[haiku-development] Re: [PATCH] libroot improvements

  • From: "Axel Dörfler" <axeld@xxxxxxxxxxxxxxxx>
  • To: haiku-development@xxxxxxxxxxxxx
  • Date: Fri, 27 Feb 2009 16:05:17 +0100 CET

Hi Artur,

Artur Wyszynski <aljen-mlists@xxxxxxx> wrote:
> looking at current libroot status i noticed a mix composed by 
> different
> versions of glibc with some features enabled/disabled and some ports 
> of 
> bsd code there.

That pretty much fits to the unfortunate current status quo.

> Then i removed completly glibc and friends mixes from libroot and 
> started porting libc and Sun math library from FreeBSD.
> Now it's working as expected, with full posix support (even c99), 
> wide 
> chars enabled, support for locales, etc.
> Don't worry, it won't break binary compatibility, because it's ported 
> & 
> enabled only for gcc4 build, gcc2 is still using old
> libroot+glibc&friends.

That sounds nice, and I think is the way we should proceed more or less 
IMO, anyway. Though we might also want to support some glibc 
extensions, too (regexp comes to mind).
Only the locale stuff will likely get changed when we develop our 
locale kit.

> Current status:
> * old libroot is moved from src/system/libroot to 
> src/system/libroot/compatibility and gcc2 builds it from there
> * new libroot is now in src/system/libroot
> * headers/posix is moved to headers/compatibility/beos/posix and they 
> are used in buildtools scripts to instruct gcc2
> where headers are and in build/jam/HeadersRules where they are when 
> building Haiku
> * new posix headers are in headers/posix

Is there any reason why you moved the whole libroot, and not just the 
parts affected by your work, most notably the "posix" subfolder?
Also, you included lots of non-POSIX folder into the POSIX directory 
(like arpa/ftp.h). I don't really like having them there, as long as 
there is a different solution that doesn't hinder ports too much.
I don't really understand your changes to Errors.h.

> Things noticed by me (and not only, thx for testing guys :P):
> * system is noticable faster compared to other builds, i mean gui and 
> boot; for virtual box on my machine (dont know how others and i 
> didn't
> tested this on real hardware yet) boot time was decrased from 24-26 
> seconds to 14 (don't know the reason)
> * gui is more responsible, i think it's because new sun math library

That doesn't really sound very likely, though. I would be happy with 
benchmarks proving me wrong, though :-)


Other related posts: