-24-from-symfony-to-django-in-two-days.rst Normal file
View File

@@ -0,0 +1,22 @@
From Symfony to Django in two days
##################################
:date: 2013-09-24T14:05:22Z
:category: blog
:tags: development,symfony,django
:url: blog/2013/9/24/from-symfony-to-django-in-two-days.html
:save_as: blog/2013/9/24/from-symfony-to-django-in-two-days.html
:status: published
:author: Gergely Polonkai
I was a Python hater for a long time, although I can’t really tell why. It didn’t fit in my mind,
maybe. I was programming in BASIC, Pascal (none of these would come to my mind now, though), C,
PHP, Perl, JavaScript, and different shell “languages” like awk, sed or bash.
After I could not fit my next Symfony app on my cloud server (it is pretty low on storage), I have
decided to move slowly to Django. My first task was simple: transition my web page (this one)
from PHP + Symfony 2 to Python + Django. The results: the “static” pages are already working, the
blog listing is almost ready (some styling issues are still around), only tagging remains. And
this is after about 6 hours of work. Oh, and the admin site is included with Django, so I don’t
have to port that. I have also decided to finally integrate a comment feature in the Django
version.

25
content/blog/2013-11-05-first-impressions-of-windows-8.rst Normal file
View File

@@ -0,0 +1,25 @@
First impressions of Windows 8
##############################
:date: 2013-11-05T08:14:50Z
:category: blog
:tags: windows
:url: blog/2013/11/5/first-impressions-of-windows-8.html
:save_as: blog/2013/11/5/first-impressions-of-windows-8.html
:status: published
:author: Gergely Polonkai
Many of you may know my commitment to Linux and Open Source Software. But this doesn’t mean I
hate proprietary software like many others do. I think everything has its own place in the world,
and this goes for software as well.
A few days ago I got my hands on a new notebook, thanks to my company. It was shipped with
Windows 8 by default, and although I installed Fedora 19 in an instant (which went smoothlessly,
even with Secure Boot enabled), I’ve decided to give a try to this new Windows Version.
Being a heavy Windows 7 user, my first thought was “What the hell is this?” But in a day, I got
totally used to it. I don’t miss the Start button at all. The applications already installed
were almost enough for me (I still need Office. Maybe I’ll also enroll for Office 365 later…),
and the games are great and beautiful too. So overall, this new version may be totally different
(by the looks), but it seems almost the same Windows as we know it. So if you don’t freak out by
touching something new, go give it a try: don’t instant-remove 8 in favour of 7!

60
content/blog/2014-07-18-list-git-branches-and-their-remote-tracking-branches-side-by-side.rst Normal file
View File

@@ -0,0 +1,60 @@
List Git branches and their remote tracking branches side by side
#################################################################
:date: 2014-07-18T21:46:45Z
:category: blog
:tags: git
:url: blog/2014/7/18/list-git-branches-and-their-remote-tracking-branches-side-by-side.html
:save_as: blog/2014/7/18/list-git-branches-and-their-remote-tracking-branches-side-by-side.html
:status: published
:author: Gergely Polonkai
I had a hard time following my own branches in a project. They got pretty numerous, and I wasn’t
sure if I pushed them to ``origin`` at all. ``git branch -a`` can list all the branches,
including remote ones, but, as my list grew too big, it was impossible to follow it any more.
Thus, I have created a small script called ``git-branches-with-remotes``, which does the work for
me. Its only requirements are git (of course), and the ``column`` command, which is pretty
obviously present on every POSIX compliant systems (even OSX).
.. code-block:: shell
#! /bin/sh
COLUMN=`which column 2> /dev/null`
if test -z $COLUMN
then
echo "\`column' is not found in PATH. Cannot continue."
exit 1
fi
current_branch=`git rev-parse --abbrev-ref HEAD`
for branch in $(git for-each-ref --shell --format='%(refname)' refs/heads | sed -e s/^\'refs\\/heads\\/// -e s/\'$//)
do
remote=`git config branch.$branch.remote`
merge=`git config branch.$branch.merge | sed -e 's/^refs\/heads\///'`
[ x"$current_branch" == x"$branch" ] && echo -n '*'
echo -n "$branch"
if ! test -z $merge
then
echo -en "\t"
echo -n $remote
echo -n /
echo -n $merge
fi
echo
done | $COLUMN -t
I just put it in my path, and ``git branches-with-remotes`` does the work!
Edit (16 August): I have added some code to mark the current branch (if any) with an asterisk.
Also, I have put this script `in a gist
<https://gist.github.com/gergelypolonkai/8af6a3e86b57dd4c250e>`_.
Edit (26 February, 2015): It turns out that ``git branch -vv`` shows the same information and some
more: it also shows if the branches are diverged, and the first line of the last commit’s message.

148
content/blog/2014-08-16-registering-an-enum-type-in-glib-glib-mkenums-magic.rst Normal file
View File

@@ -0,0 +1,148 @@
Registering an enum type in GLib, glib-mkenums magic
####################################################
:date: 2014-08-16T15:10:54Z
:category: blog
:tags: development,c,glib
:url: blog/2014/8/16/registering-an-enum-type-in-glib-glib-mkenums-magic.html
:save_as: blog/2014/8/16/registering-an-enum-type-in-glib-glib-mkenums-magic.html
:status: published
:author: Gergely Polonkai
In `this post <{filename}2013-01-06-registering-an-enum-type-in-glib-s-type-system.rst>`_ I said I
will get through the GLib Makefiles to add an enum type to GLib in a more sophisticated way.
In my other project, `SWE-GLib <https://github.com/gergelypolonkai/swe-glib>`_ I already used this
method. The following two rules in ``Makefile.am`` create ``gswe-enumtypes.h`` and
``gswe-enumtypes.c``.
.. code-block:: makefile
gswe_enum_headers = headers-that-contain-enums.h
gswe-enumtypes.h: $(gswe_enum_headers) gswe-enumtypes.h.template
$(GLIB_MKENUMS) --template $(filter %.template,$^) $(filter-out %.template,$^) > \
gswe-enumtypes.h.tmp && mv gswe-enumtypes.h.tmp gswe-enumtypes.h
gswe-enumtypes.c: $(gswe_enum_headers) gswe-enumtypes.h gswe-enumtypes.c.template
$(GLIB_MKENUMS) --template $(filter %.template,$^) $(filter-out %.template,$^) > \
gswe-enumtypes.c.tmp && mv gswe-enumtypes.c.tmp gswe-enumtypes.c
``$(GLIB_MKENUMS)`` is set in ``configure`` using
.. code-block:: m4
AC_PATH_PROG([GLIB_MKENUMS], [glib-mkenums])
This approach requires the GNU Autotools (you can get rid of it by changing ``$(GLIB_MKENUMS)`` to
the path to ``glib-mkenums`` binary), and two template files, one for the header and one for the
code. ``$(gswe_enum_headers)`` contains a list of all the header files that have enum types
defined throughout the project.
.. code-block:: c
/*** BEGIN file-header ***/
/* gswe-enumtypes.h - Enumeration types for SWE-GLib
*
* Copyright © 2013 Gergely Polonkai
*
* SWE-GLib is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 3 of the License, or
* (at your option) any later version.
*
* SWE-GLib is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#ifndef __GSWE_ENUM_TYPES_H__
#define __GSWE_ENUM_TYPES_H__
#include <glib-object.h>
/*** END file-header ***/
/*** BEGIN file-production ***/
/* enumerations from "@filename@" */
#include "@filename@"
/*** END file-production ***/
/*** BEGIN value-header ***/
GType @enum_name@_get_type(void);
#define @ENUMPREFIX@_TYPE_@ENUMSHORT@ (@enum_name@_get_type())
/*** END value-header ***/
/*** BEGIN file-tail ***/
#endif /* __GSWE_ENUM_TYPES_H__ */
/*** END file-tail ***/
.. code-block:: c
/*** BEGIN file-header ***/
/* gswe-enumtypes.c - Enumeration types for SWE-GLib
*
* Copyright © 2013 Gergely Polonkai
*
* SWE-GLib is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 3 of the License, or
* (at your option) any later version.
*
* SWE-GLib is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "swe-glib.h"
#include "gswe-enumtypes.h"
#include "@filename@"
/*** END file-header ***/
/*** BEGIN file-production ***/
/* enumerations from "@filename@" */
/*** END file-production ***/
/*** BEGIN value-header ***/
GType
@enum_name@_get_type(void)
{
static volatile gsize g_define_type_id__volatile = 0;
gswe_init();
if (g_once_init_enter(&g;_define_type_id__volatile)) {
static const G@Type@Value values[] = {
/*** END value-header ***/
/*** BEGIN value-production ***/
{
@VALUENAME@,
"@VALUENAME@",
"@valuenick@"
},
/*** END value-production ***/
/*** BEGIN value-tail ***/
{ 0, NULL, NULL }
};
GType g_define_type_id = g_@type@_register_static(
g_intern_static_string("@EnumName@"),
values
);
g_once_init_leave(&g;_define_type_id__volatile, g_define_type_id);
}
return g_define_type_id__volatile;
}
/*** END value-tail ***/

14
content/blog/2014-09-17-nyanmacs.rst Normal file
View File

@@ -0,0 +1,14 @@
NyanMacs
########
:date: 2014-09-17T12:45:42Z
:category: blog
:tags: emacs
:url: blog/2014/9/17/nyanmacs.html
:save_as: blog/2014/9/17/nyanmacs.html
:status: published
:author: Gergely Polonkai
I was a Vi/ViM user for years. For several reasons I had to change to Emacs now and then. And
then, I found `this <https://github.com/TeMPOraL/nyan-mode/>`_. I surrender. Emacs is just
better. (And this addon is working even in plain text mode without graphics)

40
content/blog/2014-10-07-rounding-numbers-to-n-decimals-in-emacs.rst Normal file
View File

@@ -0,0 +1,40 @@
Rounding numbers to N decimals in Emacs
#######################################
:date: 2014-10-07T10:28:50Z
:category: blog
:tags: emacs,development
:url: blog/2014/10/7/rounding-numbers-to-n-decimals-in-emacs.html
:save_as: blog/2014/10/7/rounding-numbers-to-n-decimals-in-emacs.html
:status: published
:author: Gergely Polonkai
I have recently faced a problem, where I had a bunch of SVG files with a large amount of fraction
numbers in the path definitions. These images were displayed in small size, so this amount of
precision was irrelevant, and these numbers took almost half of my SVG images’ size. So I created
an Elisp defun to round these numbers to 2 decimals:
.. code-block:: lisp
(defun get-number-at-point ()
(interactive)
(skip-chars-backward "0123456789.-")
(or (looking-at "[0123456789.-]+")
(error "No number at point"))
(string-to-number (match-string 0)))
(defun round-number-at-point-to-decimals (decimal-count)
(interactive "NDecimal count: ")
(let ((mult (expt 10 decimal-count)))
(replace-match (number-to-string
(/
(fround
(*
mult
(get-number-at-point)))
mult)))))
This finds the first digit of the number under point (the cursor), and reduces its digits to the
given amount (or the number given with :kbd:`C-u`). It has some drawbacks, though, as it cannot
handle exponential forms (e.g. ``1e-1234``), but these were rare in my case, and its hard to
iterate through all numbers. I will come over this latter problem soon(ish).

