main inclusion report for eina

Packaging looks good overall.

Random minor things:
- copyright still mention the SVN for the download location but looking at the version and watch files seems to imply that it's not a SVN snapshot anymore
- DEB_DH_STRIP_ARGS is probably not needed anymore with recent cdbs (should pick up proper name automatically)
- some explanation for -O3 would be welcome, this looks like just a regular toolkit which we usually build with the default -O2
- the whole SONAME/package name handling is quite ugly but is a consequence from upstream's handling
- doc installation could use dh_install
- shlibs would better be set without hardcoding the SONAME in the shlibs file, by using the CDBS var for the dh_mkshlibdeps calls

FYI this is actually not the latest upstream version (0.0.2.062).

I'm a bit scared that it detects MMX/SSE/altivec CFlags on the buildd to optimize the package; it means that people without MMX or SSE or Altivec will get SIGILL when using the binaries. It seems to be the flags aren't used in the package, but would you mind double checking? It would be best to remove the checks if they are useless or use an override to disable this support even if it's detected.

Is it also normal that the package is built without libevas support?

Tests aren't run during build.

Doc is built twice during build on i386.

Code looks ok; only did a quick check.

So I'd just like some clarifications on the SIMD checks and libevas and woud love if some of the build issues (doc built twice and testsuite not run) would be addressed; thanks!

Hi Loic,

I'm part of EFL team, so you can consider these "official" replies:

 - -O3 is recommended as this is a data types library, so needs to be high performance. That said, it's not mandatory and I know -O2 may be on par with it or -Os be even better on embedded systems. Feel free to revert it to -O2 or even per-platform settings based on benchmark results.
  - yes, SONAME is weird, but doing that until it sees first 1.0 official releases, after that it will go smoother :-) That should happen before year's end.
  - 0.0.2.062 is the last release, you're correct.
  - mmx/sse are just detected but not used at the moment. Just as in other packages (evas itself), it will use the correct code for platforms that do not support it.
  - evas as well as glib is not required for usage. These are just required to run benchmarks against those libraries and compare how eina data types will perform using the same work load.
  - test suite does not need to be run, it's a developer thing, not meant for packagers to run. If you want to run it to ensure it runs fine on platforms, okay, but IMHO it will only consume build bot CPU.

I'll subscribe Albin Tonnerre, the one upstreaming EFL into Debian so he can share his thoughts as well.

Thanks,

Ok; I'd appreciate if you'd run the testsuite during build and switch to -O2; good to go in main; there are some other easy fixes to pick from my list of minor things :-)

Some of them are already done in my git, it's just that I'm lacking time at the moment. I'll send a sync request for the .062 series once the fix is uploaded.

Regards,
Albin

Uploaded fixes requested by lool. Debdiff attached incase its useful to Albin.

Package moved to main

This bug was fixed in the package eina - 0.0.2.061-1ubuntu1

---------------
eina (0.0.2.061-1ubuntu1) karmic; urgency=low

  * Don't built with -O3
  * Build test suite
  * Update debian/copyright
  * Use dh_install for docs install
  * No need to build docs twice
  * No need to set DEB_DH_STRIP_ARGS any more
  * Closes LP: #405379

 -- Jonathan Riddell <email address hidden> Mon, 24 Aug 2009 14:16:03 +0100

better debdiff which actually runs the test suite

I'm not sure you really want to run the testsuite, actually. It has always been known to fail when the 'magic debug' feature is disabled, which is the case here. Thanks for the debdiff, anyway

Regards,
Albin