Delta XMPP Project

It may not be widely known that Psi (as well as QCA and Qt) can be built and used “in-place”. This means you configure and make, but there is no ‘make install’ step. This can be convenient if you don’t want to install these packages system-wide.

Of course, you can always just set a prefix within your home directory (or other user-writable area) when building sources if you don’t wish to install system-wide. However, in-place builds are extra convenient for the developer, as it saves the trouble of having to do ‘make install’ every time something is changed.

So while this is not anything new for Psi development, and nor is the idea particularly revolutionary, it might be something you didn’t know you could do and so I’m documenting it here. Personally, I develop everything in-place.

Qt:

Qt 3 is/was always built in-place. You extract it where you want it to end up, configure and make. There is no ‘make install’ step. With Qt 4, they changed it to behave more like other packages. Now the prefix defaults to somewhere in /usr/local, and there’s a ‘make install’. However, the old way is still supported if you pass the directory of where you extracted Qt to -prefix. That is:

$ tar zxvf qt-x11-opensource-src-[version].tar.gz
$ cd qt-x11-opensource-src-[version]
$ ./configure -prefix $PWD
$ make

This is not a hack or a way to trick Qt. It is actually a supported feature, and when Qt is done configuring you’ll notice it says:

Qt is now configured for building. Just run 'make'.
Once everything is built, Qt is installed.
You should not run 'make install'.

Another feature of Qt 4 is that qmake is aware of the installation location. This allows it to locate the necessary files it needs without QTDIR. Additionally, all subtools (e.g. moc) are invoked using their full path. This means you don’t even need the Qt 4 bin directory in your PATH. Finally, if you’re using gcc, qmake also uses rpath when linking. This means you don’t need Qt 4′s lib directory to be specified anywhere either. No need for /etc/ld.so.conf or LD_LIBRARY_PATH. You simply build Qt, go into your project directory, and run qmake.

$ cd /path/to/myproject
$ /path/to/qt/bin/qmake
$ make

QCA:

Like Qt, QCA (as of 2.0 beta2, I believe) supports in-place builds by pointing prefix to $PWD.

$ ./configure --prefix=$PWD

Note: If you’re building against an in-place Qt, and you didn’t put Qt’s bin directory in your PATH, then you’re going to have to tell configure where Qt is. Do this with –qtdir. For example:

$ ./configure --prefix=$PWD --qtdir=/path/to/qt

Unlike Qt, QCA doesn’t use rpath. This may change in the future, but for now you will need to put QCA’s lib directory in your LD_LIBRARY_PATH if you want to use it in-place:

$ export LD_LIBRARY_PATH=$PWD/lib

There’s actually one hidden secret you must do next, and it is sort of cheating. You need to copy crypto.prf into your qmake feature directory. This is so projects that need QCA can find QCA.

$ cp crypto.prf /path/to/qt/mkspecs/features

Fortunately, this file doesn’t ever change, so if you later modify QCA in-place, there’s no need to copy this file again.

QConf:

With the latest development version (as of October 22nd, 2006), qconf can be run without doing ‘make install’. In fact, you can run it from another location, and it locates its data files relative to the executable. This allows it to work similar to qmake, in that you can invoke it by specifying the full path. Unlike qmake, however, this is not dependent on the prefix. That is, you can configure qconf with the default prefix of /usr/local, and you can still run qconf from anywhere without installing it.

$ ./configure --qtdir=/path/to/qt
$ make
$ cd /path/to/myproject
$ /path/to/qconf
$ ./configure ...

Psi:

Psi can be built and used in-place. Once compiled, just run it.

$ ./configure --qtdir=/path/to/qt
$ make
$ ./psi

That’s it!

You can build and use all the Psi tools without mucking with your system, and without ever doing ‘make install’ either.