44
content/blog/2015-02-26-using-git-bisect-to-find-the-first-good-commit.rst Normal file
View File

@@ -0,0 +1,44 @@
Using Git bisect to find the first good commit
##############################################
:date: 2015-02-26T10:42:56Z
:category: blog
:tags: git
:url: blog/2015/2/26/using-git-bisect-to-find-the-first-good-commit.html
:save_as: blog/2015/2/26/using-git-bisect-to-find-the-first-good-commit.html
:status: published
:author: Gergely Polonkai
Few months ago we “implemented” a bug in our software, which was released to the customers. We
continued development for two weeks when the first customer ticket arrived about the bug. We
successfully reproduced it with the customer’s version, but not with the development sources; it
turned out that one of the developers unconsciously fixed the bug. The devs spent some hours
finding where the fix lied before coming to me like “There is ``git-bisect`` which we can use to
find the commit where we messed up things. Is there a way to find where we fixed it?”
For those who don’t know this feature, you have to mark a known “good” and “bad” commit, then
git-bisect will go through the commits between this two, present you the corresponding snapshots,
and you have to mark each of them as “good” or “bad”. At the end, you will get a commit hash
where the bug first occured.
As it turned out, our developers’ problem rooted in the naming convention of git-bisect: they
assumed that the “good” commit must be a working one, while a “bad” one must be the buggy. In
this case, we did the following:
The commit with the customer’s release tag was marked as good (even though this had the bug), and
the latest commit on our development branch was marked as “bad” (even though the bug was fixed by
then). Now with every snapshot presented by git-bisect we had to do the opposite what you usually
do: mark commits still having the bug as “good”, and commits that don’t as “bad”. At the end, we
had the hash of the commit that fixed the bug (among some other things; luckily, the developer who
pushed that commit had a workflow that introduced a lot of cherry-picking and squashing before the
push, so he could easily find the bit that actually fixed the problem in his local repository with
the same technique).
`This StackOverflow answer <http://stackoverflow.com/a/17153598/1305139>`_ suggests the very same,
but with some aliases:
.. code-block:: dosini
[alias]
bisect-fixed = bisect bad
bisect-unfixed = bisect good

19
content/blog/2015-04-25-good-bye-digital-ocean-hello-again-github.rst Normal file
View File

@@ -0,0 +1,19 @@
Good bye, Digital Ocean! Hello again, GitHub!
#############################################
:date: 2015-04-25T21:18:56Z
:category: blog
:url: blog/2015/4/25/good-bye-digital-ocean-hello-again-github.html
:save_as: blog/2015/4/25/good-bye-digital-ocean-hello-again-github.html
:status: published
:author: Gergely Polonkai
Few years ago I have signed up for a `Digital Ocean <https://www.digitalocean.com/>`_ account. I
used one single droplet for my private needs, like hosting my private Git repositories and my
blog. However, as I didn’t host anything else there except my blog, I decided to shut it down.
From now on, my blog is on `GitHub Pages <https://pages.github.com/>`_, as it provides just
everything I need (except automatically converting my resume to PDF. But I can live without
that.)
I’m really sorry, Digital Ocean Guys, your hosting is awesome and I’ll keep recommending you to
others, but paying for a droplet for one single blog is overkill.

60
content/blog/2015-04-28-cross-browser-border-radius-sass-mixin-with-varargs.rst Normal file
View File

@@ -0,0 +1,60 @@
Cross browser border-radius SASS mixin with varargs
###################################################
:date: 2015-04-27T22:59:56Z
:category: blog
:tags: css,sass
:url: blog/2015/4/28/cross-browser-border-radius-sass-mixin-with-varargs.html
:save_as: blog/2015/4/28/cross-browser-border-radius-sass-mixin-with-varargs.html
:status: published
:author: Gergely Polonkai
Few days ago I needed to create style sheets with many rounded boxes, where different corners had
to be rounded differently (think about Bootstrap’s `button groups
<http://getbootstrap.com/components/#btn-groups>`_).
CSS has this nifty shorthand to specify border width in one line, like with ``border-width: 1px
2px 3px 4px``, but it lacks the same for ``border-radius``. So I decided to create something
similar using `Sass mixins <http://sass-lang.com/guide#topic-6>`_ with dynamic parameters.
Another nice feature you get using the ``border-width`` shorthand is that you can specify less
than four parameters, and the values will be applied on different sides of your box, so in the end
all side will have the whole ``border-width`` set.
I wanted to achieve the same for my ``border-radius`` mixin, although I
could not start specifically with the `top` side. I decided to go with
the top right corner for the first parameter, while trying to keep a
sane repeating pattern. Here is the result:
.. code-block:: sass
=border-width($t, $r: $t, $b: $t, $l: $r)
border-top-width: $t
border-right-width: $r
border-bottom-width: $b
border-left-width: $l
=border-top-right-radius($value)
border-top-right-radius: $value
-moz-border-top-right-radius: $value
-webkit-border-top-right-radius: $value
=border-top-left-radius($value)
border-top-left-radius: $value
-moz-border-top-left-radius: $value
-webkit-border-top-left-radius: $value
=border-bottom-right-radius($value)
border-bottom-right-radius: $value
-moz-border-bottom-right-radius: $value
-webkit-border-bottom-right-radius: $value
=border-bottom-left-radius($value)
border-bottom-left-radius: $value
-moz-border-bottom-left-radius: $value
-webkit-border-bottom-left-radius: $value
=border-radius($tr, $br: $tr, $bl: $br, $tl: $tr)
+border-top-right-radius($tr)
+border-bottom-right-radius($br)
+border-bottom-left-radius($bl)
+border-top-left-radius($tl)

154
content/blog/2015-06-07-paramconverter-a-la-django.rst Normal file
View File

@@ -0,0 +1,154 @@
@ParamConverter à la Django
###########################
:date: 2015-06-07T18:14:32Z
:category: blog
:tags: python,django
:url: 2015/06/07/paramconverter-a-la-django/
:save_as: 2015/06/07/paramconverter-a-la-django/index.html
:status: published
:author: Gergely Polonkai
One thing I really miss from `Django <https://www.djangoproject.com/>`_ is `Symfony
<http://symfony.com/>`_’s `@ParamConverter
<http://symfony.com/doc/current/bundles/SensioFrameworkExtraBundle/annotations/converters.html>`_. It
made my life so much easier while developing with Symfony. In Django, of course, there is
`get_object_or_404
<https://docs.djangoproject.com/en/dev/topics/http/shortcuts/#get-object-or-404>`_ but, for
example, in one of my projects I had a view that had to resolve 6(!) objects from the URL, and
writing ``get_object_or_404`` six times is not what a programmer likes to do (yes, this view had a
refactor later on). A quick Google search gave me one `usable result
<http://openclassrooms.com/forum/sujet/middleware-django-genre-paramconverter-doctrine>`_ (in
French), but it was very generalized that I cannot always use. Also, it was using a middleware,
which may introduce performance issues sometimes :sup:`[citation needed]`. So I decided to go
with decorators, and at the end, I came up with this:
.. code-block:: python
import re
from django.shortcuts import get_object_or_404
from django.db import models
def convert_params(*params_to_convert, **options):
"""
Convert parameters to objects. Each parameter to this decorator
must be a model instance (subclass of django.db.models.Model) or a
tuple with the following members:
* model: a Model subclass
* param_name: the name of the parameter that holds the value to be
matched. If not exists, or is None, the model’s class name will
be converted from ModelName to model_name form, suffixed with
"_id". E.g. for MyModel, the default will be my_model_id
* the field name against which the value in param_name will be
matched. If not exists or is None, the default will be "id"
* obj_param_name: the name of the parameter that will hold the
resolved object. If not exists or None, the default value will
be the model’s class name converted from ModelName to model_name
form, e.g. for MyModel, the default value will be my_model.
The values are resolved with get_object_or_404, so if the given
object doesn’t exist, it will redirect to a 404 page. If you want
to allow non-existing models, pass prevent_404=True as a keyword
argument.
"""
prevent_404 = options.pop('prevent_404', False)
def is_model(m):
return issubclass(type(m), models.base.ModelBase)
if len(params_to_convert) == 0:
raise ValueError("Must pass at least one parameter spec!")
if (
len(params_to_convert) == 1 and \
hasattr(params_to_convert[0], '__call__') and \
not is_model(params_to_convert[0])):
raise ValueError("This decorator must have arguments!")
def convert_params_decorator(func):
def wrapper(*args, **kwargs):
converted_params = ()
for pspec in params_to_convert:
# If the current pspec is not a tuple, let’s assume
# it’s a model class
if not isinstance(pspec, tuple):
pspec = (pspec,)
# First, and the only required element in the
# parameters is the model name which this object
# belongs to
model = pspec[0]
if not is_model(model):
raise ValueError(
"First value in pspec must be a Model subclass!")
# We will calculate these soon…
param_name = None
calc_obj_name = re.sub(
'([a-z0-9])([A-Z])',
r'\1_\2',
re.sub(
'(.)([A-Z][a-z]+)',
r'\1_\2',
model.__name__)).lower()
obj_field_name = None
# The second element, if not None, is the keyword
# parameter name that holds the value to convert
if len(pspec) < 2 or pspec[1] is None:
param_name = calc_obj_name + '_id'
else:
param_name = pspec[1]
if param_name in converted_params:
raise ValueError('%s is already converted' % param_name)
converted_params += (param_name,)
field_value = kwargs.pop(param_name)
# The third element is the field name which must be
# equal to the specified value. If it doesn’t exist or
# None, it defaults to 'id'
if (len(pspec) < 3) or pspec[2] is None:
obj_field_name = 'id'
else:
obj_field_name = pspec[2]
# The fourth element is the parameter name for the
# object. If the parameter already exists, we consider
# it an error
if (len(pspec) < 4) or pspec[3] is None:
obj_param_name = calc_obj_name
else:
obj_param_name = pspec[3]
if obj_param_name in kwargs:
raise KeyError(
"'%s' already exists as a parameter" % obj_param_name)
filter_kwargs = {obj_field_name: field_value}
if (prevent_404):
kwargs[obj_param_name] = model.objects.filter(
**filter_kwargs).first()
else:
kwargs[obj_param_name] = get_object_or_404(
model,
**filter_kwargs)
return func(*args, **kwargs)
return wrapper
return convert_params_decorator
Now I can decorate my views, either class or function based, with ``@convert_params(User,
(Article, 'aid'), (Paragraph, None, 'pid'), (AnotherObject, None, None, 'obj'))`` and all the
magic happens in the background. The ``user_id`` parameter passed to my function will be popped
off, and be resolved against the ``User`` model by using the ``id`` field; the result is put in
the new ``user`` parameter. For Article, the ``aid`` parameter will be matched against the ``id``
field of the ``Article`` model putting the result into ``article``, and finally, the
``another_object_id`` will be matched against the ``id`` field of the ``AnotherObject`` model, but
in this case, the result is passed to the original function as ``obj``.

17
content/blog/2015-08-19-foss-failmeter.rst Normal file
View File

@@ -0,0 +1,17 @@
F/OSS Fail meter
################
:date: 2015-08-19T10:12:19Z
:category: blog
:tags: development
:url: 2015/08/19/foss-failmeter/
:save_as: 2015/08/19/foss-failmeter/index.html
:status: published
:author: Gergely Polonkai
I have recently bumped into `this article <http://spot.livejournal.com/308370.html>`_. Naturally,
I quickly calculated the FAIL metrics for all my projects (most of them are pretty high). To ease
calculation, I made up a `small page <{static}../failmeter/index.html>`_ based on this list
(although I have divided the points by 5; I really don’t understand why spot is using such big
points if all of them can be divided by 5). Feel free to use it, and if you have any
recommendations (point additions/removal, new categories, etc.), leave me a comment!

201
content/blog/2015-08-27-how-my-email-gets-to-that-other-guy.rst Normal file
View File

@@ -0,0 +1,201 @@
How my e-mail gets to that other guy?
#####################################
:date: 2015-08-27T21:47:19Z
:category: blog
:tags: technology
:url: 2015/08/27/how-my-email-gets-to-that-other-guy/
:save_as: 2015/08/27/how-my-email-gets-to-that-other-guy/index.html
:status: published
:author: Gergely Polonkai
A friend of mine asked me how it is possible that she pushes buttons on her keyboard and mouse,
and in an instant her peer reads the text she had in her mind. This is a step-by-step
introduction of what happens in-between.
From your mind to your computer
===============================
When you decide to write an e-mail to an acquaintance of yours, you open up your mailing software
(this document doesn’t cover using mail applications you access through your browsers, just plain
old Thunderbird, Outlook or similar programs. However, it gets the same after the mail left your
computer), and press the “New Mail” button. What happens during this process is not covered in
this article, but feel free to ask me in a comment! Now that you have your Mail User Agent (MUA)
up and running, you begin typing.
When you press a button on your keyboard or mouse, a bunch of bits gets through the wire (or
through air, if you went wireless) and get into your computer. I guess you learned about Morse
during school; imagine two `Morse operators
<http://www.uscupstate.edu/academics/education/aam/lessons/susan_sawyer/morse%20code.jpg>`_, one
in your keyboard/mouse, and one in your computer. Whenever you press a key, that tiny creature
sends a series of short and long beeps (called 0 or 1 bits, respectively) to the operator in your
computer (fun fact: have you ever seen someone typing at an amazing speed of 5 key presses per
second? Now imagine that whenever that guy presses a key on their keyboard, that tiny little
Morse operator pressing his button 16 times for each key press, with perfect timing so that the
receiving operator can decide if that was a short or long beep.)
Now that the code got to the operator inside the machine, it’s up to him to decode it. The funny
thing about keyboards and computers is that the computer doesn’t receive the message “Letter Q was
pressed”, but instead “The second button on the second row was pressed” (a number called scan
code). At this time the operator decodes this information (in this example it is most likely this
Morse code: ``···-···· -··-····``) and checks one of his tables titled “Current Keyboard Layout.”
It says this specific key corresponds to letter ‘Q’, so it forwards this information (I mean the
letter; after this step your computer doesn’t care which plastic slab you hit, just the letter
‘Q’) to your MUA, inserts it into the mail in its memory, then displaying it happily (more about
this step later).
When you finish your letter you press the send button of your MUA. First it converts all the
pretty letters and pictures to something a computer can understand (yes, those Morse codes, or
more precisely, zeros and ones, again). Then it adds loads of meta data, like your name and
e-mail address, the current date and time including the time zone and pass it to the sending parts
of the MUA so the next step can begin.
IP addresses, DNS and protocols
===============================
The Internet is a huge amount of computers connected with each other, all of them having at least
one address called IP address that looks something like this: ``123.234.112.221``. These are four
numbers between 0 and 255 inclusive, separated by dots. This makes it possible to have
4,294,967,296 computers. With the rules of address assignment added, this is actually reduced to
3,702,258,432; a huge number, still, but it is not enough, as in the era of the Internet of Things
everything is interconnected, up to and possibly including your toaster. Thus, we are slowly
transitioning to a new addressing scheme that looks like this:
``1234:5678:90ab:dead:beef:9876:5432:1234``. This gives an enormous amount of
340,282,366,920,938,463,463,374,607,431,768,211,456 addresses, with only
4,325,185,976,917,036,918,000,125,705,034,137,602 of them being reserved, which gives us only a
petty 335,957,180,944,021,426,545,374,481,726,734,073,854 available.
Imagine a large city with `that many buildings
<http://www.digitallifeplus.com/wp-content/uploads/2012/07/new-york-city-aerial-5.jpg>`_, all of
them having only a number: their IP address. No street names, no company names, no nothing. But
people tend to be bad at memorizing numbers, so they started to give these buildings names. For
example there is a house with the number ``216.58.209.165``, but between each other, people call
it ``gmail.com``. Much better, isn’t it? Unfortunately, when computers talk, they only
understand numbers so we have to provide them just that.
As remembering this huge number of addresses is a bit inconvenient, we created Domain Name
Service, or DNS for short. A “domain name” usually (but not always) consist of two strings of
letters, separated by dots (e.g. polonkai.eu, gmail.com, my-very-long-domain.co.uk, etc.), and a
hostname is a domain name occasionally prefixed with something (e.g. **www**.gmail.com,
**my-server**.my-very-long-domain.co.uk, etc.) One of the main jobs of DNS is to keep record of
hostname/address pairs. When you enter ``gmail.com`` (which happens to be both a domain name and a
hostname) in your browser’s address bar, your computer asks the DNS service if it knows the actual
address of the building that people call ``gmail.com``. If it does, it will happily tell your
computer the number of that building.
Another DNS job is to store some meta data about these domain names. For such meta data there are
record types, one of these types being the Mail eXchanger, or MX. This record of a domain tells
the world who is handling incoming mails for the specified domain. For ``gmail.com`` this is
``gmail-smtp-in.l.google.com`` (among others; there can be multiple records of the same type, in
which case they usually have priorities, too.)
One more rule: when two computers talk to each other they use so called protocols. These
protocols define a set of rules on how they should communicate; this includes message formatting,
special code words and such.
From your computer to the mail server
=====================================
Your MUA has two settings called SMTP server address SMTP port number (see about that later).
SMTP stands for Simple Mail Transfer Protocol, and defines the rules on how your MUA, or another
mail handling computer should communicate with a mail handling computer when *sending* mail. Most
probably your Internet Service Provider gave you an SMTP server name, like ``smtp.aol.com`` and a
port number like ``587``.
When you hit that send button of yours, your computer will check with the DNS service for the
address of the SMTP server, which, for ``smtp.aol.com``, is ``64.12.88.133``. The computer puts
this name/address pair into its memory, so it doesn’t have to ask the DNS again (this technique is
called caching and is widely used wherever time consuming operations happen).
Then it will send your message to the given port number of this newly fetched address. If you
imagined computers as office buildings, you can imagine port numbers as departments and there can
be 65535 of them in one building. The port number of SMTP is usually 25, 465 or 587 depending on
many things we don’t cover here. Your MUA prepares your letter, adding your e-mail address and
the recipients’, together with other information that may be useful for transferring your mail.
It then puts this well formatted message in an envelope and writes “to building ``64.12.88.133``,
dept. ``587``”, and puts it on the wire so it gets there (if the wire is broken, the building does
not exist or there is no such department, you will get an error message from your MUA). Your
address and the recipient’s address are inside the envelope; other than the MUA, your own computer
is not concerned about it.
The mailing department (or instead lets call it the Mail Transfer Agent, A.K.A. MTA) now opens
this envelope and reads the letter. All of it, letter by letter, checking if your MUA formatted
it well. More than probably it also runs your message through several filters to decide if you
are a bad guy sending some unwanted letter (also known as spam), but most importantly it fetches
the recipients address. It is possible, e.g. when you send an e-mail within the same
organization, that the recipient’s address is handled by this very same computer. In this case
the MTA puts the mail to the recipient’s mailbox and the next step is skipped.
From one server to another
==========================
Naturally, it is possible to send an e-mail from one company to another, so these MTAs don’t just
wait for e-mails from you, but also communicate with each other. When you send a letter from your
``example@aol.com`` address to me at ``gergely@polonkai.eu``, this is what happens.
In this case, the MTA that initially received the e-mail from you (which happened to be your
Internet Service Provider’s SMTP server) turns to the DNS again. It will ask for the MX record of
the domain name specified by the e-mail address, (the part after the ``@`` character, in my case,
``polonkai.eu``), because the server mentioned that must be contacted, so they can deliver your
mail for me. My domain is configured so its primary MX record is ``aspmx.l.google.com`` and the
secondary is ``alt1.aspmx.l.google.com`` (and 5 more. Google likes to play it safe.) The MTA
then gets the first server name, asks the DNS for its address, and tries to send a message to the
``173.194.67.27`` (the address of ``aspmx.l.google.com``), same department. But unlike your MUA,
MTAs don’t have a pre-defined port number for other MTAs (although there can be exceptions).
Instead, they use well-known port numbers, ``465`` and ``25``. If the MTA on that server cannot
be contacted for any reason, it tries the next one on the list of MX records. If none of the
servers can be contacted, it will retry based on a set of rules defined by the administrators,
which usually means it will retry after 1, 4, 24 and 48 hours. If there is still no answer after
that many attempts, you will get an error message back, in the form of an e-mail sent directly by
the SMTP server.
Once the other MTA could be contacted, your message is sent there. The original envelope you used
is discarded, and a new one is used with the address and dept. number (port) of the receiving MTA.
Also, your message gets altered a little bit, as most MTAs are kind enough (ie. not sneaky) to add
a clause to your message stating “the MTA at <organization> has checked and forwarded this
message.”
It is possible, though not likely, that your message gets through more than two MTAs (one at your
ISP and one at the receiver’s) before arriving to its destination. At the end, an MTA will say
that “OK, this recipient address is handled by me”, your message stops and stays there, put in
your peer’s mailbox.
The mailbox
-----------
Now that the MTA has passed your mail to the mailbox team (I call it a team instead of department
because the tasks described here are usually handled by the MTA, too), it reads it. (Pesky little
guys these mail handling departments, aren’t they?) If the mailbox has some filtering rules, like
“if XY sends me a letter, mark it as important” or “if the letter has a specific word in its
subject, put it in the XY folder”, it executes them, but the main point is to land the message in
the actual post box of the recipient.
From the post box to the recipients computer
============================================
When the recipient opens their MUA, it will look to a setting usually called “Incoming mail
server”. Just like the SMTP server, it has a name and port number, along with a server type.
This type can vary from provider to provider, and is usually one of POP3 (pretty old protocol,
doesn’t even support folders on its own), IMAP (a newer one, with folders and message flags like
“important”), MAPI (a dialect of IMAP, created by Microsoft as far as I know), or plain old mbox
files on the receiving computer (this last option is pretty rare nowadays, so I don’t cover this
option. Also, if you use these, you most probably don’t really need this article to understand
how these things work.) This latter setting defines the protocol, telling your MUA how to “speak”
to the post box.
So your MUA turns to the DNS once more to get the address of your incoming mail server and
contacts it, using the protocol set by the server type. At the end, the recipients computer will
receive a bunch of envelopes including the one that contains your message. The MUA opens them one
by one and reads them, making a list ordered by their sender or subject, or the date of sending.
From the recipient’s comupter to their eyes
===========================================
When the recipient then clicks on one of these mails, the MUA will fetch all the relevant bits
like the sender, the subject line, the date of sending and the contents itself and sends it to the
“printing” department (I use quotes as they don’t really print your mail on paper, they just
convert it to a nice image so the recipient can see it. This is sometimes referred to as a
rendering engine). Based on a bunch of rules they pretty-print it and send it to your display as
a new series of Morse codes. Your display then decides how it will present it to the user: draw
the pretty pictures if it is a computer screen, or just raise and lower some hard dots that
represents letters on a Braille terminal.

131
content/blog/2016-01-13-emacs-implement-a-gobject-s-virtual-function.rst Normal file
View File

@@ -0,0 +1,131 @@
Emacs: Implement a GObject’s virtual function
#############################################
:date: 2016-01-13T13:31:12Z
:category: blog
:tags: c,development,emacs
:url: 2016/01/13/emacs-implement-a-gobject-s-virtual-function/
:save_as: 2016/01/13/emacs-implement-a-gobject-s-virtual-function/index.html
:status: published
:author: "Gergely Polonkai"
I have recently started creating a GLib implementation of the Matrix.org API. For that, I have
created a GObject interface, MatrixAPI, which has as many virtual functions as API calls (which is
a lot, and expanding). This way I ended up with the following scenario.
In ``matrix-api.h`` I had a struct like this, with a lot more elements:
.. code-block:: c
typedef struct {
void (*initial_sync)(MatrixAPI *api,
MatrixAPICallback callback,
gpointer user_data,
GError **error);
void (*sync)(MatrixAPI *api,
MatrixAPICallback callback,
gpointer user_data,
GError **error);
…
And in ``matrix-http-api.c``, which implements ``MatrixAPI``, I have a function like this (again,
with a lot more elements):
.. code-block:: c
static void
matrix_http_api_matrix_api_init(GObjectInterface *iface)
{
iface->initial_sync = i_initial_sync;
iface->sync = i_sync;
…
}
And every time I wanted to implement a new function from the vtable, I had to copy the prototype,
and add an ``iface->foo_bar = i_foo_bar`` line and an actual function header for ``i_foo_bar``
with the same parameters. That’s a cumbersome job for more than 40 function headers. But Emacs
comes to the rescue!
.. code-block:: lisp
(require 'thingatpt)
(defun get-point(symbol &optional arg)
"Get point, optionally running a command beforehand"
(funcall symbol arg)
(point))
(defun copy-symbol-at-point()
"Copy the symbol under point"
(interactive)
(save-excursion
(let ((beg (get-point 'beginning-of-thing 'symbol))
(end (get-point 'end-of-thing 'symbol)))
(copy-region-as-kill beg end))))
(defun implement-gobject-vfunc()
"Change a vtable line of a GObject interface to an implementation line like:
void (*my_iface_func)(type1 param1, type2 param2, ...);
to
iface->my_iface_func = i_my_iface_func;"
(interactive)
(save-excursion
(let ((beg ((lambda()
(search-forward "(*")
(point))))
(end ((lambda()
(back-to-indentation)
(point)))))
(kill-region beg end))
(copy-symbol-at-point)
(insert "iface->")
(end-of-thing 'symbol)
(delete-char 1)
(let ((beg (point))
(end ((lambda()
(find-list-end)
(point)))))
(kill-region beg end))
(insert " = i_")
(yank 2))
(next-line)
(beginning-of-line))
(defun implement-gobject-vfunc-prototype()
"Change a vtable line of a GObject interface to an implementation prototype line like:
void (*my_iface_func)(type1 param1, type2 param2, ...);
to
static void
i_my_iface_func(type1 param1, type2 param2, ...)"
(interactive)
(let ((beg ((lambda()
(back-to-indentation)
(point))))
(end ((lambda()
(beginning-of-line)
(point)))))
(kill-region beg end))
(insert "static ")
(search-forward "(*")
(delete-char -3)
(newline)
(insert "i_")
(end-of-thing 'symbol)
(delete-char 1)
(let ((beg (point))
(end ((lambda()
(find-list-end)
(point)))))
(indent-region beg end))
(delete-char 1))
Now all I have to do is to copy the whole vtable entry into ``matrix_http_api_matrix_api_init()``,
execute :kbd:`M-x implement-gobject-vfunc`, then put the same vtable entry somewhere before the
interface init function, and execute :kbd:`M-x implement-gobject-vfunc-prototype`.

32
content/blog/2016-02-26-vala-interface-madness.rst Normal file
View File

@@ -0,0 +1,32 @@
Vala interface madness
######################
:date: 2016-02-26T13:07:52Z
:category: blog
:tags: vala,development
:url: 2016/02/26/vala-interface-madness/
:save_as: 2016/02/26/vala-interface-madness/index.html
:status: published
:author: "Gergely Polonkai"
Although I have just started making it in C, I decided to move my Matrix GLib SDK to Vala. First
to learn a new language, and second because it is much easier to write GObject based stuff with
it.
For the first step I created a ``.vapi`` file from my existing sources, so the whole SDK prototype
was available for me in Vala.
I had a ``MatrixEvent`` class that implemented the ``GInitable`` interface, and many others were
subclassed ``MatrixEvent``. For some reason I don’t remember, I created the following header for
one of the event classes:
.. code-block:: vala
public class MatrixPresenceEvent : GLib.Object, GLib.Initable {
This is nice and everything, but as I didn’t create an ``init()`` method for
``MatrixPresenceEvent``, it tried to use the one from the parent class and somehow got into an
infinite loop. The Vala transformer (``valac``), however, doesn’t mention this.
Lessons learned: if you implement an interface on a subclass that is implemented by the parent
don’t forget to add the necessary functions to the subclass.

38
content/blog/2016-09-28-emacs-package-to-generate-gobject-boilerplate.rst Normal file
View File

@@ -0,0 +1,38 @@
Emacs package to generate GObject boilerplate
#############################################
:date: 2016-09-28T15:40:15Z
:category: blog
:tags: gnome,development
:url: 2016/09/28/emacs-package-to-generate-gobject-boilerplate/
:save_as: 2016/09/28/emacs-package-to-generate-gobject-boilerplate/index.html
:status: published
:author: Gergely Polonkai
Before I started using Vala (and sometimes even after that) I often needed to generate new classes
based on `GObject <https://developer.gnome.org/gobject/stable/>`_.
If you have ever worked with GObject in C, you know how tedious it can be. You need a pretty long
boilerplate just to register your class, and, if you want to be introspectable (and readable,
actually), your function names can grow really long.
To overcome this problem back in my ViM days, I used template files, where I could replace class
prefixes and names with a few keyboard macros. As I never really dug into ViM scripting other
than using some plugins, I never got farther than that. `Then came Emacs
<{filename}2014-09-17-nyanmacs.rst>`_.
I use Emacs for about two years now very extensively, up to and including GLib-based development.
I tried the template approach, but it felt to be a really poor experience, especially given that I
made my feet wet with Emacs Lisp. So I dug deeper, and created a package for that.
.. image:: {static}../images/screenshot-gobgen.png
:alt: A screenshot of GobGen in action
GobGen has its own buffer with some widgets, a bit similar to ``customize``. You can enter the
name of your new object and its parent, specify some settings. Then you press Generate, and you
are presented with two new buffers, one for the ``.c`` and another for the ``.h`` boilerplate.
There are a lot of things to do, actually. There is already an open issue for creating a major
mode for this buffer, and there are some minor switches I’d like to add, but it is already usable.
You can grab it from `MELPA <https://melpa.org/#/gobgen>`_ (my first package there; woo!) or from
my `GitHub account <https://github.com/gergelypolonkai/gobgen.el>`_.

