[glob2-devel] Re: Patches for Globulation 2 and things about 0.8.22 release.

[Stéphane Magnenat]

Hello,

This is my answer to a mail from David Martínez Moreno, the debian maintainer 
of glob2. I cc it to the ml so that the info circulate.

>       Hello, Stéphane.  I have several things for you. :-)

Thanks. While I still contribue to glob2, I'm not the official maintainer 
anymore. "Bradley Arsenault" <address@hidden> is. Problem is, he seems to 
have disappeared just after last release, he is probably in holidays. The 
best for reporting such problems would be the development mailing list.
For now, I will answer your points. Thanks a lot for you help :-)

>       Please do not take none of this personally.  As for you (I guess), 
> English
> is not my mother tongue.  I like to go straight to the point, so do not get
> angry. ;-)

No problem, glob2 has several problems, all of them related to the lack of 
developers which induce a critical lack of testing.

>       First of all, I saw the advertisement in the download page about the
> released tarball and the subsequent recommended CVS pull of 'alpha22' tag. 
> While not unconfortable of fiddling with CVS, I ran a complete diff against
> your tree and the only diff I could find was the commented assert() in
> ConfigFiles.h. It should be *FAR* better to simply remove the tarball, and
> replace with the one with the fixed file.  But I have a couple of further
> complaints (below).

I fully agree. We are discussing it right now. Hopefully once Bradley appears 
again the bug will be solved.

>       Several other things that I found in the 0.8.22 tar.gz:
>
> - You keep leaving CVS/ directory under src/ (and you shouldn't, given that
> it is a source release, without revision information).  We (packagers) must
> remove it and repackage the tar.gz.

I don't understand why this directory got included in the tar.gz

> - Maps were not included.
> - Campaigns were not included.

This is a bug, teh people who did the deb forgot to do the syncdata syncmaps

>       Then I found how you sync the maps and campaign data: rsync.  Why don't
> you put that data in CVS?  Now I have maps and campaigns from 20070407,
> that I can *only* suppose that will work well with 0.8.22, as they are
> newer.  Worse than that, I could have decided to package glob2 in a month,
> so I would have maps and campaign data much newer than (and why not,
> incompatible with) the original ones that should have been shipped with the
> original tarball.

The reason was not to overload savannah cvs with binary files. We are 
currently thinking to switching to a more distributed source management 
system.

>       I know that you have been doing a lot of work and going through lots of
> problems (for example, the server's death).  But I think that you should
> take the original tarball, add the fix, maps and campaigns, and put that
> one for release (or make a 0.8.22.1 point release if it is much trouble for
> SF to replace files).  I volunteer for that task, if you want, as I have a
> custom tar.gz for glob2-0.8.22 with latest maps and campaigns.

That is nice. I would like to have a bug fixed release asap but in the mean 
time, if you register on savannah you'l be able to upload your custom .tar.gz 
in place of the other one there.

>       That said, I have several patches for you (attached):
>
> - 01_remove_home_directory_browsing:
> Isolated fix for Debian bug #408733.
> The problem is that you are putting some local directory path *in compile
> time* as a string inside the sources (and thus the binary), so the user
> playing with your compiled binary have information about your system.  The
> problem here is not leaking information, but having in the first place to
> look when browsing files the packager's compile directory.  With
> automounter, that is terrible, as you see lots of system errors from trying
> to
> mount /home/ender/whatever.  Treat it as you wish.

For now, I leave it as it is.

> - 02_gcc-4.3_fixes:
> Fix to compile Globulation 2 with GCC 4.3.
> The package fails to build with GCC 4.3.  Version 4.3 has not been released
> yet but I'm building with a snapshot in order to find errors and give
> people an advance warning.  In GCC 4.3, the C++ header dependencies have
> been cleaned up.  The advantage of this is that programs will compile
> faster.  The downside is that you actually need to directly #include
> everything you use (but you really should do this anyway, otherwise your
> program won't work with any compiler other than GCC).  Some background of
> this can be found at http://gcc.gnu.org/PR28080.  Patch by Martin
> Michlmayr.

Applied to HEAD, thanks.

> - 03_freedesktop_compliance:
> Fix to glob2.desktop in order to make it compliant with freedesktop
> standards.

Applied to HEAD, thanks.

>       The two next patches patch directly 'configure', but the concept should 
> be
> clear.
>
> - 04_CXXFLAGS_abuse:
> You forced to use -O3 -march=i686 as CXXFLAGS, so I had to disable it.  We
> use 12 architectures, and I found -O3 and -march=i686 excessive and
> i386-centric. Treat it as you wish.

Already fixed in HEAD.

> - 05_remove_vorbis_useless_dependency:
> checklib (a Debian tool) sorted out a useless direct dependency on
> libvorbis (in fact it is accesed through libvorbis{file,enc}).  So I
> removed from configure.  But I suspect that this dependency is generated
> from something else in the system, and not your fault...so you can ignore
> me on this one. :-)

It is, from configure.in
XIPH_PATH_VORBIS macro from vorbis package generates those lines. Patch not 
applied.

>       And given that you lack a manpage, I took the liberty to write one.  I
> send it to you attached as well.

Thanks. In fact there was already one in debian/ subdirectory, but yours is 
more recent. Perhaps a fuse of both would be good? Do you have the lines I 
should add to configure.in to have it installed automaticaly?

Steph

-- 
http://nct.ysagoon.com