[Bf-committers] Meeting minutes, nov 7, 2010

[Campbell Barton]

On Sun, Nov 7, 2010 at 5:13 PM, Ton Roosendaal <ton at blender.org> wrote:
> Hi,
>
> 1) Current projects
>
> - Campbell Barton has a patch in the tracker to place background
> images with empty objects.
>
> - Meeting shortly discussed progress in particle branch, which
> involves node-based particle editing too.
>
> 2) Blender 2.5x topics
>
> - Janne Karhu proposed to make pointcache baking (like for particles)
> threaded, but he was getting thread issues with scene drawing, mostly
> due to an ancient global variable.
>
> Meeting then went into review of how threaded render works in general,
> which is not stable yet. Brecht reminded that there's a "Debug" test
> 101 for it (CTRL+ALT+D, set to 101), which copies the entire database
> (but there's issues with Images still).
> Ton confirms this needs further research, and is defintely a 2.5 target.
>
> - Andrea Weikert noticed 'noise depth' on Perlin-noise based textures
> had a too low maximum setting. Will be fixed. Example: http://www.pasteall.org/pic/show.php?id=6856
>
> - Campbell mentions there are too many "Blender doesn't start" bug
> reports. He suspects gfx cards to be the issue, and proposes to get a
> couple of different cards to configure systems at home with.
>
> - Nathan Letwory suspects that (for Windows) a "doesn't start" bug is
> also related to failing installer issues. The installer doesn't handle
> priviledged vs nonpriviliged files well, and there are problems with
> installation path finding code.
> Maybe we should drop the installer, and just provide a
> register_blender.bat or install.bat?
>
> - Campbell and Thomas Dinges will look into removing unused files from
> svn.
>
> - Campbell proposes to ditch old Makefiles, Ton protests, but will
> test Cmake in due time. Andrea mentions Cmake to MSVC has a nasty bug,
> creating directories for every C file.
>
> - Question: is anyone even supporting Irix for 2.5x?
>
> -Ton-

For the record (ie. in-case Ton changes his mind). just did some tests
with CMake Makefiles vs NAN-Makefiles.
Single threaded build:
 CMake: 4min, 18 seconds.
 NAN-Makefiles: 7min, 45seconds.

More worrying is that nan-makefiles don't seem to be able to do
threaded builds, I couldn't even get "make -j2" to build without
continually re-running, after 8 or so times I gave up, it finished ok
single threaded.

CMake Makefiles on my 6 core AMD, system: 1min, 0.07 seconds.

Ofcourse its not all about speed but slower + less-reliable isn't a
great combination, esp with most systems multi-core these days.