63
content/blog/2016-10-04-git-merge-stages.rst Normal file
View File

@@ -0,0 +1,63 @@
git-merge stages
################
:date: 2016-10-04T12:46:00Z
:category: blog
:tags: git
:url: 2016/10/04/git-merge-stages/
:save_as: 2016/10/04/git-merge-stages/index.html
:status: published
:author: Gergely Polonkai
This was a mail to my company’s internal Git mailing list, after I realised many colleagues can’t
wrap their heads around merge conflicts.
Hello all,
I just saw this on the `git-users <https://groups.google.com/forum/#!forum/git-users>`_ list
and thought it could help you when you bump into a merge conflict. It is an excerpt from a
mail by Konstantin Khomoutov (one of the main contributors on the list), with a few
modifications of mine. Happy debugging :)
When a merge conflict is detected for a file, Git:
1. Updates the entry for that file in the index to make it contain
several so-called “stages”:
* `0`: “Ours” version – that one which was there in this index entry
before we begun to merge. At the beginning of the conflict, like
right after the `git merge` or `git rebase` command this won’t
exist (unless you had the file in the index, which you didn’t, did
you?). When you resolve the conflict and use `git add
my/conflicting/file.cc`, this will be the version added to the
staging area (index), thus, the resolution of the conflict.
* `1`: The version from the common ancestor commit, ie. the version
of the file both of you modified.
* `2`: The version from `HEAD`. During a merge, this is the current
branch. During a rebase, this is the branch or commit you are
rebasing onto, which usually will be `origin/develop`).
* `3`: The version being merged, or the commit you are rebasing.
2. Updates the file in the work tree to contain conflict markers and
the conflicting chunks of text between them (and the text from the
common ancestor if the `diff3` style of conflict markers was set).
Now you can use the numbers in point 1 to access the different stages
of the conflicting file. For example, to see the common ancestor (the
version both of you modified), use
.. code-block:: shell
git show :1:my/conflicting/file.cc
Or, to see the difference between the two conflicting versions, try
.. code-block:: shell
git diff :2:my/conflicting/file.cc :3:my/conflicting/file.cc
**Note** that you can’t use the ``:0:`` stage *before* you stage your resolution with ``git
add``, and you can’t use the ``:2:`` and ``:3:`` stages *after* you staged the resolution.
Fun fact: behind the scenes, these are the files (*revisions*) ``git mergetool`` accesses when
it presents you the conflict visually.

76
content/blog/2016-11-03-how-i-started-with-emacs.rst Normal file
View File

@@ -0,0 +1,76 @@
How I started with Emacs
########################
:date: 2016-11-03T09:58:41Z
:category: blog
:tags: emacs
:url: 2016/11/03/how-i-started-with-emacs/
:save_as: 2016/11/03/how-i-started-with-emacs/index.html
:status: published
:author: Gergely Polonkai
Sacha Chua has a nice `Emacs chat intro <http://sachachua.com/blog/2013/04/emacs-chat-intro/>`_
article back from 2013. I write this post half because she asks there about my (OK, anyone’s)
first Emacs moments, and half because I plan to do it for months now.
I wanted to start using Emacs 6(ish) years ago, and I was like “:kbd:`C-x` what”? (Note that back
around 1998, I was among the people who exited ``vi`` by killing it from another terminal after a
bunch of tries & fails like `these <http://osxdaily.com/2014/06/12/how-to-quit-vim/>`_.)
I tried to come back to Emacs a lot of times. And I mean a *lot*, about every two months. I
suddenly learned what these cryptic key chord descriptions mean (``C`` is for :kbd:`Control` and
``M`` is for :kbd:`Meta`, which is actually :kbd:`Alt`), but somehow it didn’t *click*. I
remained a ViM power user with a huge pile of 3:sup:`rd` party plugins. Then `I found Nyan-macs
<{filename}2014-09-17-nyanmacs.rst>`_), which converted me to Emacs, and it is final now. Many of
my friends thought I’m just kidding this being the cause, but I’m not. I’m a huge fan of Nyan cat
(did you know there is even a site called `nyan.cat <http://nyan.cat/>`_?) and since then I have
it in my mode line:
.. image:: {static}../images/nyan-modeline.png
:alt: Nyan modeline
…in my ``eshell`` prompt:
.. image:: {static}../images/nyan-eshell.png
:alt: eshell prompt with a Nyan cat
…and I also `zone out <https://www.emacswiki.org/emacs/ZoneMode>`_ with Nyan cat:
.. image:: {static}../images/nyan-zone.png
:alt: a text-based animation with Nyan cat
Now on to more serious stuff. After browsing through all the packages provided by `ELPA
<http://elpa.gnu.org/>`_, I found tons of useful (and sometimes, less useful) packages, like `Helm
<https://github.com/emacs-helm/helm/wiki>`_, `company <http://company-mode.github.io/>`_, `gtags
<https://www.emacswiki.org/emacs/GnuGlobal>`_ (which introduced me to GNU Global, removing
Exuberant ctags from my life), `magit <https://magit.vc/>`_, `Projectile
<http://batsov.com/projectile/>`_, and `Org <http://orgmode.org/>`_ (OK, it’s actually part of
Emacs for a while, but still). I still use these few, but in a month or two, I started to
`version control <https://github.com/gergelypolonkai/my-emacs-d>`_ my ``.emacs.d`` directory, so I
can easily transfer it between my home and work machine (and for a few weeks now, even to my
phone: I’m using Termux on Android). Then, over these two years I wrote some packages like
`GobGen <https://github.com/gergelypolonkai/gobgen.el>`_, and a small addon for Calendar providing
`Hungarian holidays <https://github.com/gergelypolonkai/hungarian-holidays>`_, and I found a lot
more (in no particular order):
* `git-gutter <https://github.com/syohex/emacs-git-gutter>`_
* `multiple-cursors <https://github.com/magnars/multiple-cursors.el>`_
* `origami <https://github.com/gregsexton/origami.el>`_
* `ace-window <https://github.com/abo-abo/ace-window>`_
* `avy <https://github.com/abo-abo/avy>`_
* `beacon <https://github.com/Malabarba/beacon>`_
…and a lot more.
What is more important (to me) is that I started using the `use-package
<https://github.com/jwiegley/use-package>`_ package, which can automatically download packages
that are not installed on my current local system. Together with `auto-package-update
<https://github.com/rranelli/auto-package-update.el>`_, it is *very* practical.
In addition, I started to follow the blogs of a bunch of Emacs users/gurus. I’ve already
mentioned `Sacha Chua <http://sachachua.com/>`_. She’s a charming, cheerful person, writing a lot
about Emacs and project management (among other things). Another one is `Bozhidar Batsov
<http://batsov.com/>`_, who, among other things, had an initiate to lay down the foundation of a
`common Elisp coding style <https://github.com/bbatsov/emacs-lisp-style-guide>`_. Another
favourite of mine is `Endless Parentheses <http://endlessparentheses.com/>`_, whence I got a lot
of ideas.

38
content/blog/2016-11-10-edit-file-as-other-user-in-emacs.rst Normal file
View File

@@ -0,0 +1,38 @@
Edit file as another user in Emacs
##################################
:date: 2016-11-10T08:57:12Z
:category: blog
:tags: development,emacs
:url: 2016/11/10/edit-file-as-other-user-in-emacs/
:save_as: 2016/11/10/edit-file-as-other-user-in-emacs/index.html
:status: published
:author: Gergely Polonkai
I have recently found `this article <http://emacsredux.com/blog/2013/04/21/edit-files-as-root/>`_
by Bozhidar Batsov on opening the current file as root. I barely use `tramp
<https://www.gnu.org/software/tramp/>`_ for sudo access, but when I do, I almost never use root as
the target user. So I decided to fix it for my needs.
.. code-block:: lisp
(defun open-this-file-as-other-user (user)
"Edit current file as USER, using `tramp' and `sudo'. If the current
buffer is not visiting a file, prompt for a file name."
(interactive "sEdit as user (default: root): ")
(when (string= "" user)
(setq user "root"))
(let* ((filename (or buffer-file-name
(read-file-name (format "Find file (as %s): "
user))))
(tramp-path (concat (format "/sudo:%s@localhost:" user) filename)))
(if buffer-file-name
(find-alternate-file tramp-path)
(find-file tramp-path))))
If the user is not specified, the default is still root. Also, if the current buffer is not
visiting a file, I prompt for a filename. As I’m not an ``ido`` user, I didn’t bother calling
``ido-read-file-name``; `helm <https://github.com/emacs-helm/helm/wiki>`_ overrides
``read-file-name`` for me anyway.
Unlike Bozhidar, I barely use this feature, so I didn’t bind this to a key.

46
content/blog/2016-11-18-get-passwords-from-id-manager.rst Normal file
View File

@@ -0,0 +1,46 @@
Get account data programatically from id-manager
################################################
:date: 2016-11-18T12:43:13Z
:category: blog
:tags: emacs
:url: 2016/11/18/get-passwords-from-id-manager/
:save_as: 2016/11/18/get-passwords-from-id-manager/index.html
:status: published
:author: Gergely Polonkai
I recently started using `id-manager <https://github.com/kiwanami/emacs-id-manager>`_. It is a
nice little package that can store your passwords, encrypting them with GPG. My original reason
was to store my GitHub access token for `github-notifier
<https://github.com/xuchunyang/github-notifier.el>`_, but it soon turned out, it’s not *that*
easy.
``id-manager`` is a nice package when it comes to storing your password and retrieving them for
your own eyes. But it cannot retrieve account data programatically. Taking a look into its
source code, I came up with this solution:
.. code-block:: lisp
(defun gpolonkai/idm-record-get-field (record field)
"Get FIELD of an id-manager RECORD."
(let ((funcname (intern (concat "idm-record-" (symbol-name field)))))
(when (fboundp funcname)
(funcall funcname record))))
(defun gpolonkai/idm-get-field-for-account (account field)
"Get id-manager password for ACCOUNT."
(let ((db (idm-load-db))
(lookup-record nil))
(dolist (record (funcall db 'get-all-records) password)
(when (string= account (idm-record-name record))
(setq lookup-record (gpolonkai/idm-record-get-field record field))))
lookup-record))
(defmacro gpolonkai/idm-get-password-for-account (account)
`(gpolonkai/idm-get-field-for-account ,account 'password))
(defmacro gpolonkai/idm-get-id-for-account (account)
`(gpolonkai/idm-get-field-for-account ,account 'account-id))
I currently need only the account ID (ie. the username) and the password, but it’s pretty easy to
add a macro to get the ``memo`` or ``update-time`` fields, too.

