Skip to contentSkip to conversation

Solving the mesa3d build issue

Buildroot Thomas Petazzoni 3 participants 4 messages Nov 17, 2012 – Nov 22, 2012

lore
Share & tools
lore
#0 lore raw

Hello,

The mesa3d build is currently broken because it generates during its build process a bunch of C files from a XML description of the OpenGL APIs. This generation process is done through Python scripts that require the libxml2 Python bindings to be installed.

Unfortunately, those libxml2 Python bindings are part of the libxml2 source code itself, so we would have to enable host-python as a dependency of host-libxml2 in order to get those Python bindings built and installed. This means that all other users of host-libxml2 would pay the price of building host-python even if they don't need it, which is quite annoying.

We have several solutions to address this problem:

(1) The solution proposed by Will Wagner/Noel Vellemans at
http://lists.busybox.net/pipermail/buildroot/2012-November/061594.html
and
http://lists.busybox.net/pipermail/buildroot/2012-November/061595.html.

This solution consists in adding a BR2_PACKAGE_HOST_LIBXML2_PYTHON hidden configuration symbol, that mesa3d selects. It sells the host-libxml2 package that it should build the Python binding, and therefore depend on host-python.

It is quite nice in the sense that it is simple and only builds host-python as a dependency of host-libxml2 when needed. However, it adds something entirely non-standard: a hidden kconfig option that tunes the compilation of a host package. Do we want to do this?

(2) Workaround the problem by adding the generated C files as patches to the mesa3d package so that generating them at build time is not needed. Unfortunately, those generated files are quite large, so I'm not sure we want them as patches. And it also means they would have to be updated whenever we bump the mesa3d version.

-rw-r--r-- 1 thomas thomas 181158 Nov 17 13:02 api_exec_es1.c -rw-r--r-- 1 thomas thomas 94023 Nov 17 13:02 api_exec_es2.c

(3) Rewrite the small Python script that generates those C files so that it uses the XML APIs available in the standard Python library instead of the libxml2 XML APIs. This is not too complicated, but doing it on the old 7.10.1 version we're using seems like a waste of time (we'd better do it on a more recent version so that it could maybe be upstreamed at some point). I've tried bumping mesa3d. It is certainly doable, but requires the bumping of quite a few other packages.

We really need to solve that problem, so I would be grateful if you could comment on it. Peter?

Thanks!

Thomas

Signature · 4 lines
Thomas Petazzoni, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com
#1 lore raw
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Thomas" == Thomas Petazzoni <[email protected]> writes:

Hi,

Thomas> We really need to solve that problem, so I would be grateful if you Thomas> could comment on it. Peter?

As discussed on irc, option 1 seems the best option until we can go for 3.

Signature · 1 line
Bye, Peter Korsgaard
#2 lore raw
In reply to Thomas Petazzoni
Quoted 26 lines · Thomas Petazzoni

Hello,

The mesa3d build is currently broken because it generates during its build process a bunch of C files from a XML description of the OpenGL APIs. This generation process is done through Python scripts that require the libxml2 Python bindings to be installed.

Unfortunately, those libxml2 Python bindings are part of the libxml2 source code itself, so we would have to enable host-python as a dependency of host-libxml2 in order to get those Python bindings built and installed. This means that all other users of host-libxml2 would pay the price of building host-python even if they don't need it, which is quite annoying.

We have several solutions to address this problem:

(1) The solution proposed by Will Wagner/Noel Vellemans at
http://lists.busybox.net/pipermail/buildroot/2012-November/061594.html
and
http://lists.busybox.net/pipermail/buildroot/2012-November/061595.html.

This solution consists in adding a BR2_PACKAGE_HOST_LIBXML2_PYTHON hidden configuration symbol, that mesa3d selects. It sells the host-libxml2 package that it should build the Python binding, and therefore depend on host-python.

Quoted 6 lines

It is quite nice in the sense that it is simple and only builds host-python as a dependency of host-libxml2 when needed. However, it adds something entirely non-standard: a hidden kconfig option that tunes the compilation of a host package. Do we want to do this?

[snip (2) and (3)]

A fourth solution would be to create a new host-only package libxml2-python
that builds libxml2 with python bindings. It would just reuse the libxml2 package with a slightly different CONF_OPT.

The differences with the currently committed solution (1) are:

- No need for a HOST-config symbol.

- host-libxml2 will be built twice, once without and once with python bindings.

There may even be a workaround for the second step, but at the moment I assume
it will be built twice.

What do you think?

Regards,
Arnout

Signature · 6 lines
Arnout Vandecappelle                               arnout at mind be
Senior Embedded Software Architect                 +32-16-286540
Essensium/Mind                                     http://www.mind.be
G.Geenslaan 9, 3001 Leuven, Belgium                BE 872 984 063 RPR Leuven
LinkedIn profile: http://www.linkedin.com/in/arnoutvandecappelle
GPG fingerprint:  7CB5 E4CC 6C2E EFD4 6E3D A754 F963 ECAB 2450 2F1F
#3 lore raw

Arnout,

Quoted 6 lines · Arnout Vandecappelle

A fourth solution would be to create a new host-only package libxml2-python
that builds libxml2 with python bindings. It would just reuse the libxml2 package with a slightly different CONF_OPT.

The differences with the currently committed solution (1) are:

Quoted 8 lines

- No need for a HOST-config symbol.

- host-libxml2 will be built twice, once without and once with python bindings.

There may even be a workaround for the second step, but at the moment I assume
it will be built twice.

What do you think?

This is indeed another solution. But just like the kconfig knob for a host package creates a precedent, having two separate binary packages for one single source package is also creating another precedent. I'm not sure which is not is better, really.

Or maybe it's time for Buildroot to make the distinction between source packages and binary packages? But this is going to be a very large change.

Best regards,

Thomas

Signature · 4 lines
Thomas Petazzoni, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com