Category: Tips

Auto-Detect Dependencies when Building .debs Using CMake

CMake (via CPack) as a great feature that allows you to automatically generate Debian/Ubuntu (.deb) packages. One of the annoying things to do when you create a package is listing its dependencies. CMake asks you do it via the CPACK_DEBIAN_PACKAGE_DEPENDS variable. For example:

set (CPACK_DEBIAN_PACKAGE_DEPENDS "libc6 (>=2.7-18)")

But what happens when you work on a more complex project? Keeping track of all the dependencies by hand is a tedious task. Debian provides a tool named dpkg-shlibdeps which makes this task easier by updating the debian/control file with dependencies extracted from the dynamic libraries needed by a given executable. Luckily since CMake 2.8.3, CMake also supports running this tool automatically to figure out the required dependencies. The documentation is sparse, and I had hard time finding how to do so (I actually found it via a bug report and a commit message, but afterwards I’ve seen it the official documentation too). To enable it, you need to add the following line to your CMakeLists.txt file:

# autogenerate dependency information
set (CPACK_DEBIAN_PACKAGE_SHLIBDEPS ON)

Conditional Compilation in Autoconf and Automake

While working on my audio based random password generator (you view the source in github), I wanted to do some conditional compilation: Compiling certain parts of the program only in case some option is passed to the configure script. As it usually happens with GNU’s autotools, it kind of hell to do it. Documentation is spread across dozens of sources, each provides only a specific part of what to do. I’m writing it here in the blog, in hope I’ll never have to search how to do so again.
Continue reading Conditional Compilation in Autoconf and Automake

Fixing virtualenv after Upgrading Your Distribution/Python

After you upgrade your python/distribution (specifically this happened to me after upgrading from Ubuntu 11.10 to 12.04), your existing virtualenv environments may stop working. This manifests itself by reporting that some modules are missing. For example when I tried to open a Django shell, it complained that urandom was missing from the os module. I guess almost any module will be broken.

Apparently, the solution is dead simple. Just re-create the virtualenv environment:

virtualenv /PATH/TO/EXISTING/ENVIRONMENT

or

virtualenv --system-site-packages /PATH/TO/EXISTING/ENVIRONMENT

(depending on how you created it in the same place). All the modules you’ve already installed should keep working as before (at least it was that way for me).

Bye Bye OmniCppComplete, Hello Clang Complete

For years OmniCppComplete has been the de facto standard for C++ completion in Vim. But as time progressed, I got more and more annoyed by it’s shortcomings. OmniCppComplete is based on tokenizing provided by ctags. The ctags parsing of C++ code is problematic, you can’t even run it on libstdc++ headers (you need to download modified headers). You want to use an external library? You’ll need to run ctags seperatly on each library. Not to mention it’s inablity to deduce types of anything more than trivial. The core of the problem is that OmniCppComplete isn’t a compiler and you can’t expect something that isn’t a compiler to fully understand code. This what makes Visual Studio’s IntelliSense so great: it uses the Visual C++ compiler for parsing, it isn’t making wild guess at types and what is the current scope – it knows it.
Continue reading Bye Bye OmniCppComplete, Hello Clang Complete

Starting Djano upon Reboot using Cron

Three years ago I wrote about starting services as user via cron instead of init.d (specifically Trac). However this method has a serious downside: it has no support for dependencies. This really bothered me when I used cron to start a Django server, has it would attempt to load before MySQL was running. This made the cron useless, as always after rebooting I would receive an error email saying it couldn’t connect to the MySQL server, and I would have to login and start the Django Server manually. Yesterday I got sick of it, and decided to hack something that will work properly. So my crontab had the following line:

@reboot python ./manage.py runfcgi ...options...

which I changed into:

@reboot until /sbin/status mysql | grep start/running ; do echo "Mysql isn't running yet...>&2"; sleep 1; done; python ./manage.py runfcgi ...options...

Basically it loops and checks whether the MySQL service got started. If so, it would start Django as it did before. On the other hand, if MySQL isn’t running it would just sleep for a second and repeat the check.

A small issue is that if for some reason MySQL won’t start at all, it will loop forever. If this happens, it would mean that I’ll have to manually kill that cronjob, but I would have to login anyway to see what wrong with the MySQL. So, while this method can’t support dependencies like init.d does, it does provide a good-enough solution.

Update 2012-11-23: Fixed the crontab line (it would fail when mysql was in start/post-start state).

Gmail backup: getmail vs. OfflineIMAP

I’m currently reviewing my backup plans and decided it’s a good occasion to finally start backing up my Gmail account. Firstly, I didn’t seriously consider Desktop clients as the main backup tool, as they are hard to automate. The two main options are: OfflineIMAP and getamil. Both are available from Ubuntu’s repositories, so installation is easy with both and both have good tutorials, Matt Cutts’ getmail and EnigmaCurry’s OfflineIMAP.

OfflineIMAP claims to be faster, but I haven’t really checked it (and I’m not sure how important that is giving that it runs in the background). From what I saw configuring them is mainly a task of cut-and-paste, but getmail requires to list every label you want to backup, which I consider is a major downside. As both are able to save the mails to maildir format, it should be easy to back it up using duplicity.

Conclusion: This was a short comparison, mainly to guide me in choosing the right backup for me, you may have different opinions (which, of course, I would gladly hear). I finally chose OfflineIMAP, mainly due to the labels issue.

Note on desktop clients: It seems that every decent one can be configured to work with a local maildir, so you can use them to read the backups. As I prefer Gmail’s interface, I will only use desktop clients in case I’m offline, so read-only access from desktop client seems good enough for me.

GNOME_COMPILE_WARNINGS(maximum) – Syntax Error in configure

I’m still encountering migration issues from Gentoo to Ubuntu. Apperantly, Gentoo is much more user friendly than Ubuntu when it comes to compiling packages. In Gentoo you’ve got almost all the major dependencies you need. In Ubuntu, on the other hand, you need to hunt them down. It’s much easier with the main ones, as they are listed. But there are some small ones which are harder to track. I came across the following error while trying to compile gitg, a GUI for Git, today:

./configure: line 14447: syntax error near unexpected token `maximum'
./configure: line 14447: `GNOME_COMPILE_WARNINGS(maximum)'

After not so short investigation I found out I was missing gnome-common

sudo apt-get install gnome-common

Why can’t be one distribution which is user-friendly like Ubuntu and in the same time developer-friendly like Gentoo?

GCC Usage quirks

This is more of a note to myself, as it’s an error I keep bumping into. When compiling manually using gcc (or g++) you should pass all the libraries you’re compiling against at the end of the argument list. E.g.:

g++ -Wall -lboost_thread my_file.cpp

will result in error like:

undefined reference to `boost::thread::join()'
undefined reference to `boost::thread::~thread()'

while

g++ -Wall vortex5_brute.cpp -lboost_thread

works fine.

Number Exercises Separately in LyX

Say you’ve got a document with a bunch of exercises and few lemmas. You may want the exercises numbered separately from the numbering of the lemmas and theorem, unlike LyX’s default behavior. This can be achieved by redefining xca, the environment LyX uses for exercises. Add the following to your LaTeX preamble:

\let\xca\@undefined
\theoremstyle{plain}
\newtheorem{xca}{\protect\exercisename}

LyX will still display the incorrect numbering, but the output will be correct nonetheless. The first line, undefines the LyX’s definition of xca, then we set the style to match the old one and we redefine xca, this time without a reference to the theorems’ counter.