80
content/blog/2016-11-30-add-python-docstring-to-the-beginning-of-anything.rst Normal file
View File

@@ -0,0 +1,80 @@
Add Python docstring to the beginning of anything in Emacs
##########################################################
:date: 2016-11-30T07:52:37Z
:category: blog
:tags: development,python,emacs
:url: 2016/11/30/add-python-docstring-to-the-beginning-of-anything/
:save_as: 2016/11/30/add-python-docstring-to-the-beginning-of-anything/index.html
:status: published
:author: Gergely Polonkai
Now that I write Python code for a living, I write a lot of functions, classes, and modules. What
I still tend to forget, and also find tedious, is adding docstrings. Unlike many developers,
writing documentation is not an enemy of mine, but it usually comes to my mind when I finish
implementation. The procedure, roughly, is this:
* Decide where I am (in a function, in a class but not in one of its methods, or not inside such a
block at all)
* Go to the beginning of the thing
* Insert ``"""``
* Leave a blank line
* Insert ``"""``
One of my mottos is if something takes more than one step and you have to do it more than twice,
you should automate it after the first time. This puts a small(ish) overhead on the second
invocation (when you implement the automation), but it usually worth the time.
Since I use Emacs for pretty much everything coding-related (and many more, but that’s the topic
of a different post), I wrote a small function to do it for me.
.. code-block:: lisp
(defun gpolonkai/prog-in-string-p ()
"Return `t' if point is inside a string."
(nth 3 (syntax-ppss)))
(defun gpolonkai/prog-in-comment-p ()
"Return `t' if point is inside a comment."
(nth 4 (syntax-ppss)))
(defun gpolonkai/python-add-docstring ()
"Add a Python docstring to the current thing. If point is
inside a function, add docstring to that. If point is in a
class, add docstring to that. If neither, add docstring to the
beginning of the file."
(interactive)
(save-restriction
(widen)
(beginning-of-defun)
(if (not (looking-at-p "\\(def\\|class\\) "))
(progn
(goto-char (point-min))
(back-to-indentation)
(forward-char)
(while (gpolonkai/prog-in-comment-p)
(forward-line)
(back-to-indentation)
(forward-char)))
(search-forward ":")
(while (or (gpolonkai/prog-in-string-p)
(gpolonkai/prog-in-comment-p))
(search-forward ":")))
(if (eq 1 (count-lines 1 (point)))
(open-line-above)
(open-line-below))
(insert "\"\"\"")
(open-line-below)
(insert "\"\"\"")
(open-line-above)))
There are still a lot of things to improve:
* it always inserts double quotes (althoug I couldn’t show a use-case when single quotes are
preferred)
* it doesn’t check for an existing docstring, just happily inserts a new one (leaving the old one
alone, but generating a syntax error this way)
* it would also be nice if I could jump to the beginning of a file even from a class method. I
guess I will use prefix keys for that, but I’m not sure yet.
You can bet I will implement these features, so check back soon for an updated version!

40
content/blog/2016-12-08-slugify-in-python3.rst Normal file
View File

@@ -0,0 +1,40 @@
Slugify in Python 3
###################
:date: 2016-12-08T12:54:19Z
:category: blog
:tags: development,python
:url: 2016/12/08/slugify-in-python3/
:save_as: 2016/12/08/slugify-in-python3/index.html
:status: published
:author: Gergely Polonkai
Today I needed a function to create a slug (an ASCII-only representation of a string). I went
Googling a bit, and found an excellend `Flask snippet <http://flask.pocoo.org/snippets/5/>`_.
Problem is, it is designed for Python 2, so I came up with a Python 3 version.
.. code-block:: python
import re
from unicodedata import normalize
_punctuation_re = re.compile(r'[\t !"#$%&\'()*\-/<=>?@\[\\\]^_`{|},.]+')
def slugify(text, delim='-'):
"""
Generate an ASCII-only slug.
"""
result = []
for word in _punctuation_re.split(text.lower()):
word = normalize('NFKD', word) \
.encode('ascii', 'ignore') \
.decode('utf-8')
if word:
result.append(word)
return delim.join(result)
As I don’t really like the transliteration done in the first example (e.g. converting ü to ue), I
went with the second example.

246
content/blog/2016-12-22-finding-non-translated-strings-in-python-code.rst Normal file
View File

@@ -0,0 +1,246 @@
Finding non-translated strings in Python code
#############################################
:date: 2016-12-22T09:35:11Z
:category: blog
:tags: development,python
:url: 2016/12/22/finding-non-translated-strings-in-python-code/
:save_as: 2016/12/22/finding-non-translated-strings-in-python-code/index.html
:status: published
:author: Gergely Polonkai
When creating multilingual software, be it on the web, mobile, or desktop, you will eventually
fail to mark strings as translatable. I know, I know, we developers are superhuman and never do
that, but somehow I stopped trusting myself recently, so I came up with an idea.
Right now I assist in the creation of a multilingual site/web application, where a small part of
the strings come from the Python code instead of HTML templates. Call it bad practice if you
like, but I could not find a better way yet.
As a start, I tried to parse the source files with simple regular expressions, so I could find
anything between quotation marks or apostrophes. This attempt quickly failed with strings that
had such characters inside, escaped or not; my regexps became so complex I lost all hope. Then
the magic word “lexer” came to mind.
While searching for ready made Python lexers, I bumped into the awesome ``ast`` module. AST
stands for Abstract Syntax Tree, and this module does that: parses a Python file and returns a
tree of nodes. For walking through these nodes there is a ``NodeVisitor`` class (among other
means), which is meant to be subclassed. You add a bunch of ``visitN`` methods (where ``N`` is an
``ast`` class name like ``Str`` or ``Call``), instantiate it, and call its ``visit()`` method with
the root node. For example, the ``visitStr()`` method will be invoked for every string it finds.
How does it work?
=================
Before getting into the details, let’s me present you the code I made:
.. code-block:: python
import ast
import gettext
from gettext import gettext as _
import sys
def get_func_name(node):
cls = node.__class__.__name__
if cls == 'Call':
return get_func_name(node.func)
elif cls == 'Attribute':
return '{}.{}'.format(
get_func_name(node.value),
node.attr)
elif cls == 'Name':
return get_func_name(node.id)
elif cls == 'str':
return node
elif cls == 'Str':
return "<String literal>"
elif cls == 'Subscript':
return '{}[{}]'.format(get_func_name(node.value),
get_func_name(node.slice))
elif cls == 'Index':
return get_func_name(node.value)
else:
print('ERROR: Unknown class: {}'.format(cls))
class ShowStrings(ast.NodeVisitor):
TRANSLATION_FUNCTIONS = [
'_', # gettext.gettext is often imported under this name
'gettext',
'gettext.gettext',
# FIXME: this list is pretty much incomplete
]
UNTRANSLATED = 'untranslated 9'
def __init__(self, filename=None):
super(ShowStrings, self).__init__()
self.in_call = []
self.filename = filename or '<parsed string>'
def visit_with_trace(self, node, func):
self.in_call.append((func, node.lineno, node.col_offset))
self.visit(node)
self.in_call.pop()
def visit_Str(self, node):
# TODO: make it possible to ignore untranslated strings
# TODO: make this ignore docstrings
# if we are not in a translator function, issue a warning
if not self.in_call or \
self.in_call[-1][0] not in self.TRANSLATION_FUNCTIONS:
try:
funcname = self.in_call[-1][0]
except IndexError:
funcname = None
funcall_msg = "outside a function call" if funcname is None \
else "inside a call to {funcname}".format(
funcname=funcname)
print("WARNING: Untranslated string found at "
"{filename}:{line}:{col} {funcall_msg}".format(
filename=self.filename,
line=node.lineno,
col=node.col_offset,
funcall_msg=funcall_msg))
def visit_Call(self, node):
# if we are in a translator function, issue a warninc
if self.in_call and self.in_call[-1][0] in self.TRANSLATION_FUNCTIONS:
print("WARNING: function call within a translation function at "
"{filename}:{line}:{col}".format(filename=self.filename,
line=node.lineno,
col=node.col_offset))
funcname = get_func_name(node)
for arg in node.args:
self.visit_with_trace(arg, funcname)
for kwarg in node.keywords:
self.visit_with_trace(kwarg.value, funcname)
def generic_visit(self, node):
# if we are inside a translator function, issue a warning
if self.in_call and self.in_call[-1][0] in self.TRANSLATION_FUNCTIONS:
# Some ast nodes, like Add don’t have position information
if hasattr(node, 'lineno'):
print("WARNING: something not a string ({klass}) found in a "
"translation function at {filename}:{line}:{col}".format(
filename=self.filename,
klass=node.__class__.__name__,
line=node.lineno,
col=node.col_offset))
else:
print("WARNING: something not a string ({klass}) found in a "
"translation function. Position unknown; function call "
"is at {filename}:{line}:{col}".format(
filename=self.filename,
klass=node.__class__.__name__,
line=self.in_call[-1][1],
col=self.in_call[-1][2]))
super(ShowStrings, self).generic_visit(node)
def tst(*args, **kwargs):
pass
def actual_tests():
_('translated 1')
tst(_('translated 2'))
tst(gettext.gettext('translated 3'))
tst(_('translated 4') + 'native 1')
tst('native 2'
'native 3')
tst(_('native 4' + 'native 5'))
tst('native 6', b='native 7')
tst(_(tst('hello!')))
if __name__ == '__main__':
try:
filename = sys.argv[1]
except IndexError:
filename = __file__
print("INFO: No filename specified, checking myself.")
with open(filename, 'r') as f:
code = f.read()
root = ast.parse(code)
show_strings = ShowStrings(filename=filename)
show_strings.visit(root)
The class initialization does two things: creates an empty ``in_call`` list (this will hold our
primitive backtrace), and saves the filename, if provided.
``visitCall``, again, has two tasks. First, it checks if we are inside a translation function.
If so, it reports the fact that we are translating something that is not a raw string. Although
it is not necessarily a bad thing, I consider it bad practice as it may result in undefined
behaviour.
Its second task is to walk through the positional and keyword arguments of the function call. For
each argument it calls the ``visit_with_trace()`` method.
This method updates the ``in_call`` property with the current function name and the position of
the call. This latter is needed because ``ast`` doesn’t store position information for every node
(operators are a notable example). Then it simply visits the argument node, which is needed
because ``NodeVisitor.visit()`` is not recursive. When the visit is done (which, with really
deeply nested calls like ``visit(this(call(iff(you(dare)))))`` will be recursive), the current
function name is removed from ``in_call``, so subsequent calls on the same level see the same
“backtrace”.
The ``generic_visit()`` method is called for every node that doesn’t have a named visitor (like
``visitCall`` or ``visitStr``. For the same reason we generate a warning in ``visitCall``, we do
the same here. If there is anything but a raw string inside a translation function call,
developers should know about it.
The last and I think the most important method is ``visitStr``. All it does is checking the last
element of the ``in_call`` list, and generates a warning if a raw string is found somewhere that
is not inside a translation function call.
For accurate reports, there is a ``get_func_name()`` function that takes an ``ast`` node as an
argument. As function call can be anything from actual functions to object methods, this goes all
down the node’s properties, and recursively reconstructs the name of the actual function.
Finally, there are some test functions in this code. ``tst`` and
``actual_tests`` are there so if I run a self-check on this script, it will
find these strings and report all the untranslated strings and all the
potential problems like the string concatenation.
Drawbacks
=========
There are several drawbacks here. First, translation function names are built in, to the
``TRANSLATION_FUNCTIONS`` property of the ``ShowString`` class. You must change this if you use
other translation functions like ``dngettext``, or if you use a translation library other than
``gettext``.
Second, it cannot ignore untranslated strings right now. It would be great if a pragma like
``flake8``’s ``# noqa`` or ``coverage.py``’s ``# pragma: no cover`` could be added. However,
``ast`` doesn’t parse comment blocks, so this proves to be challenging.
Third, it reports docstrings as untranslated. Clearly, this is wrong, as docstrings generally
don’t have to be translated. Ignoring them, again, is a nice challenge I couldn’t yet overcome.
The ``get_func_name()`` helper is everything but done. As long as I cannot remove that final
``else`` clause, there may be error reports. If that happens, the reported class should be
treated in a new ``elif`` branch.
Finally (and the most easily fixed), the warnings are simply printed on the console. It is nice,
but it should be optional; the problems identified should be stored so the caller can obtain it as
an array.
Bottom line
===========
Finding strings in Python sources is not as hard as I imagined. It was fun to learn using the
``ast`` module, and it does a great job. Once I can overcome the drawbacks above, this script
will be a fantastic piece of code that can assist me in my future tasks.

39
content/blog/2017-01-02-rename-automatically-named-foreign-keys-with-alembic.rst Normal file
View File

@@ -0,0 +1,39 @@
Rename automatically named foreign keys with Alembic
####################################################
:date: 2017-01-02T09:41:23Z
:category: blog
:tags: mysql,development,flask,python
:url: 2017/01/02/rename-automatically-named-foreign-keys-with-alembic/
:save_as: 2017/01/02/rename-automatically-named-foreign-keys-with-alembic/index.html
:status: published
:author: Gergely Polonkai
I have recently messed up my Alembic migrations while modifying my SQLAlchemy models. To start
with, I didn’t update the auto-generated migration files to name the indexes/foreign keys a name,
so Alembic used its own naming scheme. This is not an actual problem until you have to modify
columns that have such constraints. I have since fixed this problem, but first I had to find
which column references what (I had no indexes other than primary key back then, so I could go
with foreign keys only). Here is a query I put together, mostly using `this article
<http://www.binarytides.com/list-foreign-keys-in-mysql/>`_.
.. code-block:: sql
SELECT constraint_name,
CONCAT(table_name, '.', column_name) AS 'foreign key',
CONCAT(referenced_table_name, '.', referenced_column_name) AS 'references'
FROM information_schema.key_column_usage
WHERE referenced_table_name IS NOT NULL AND
table_schema = 'my_app';
Now I could easily drop such constraints using
.. code-block:: python
alembic.op.drop_constraint('users_ibfk1', 'users', type_='foreignkey')
and recreate them with
.. code-block:: python
alembic.op.create_foreign_key('fk_user_client', 'users', 'clients', ['client_id'], ['id'])

85
content/blog/2017-03-26-category-based-logging-with-flask.rst Normal file
View File

@@ -0,0 +1,85 @@
Category-based logging with Flask
#################################
:date: 2017-03-26T22:00:52Z
:category: blog
:tags: development,python,flask
:url: 2017/03/27/category-based-logging-with-flask/
:save_as: 2017/03/27/category-based-logging-with-flask/index.html
:status: published
:author: Gergely Polonkai
I’m in a team who are developing a Flask-based web application, which uses logging extensively.
For a while now it spews out a lot of lines so the need arose to index them in ElasticSearch, and
more importantly, to search through them for auditing purposes. This latter user story brought up
one more question: why don’t we categorize our log messages? I quickly came up with an extended
log format (``[auth]`` is the new category name):
.. code-block:: log
[2017-01-14 00:55:42,554] [8286] [INFO] [auth] invalid password for john@example.com [at __init__.py:12, in function utils.validate_login]
Here, ``[auth]`` is the category name. In the ideal solution, all I’d have to
do is adding ``%(category)s`` to my formatter, and I could call
.. code-block:: python
app.logger.info(message, extra={'auth': 'invalid password'})
to achieve this output. Unfortunately, ``Flask.logger`` (and, in the background, the ``logging``
module) is not that easy to tame.
As it turns out, a Flask application’s ``logger`` property is a subclass of ``logging.Logger``, so
my first idea was to monkey patch that class. When the app’s logger is initialised, it subclasses
``logging.Logger`` and tweaks the log level so it goes down to ``logging.DEBUG`` if the app is
running in debug mode. This is done by using a different logger class depending on the app
config. Fortunately it doesn’t directly subclass ``logging.Logger``; it calls
``logging.getLoggerClass()`` to find which class it should extend. To achieve my goals, all I had
to do is to subclass the original logger class, and pass it to ``logging.setLoggerClass()``
*before* I initialise my app, and I have a fail-safe(ish) solution. So far so good, on to the
extra category parameter.
Now if you add a new variable to the formatter like my new ``%(category)s``, you get a nifty
``KeyError`` saying there is no ``category`` in the format expansion dictionary. If you add
``category='auth'`` to the ``app.logger.info()`` calls and its cousins, it’s fine, because these
methods use the magic ``**kwarg`` argument to swallow it. Everything goes well until control
arrives to the ``_log()`` method: it complains about that extra ``category`` keyword argument.
Taking a peek at Python’s internals, I found two things: ``info()`, `error()``, and co. pass
``*args`` and ``**kwargs`` to ``_log()`` unmodified, and the ``_log()`` method doesn’t have
``**kwargs`` present in its definition to swallow it. A little doc reading later I found that if
I want to pass extra arguments for such a formatter, I should do it via the ``extra`` keyword
argument to ``_log()``. A call like ``app.logger.info('invalid password', extra={'category':
'auth'})`` solved the problem. Now *that* is tedious.
My first idea was to override all the standard logging methods like ``info()`` and ``error()``,
and handle ``category`` there. But this resulted in lots of repeating code. I changed the
specification a bit, so my calls would look like ``info('message', category='auth)`` instead of
the original plan of ``info('auth', 'message')``: as the logging methods pass all keyword
arguments to ``_log()``, I can handle it there. So at the end, my new logger class only patches
``_log()``, by picking out ``category`` from the kwarg list, and inserting it to ``extra`` before
calling ``super``.
As you can see, this is a bit ugly solution. It requires me, the app
author, to know about Flask’s internals (that I can set my own logging class
before the app is created, and so the app will use it.) This means if Flask
developers change the way how logging is done, I have to adapt and find a
workaround for the new version (well, unless they let me directly set the
logging class. That would make it easy.)
What is worse, I must know about Python internals. I know the ``extra`` kwarg is documented well
(I just failed to notice), but this made adding a new formatter variable so hard. Python version
doesn’t change as often as Flask version in this project, and I think the standard library won’t
really change until 4.0, so I don’t think my tampering with a “protected” method will cause any
trouble in the future. Still, this makes me feel a bit uneasy.
All the above can be circumvented if this class, and the whole solution have some tests. As my
class uses the same method as Flask (use ``logging.getLoggerClass()`` as a base class instead of
using ``logging.Logger()`` directly), if the base logger class changes in Python or in the running
environment, my app won’t care. By checking if the app logger can use my special ``category``
variable (ie. it doesn’t raise an exception *and* the category actually gets into the log output),
I made sure my class is used as a base in Flask, so if they change the way they construct
``app.logger``, I will know about it when I first run my tests after upgrading Flask.
If you are interested in such functionality (and more), you can grab it from `GitHub
<https://github.com/gergelypolonkai/flask-logging-extras>`_, or via `PyPI
<https://pypi.python.org/pypi/Flask-Logging-Extras/>`_.

29
content/blog/2017-10-02-add-sysadmin-day-to-emacs-calendar.rst Normal file
View File

@@ -0,0 +1,29 @@
Add SysAdmin day to Emacs Calendar
##################################
:date: 2017-10-02T09:37:52Z
:category: blog
:tags: emacs
:url: 2017/10/02/add-sysadmin-day-to-emacs-calendar/
:save_as: 2017/10/02/add-sysadmin-day-to-emacs-calendar/index.html
:status: published
:author: Gergely Polonkai
I’m a SysAdmin since 1998. Maybe a bit earlier, if you count managing our home computer. This
means `SysAdmin Day <http://sysadminday.com/>`_ is also celebrating me. However, my Emacs
Calendar doesn’t show it for some reason.
The solution is pretty easy:
.. code-block:: lisp
(add-to-list 'holiday-other-holidays '(holiday-float 7 5 -1 "SysAdmin Day") t)
Now invoke :kbd:`M-x holidays-list` for any year, choosing “Other” as the category, and there you
go:
.. code-block:: log
…
Friday, July 28, 2017: SysAdmin Day
…

54
content/blog/2018-06-26-please-welcome-calendar-social.rst Normal file
View File

@@ -0,0 +1,54 @@
Please welcome Calendar.social
##############################
:date: 2018-06-26T05:36:00Z
:category: blog
:tags: development
:url: 2018/06/26/please-welcome-calendar-social/
:save_as: 2018/06/26/please-welcome-calendar-social/index.html
:status: published
:author: Gergely Polonkai
I started looking at decentralised/federated tools some years ago, but other than Matrix I didn’t
use any of them until recently. Then this February I joined the Fediverse (federated universe) by
spinning up my own `Mastodon <https://joinmastodon.org/>`_ instance. I’m not going to lie, this
place is pure awesome. I follow only 53 people but unlike on Twitter or Facebook, I can’t just
scroll through my timeline; I have to read them all. These 53 accounts are real persons over the
Internet with meaningful posts. I could never find this on the noisy Twitter or Facebook timeline
during the last 10 or so years.
Bragging aside, and given my strive for learning, I wanted to give back something to this
community. I thought about an image sharing platform where people can share their photo albums
with each other, but I realised I’m not that good at image handling. So I decided to make a
calendar instead.
My app, creatively codenamed Calendar.social, aims to be a calendar service similar to Google
Calendar (and, obviously, any calendar app you can find out there) but instead of using emails, it
will use ActivityPub to share all the details (although I might add e-mail support sooner or
later.)
I have a *lot* on my mind about what this tool should/could do when it’s done. In no particular
order, here’s a list of them:
* events that can be private (only you and the (optional) guests see them), or public (anyone can
see them). They will have all the usual fields like start/end time, location, and maybe an icon
and a cover photo
* multiple calendars you would expect from a calendar app. This way you can separate your
birthday reminders from the dentist appointments
* event sharing over ActivityPub and other channels (to be decided, but I think you can take email
and maybe Matrix for granted.)
* full calendar sharing. The other party can get access from a very basic free/busy level to full
write access (which might be a good idea for family or company wide calendars.)
* Holiday calendars that store national/bank holidays. Users can subscribe to them to see the
holidays of a given country/area, and optionally set them as busy (on holiday weekdays) or free
(on weekends that are actually workdays for some reason.)
* Reminders! Because you obviously don’t want to forget the birthday of your significant other,
your anniversary, or your barber appointment.
* All this developed with time zones, localisation, and accessibility in mind.
That, and anything more that comes to my mind.
You can follow the development `here <https://gitea.polonkai.eu/gergely/calendar-social>`_. Also,
feel free to ping me with your ideas on my `Mastodon account <https://social.polonkai.eu/@gergely>`_, `Matrix
<https://matrix.to/#/@gergely:polonkai.eu>`_, or any other channels you can find under the
“Contact me” menu.

37
content/blog/2018-07-19-recurring-events.rst Normal file
View File

@@ -0,0 +1,37 @@
Recurring events are hard
#########################
:date: 2018-07-19T13:22:00Z
:category: blog
:tags: development
:url: 2018/07/19/recurring-events/
:save_as: 2018/07/19/recurring-events/index.html
:status: published
:author: Gergely Polonkai
It was almost a month ago when I `announced
<{filename}2018-06-26-please-welcome-calendar-social.rst>`_ the development of Calendar.social.
Since then I’m over some interesting and some less interesting stuff; (web) development, after
all, is just a recurrence of patterns. Speaking of recurrence, I arrived to a really interesting
topic: recurring events.
My initial thought was like “oh, that’s easy! Let’s insert all future occurences as a separate
``Event`` object, linking to the original one for the details. That makes handling exceptions
easy, as I just have to update/delete that specific instance.” Well, not really. I mean, an
event repeating daily *forever* would fill up the database quickly, isn’t it? That’s when I
decided to look how other projects do it.
As it turns out, my first thought is about the same as everyone else has their mind, with about
the same reasons. Then, they usually turn down the idea just like I did. And instead, they
implement recurrence patterns and exception patterns.
My favourite is `this article
<https://github.com/bmoeskau/Extensible/blob/master/recurrence-overview.md>`_ so far. The author
suggests to use the recurrence patterns specced by `RFC2445
<http://www.ietf.org/rfc/rfc2445.txt>`_ (the spec for the iCalendar format). The interesting part
in this solution is how to query recurring events: you simply store the timestamp of the last
occurence of the events (or, if the event repeats forever, the greatest timestamp your database
supports.)
Choosing the maximum date seemed to be the tricky one, but it turned out both Python and popular
SQL backends support dates up to the end of year 9999.

220
content/blog/2018-07-26-check-if-last-git-commit-has-test-coverage.rst Normal file
View File

@@ -0,0 +1,220 @@
Check if the last Git commit has test coverage
##############################################
:date: 2018-07-26T12:49:52Z
:category: blog
:tags: python,development,testing
:url: 2018/07/26/check-if-last-git-commit-has-test-coverage/
:save_as: 2018/07/26/check-if-last-git-commit-has-test-coverage/index.html
:status: published
:author: Gergely Polonkai
I use Python at work and for private projects. I also aim to write tests for my code, especially
recently. And as I usually don’t start from 100% code coverage (TDD is not my game), I at least
want to know if the code I just wrote have full coverage.
The trick is to collect all the lines that changed, and all the lines that has no coverage. Then
compare the two, and you have the uncovered lines that changed!
Getting the list of changed lines
=================================
Recently, I bumped into
`this article <https://adam.younglogic.com/2018/07/testing-patch-has-test/>`_. It is a great awk
script that lists the lines that changed in the latest commit. I have really no problem with awk,
but I’m pretty sure it can be done in Python, as that is my main language nowadays.
.. code-block:: python
def get_changed_lines():
"""Get the line numbers that changed in the last commit
"""
git_output = subprocess.check_output('git show', shell=True).decode('utf-8')
current_file = None
lines = {}
left = 0
right = 0
for line in git_output.split('\n'):
match = re.match(r'^@@ -([0-9]+),[0-9]+ [+]([0-9]+),[0-9]+ @@', line)
if match:
left = int(match.groups()[0])
right = int(match.groups()[1])
continue
if re.match(r'^\+\+\+', line):
current_file = line[6:]
continue
if re.match(r'^-', line):
left += 1
continue
if re.match(r'^[+]', line):
# Save this line number as changed
lines.setdefault(current_file, [])
lines[current_file].append(right)
right += 1
continue
left += 1
right += 1
return lines
OK, not as short as the awk script, but works just fine.
Getting the uncovered lines
===========================
Coverage.py can list the uncovered lines with ``coverage report --show-missing``. For
Calendar.social, this looks something like this:
.. code-block:: log
Name Stmts Miss Cover Missing
----------------------------------------------------------------------
calsocial/__init__.py 173 62 64% 44, 138-148, 200, 239-253, 261-280, 288-295, 308-309, 324-346, 354-363
calsocial/__main__.py 3 3 0% 4-9
calsocial/account.py 108 51 53% 85-97, 105-112, 125, 130-137, 148-160, 169-175, 184-200, 209-212, 221-234
calsocial/app_state.py 10 0 100%
calsocial/cache.py 73 11 85% 65-70, 98, 113, 124, 137, 156-159
calsocial/calendar_system/__init__.py 10 3 70% 32, 41, 48
calsocial/calendar_system/gregorian.py 77 0 100%
calsocial/config_development.py 11 11 0% 4-17
calsocial/config_testing.py 12 0 100%
calsocial/forms.py 198 83 58% 49, 59, 90, 136-146, 153, 161-169, 188-195, 198-206, 209-212, 228-232, 238-244, 252-253, 263-267, 273-277, 317-336, 339-342, 352-354, 362-374, 401-413
calsocial/models.py 372 92 75% 49-51, 103-106, 177, 180-188, 191-200, 203, 242-248, 257-268, 289, 307, 349, 352-359, 378, 392, 404-409, 416, 444, 447, 492-496, 503, 510, 516, 522, 525, 528, 535-537, 545-551, 572, 606-617, 620, 652, 655, 660, 700, 746-748, 762-767, 774-783, 899, 929, 932
calsocial/security.py 15 3 80% 36, 56-58
calsocial/utils.py 42 5 88% 45-48, 52-53
----------------------------------------------------------------------
TOTAL 1104 324 71%
All we have to do is converting these ranges into a list of numbers, and compare it with the
result of the previous function:
.. code-block:: python
def get_uncovered_lines(changed_lines):
"""Get the full list of lines that has not been covered by tests
"""
column_widths = []
uncovered_lines = {}
for line in sys.stdin:
line = line.strip()
if line.startswith('---'):
continue
if line.startswith('Name '):
match = re.match(r'^(Name +)(Stmts +)(Miss +)(Cover +)Missing$', line)
assert match
column_widths = [len(col) for col in match.groups()]
continue
name = [
line[sum(column_widths[0:idx]):sum(column_widths[0:idx]) + width].strip()
for idx, width in enumerate(column_widths)][0]
missing = line[sum(column_widths):].strip()
for value in missing.split(', '):
if not value:
continue
try:
number = int(value)
except ValueError:
first, last = value.split('-')
lines = range(int(first), int(last) + 1)
else:
lines = range(number, number + 1)
for lineno in lines:
if name in changed_lines and lineno not in changed_lines[name]:
uncovered_lines.setdefault(name, [])
uncovered_lines[name].append(lineno)
return uncovered_lines
At the end we have a dictionary that has filenames as keys, and a list of changed but uncovered
lines.
Converting back to ranges
=========================
To make the final result more readable, let’s convert them back to a nice ``from_line-to_line``
range list first:
.. code-block:: python
def line_numbers_to_ranges():
"""List the lines that has not been covered
"""
changed_lines = get_changed_lines()
uncovered_lines = get_uncovered_lines(changed_lines)
line_list = []
for filename, lines in uncovered_lines.items():
lines = sorted(lines)
last_value = None
ranges = []
for lineno in lines:
if last_value and last_value + 1 == lineno:
ranges[-1].append(lineno)
else:
ranges.append([lineno])
last_value = lineno
range_list = []
for range_ in ranges:
first = range_.pop(0)
if range_:
range_list.append(f'{first}-{range_[-1]}')
else:
range_list.append(str(first))
line_list.append((filename, ', '.join(range_list)))
return line_list
Printing the result
===================
Now all that is left is to print the result on the screen in a format digestable by a human being:
.. code-block:: python
def tabular_print(uncovered_lines):
"""Print the list of uncovered lines on the screen in a tabbed format
"""
max_filename_len = max(len(data[0]) for data in uncovered_lines)
for filename, lines in uncovered_lines:
print(filename.ljust(max_filename_len + 2) + lines)
And we are done.
Conclusion
==========
This task never seemed hard to accomplish, but somehow I never put enough energy into it to make
it happen. Kudos to Adam Young doing some legwork for me!