Re: [Boost-build] [UBLAS] [RC_1_34] Reason for large scale test failure

From: Vladimir Prus (ghost_at_[hidden])
Date: 2006-06-08 01:36:24

On Wednesday 07 June 2006 16:28, Markus Schöpflin wrote:
> Here is the log of a manual execution of the UBLAS test1 test:
>
....
> It looks like BBv2 somehow forgets the extra test1 when generating the name
> for the executable. IOW, instead of generating .../debug/test1/test1 it
> generates .../debug/test1 which of course collides with the directory name
> of the test which happens to be .../debug/test1.

In fact, V2 always creates the executable in the build directory, ignoring any 
relative paths on sources. In fact, you can have two source files from 
different directories -- how can we decide where to put the executable?

The previous logic of V2 was this:

* 'test1' binary is places to build dir.
* 'test1.o' object is placed to build dir.

Recently, on HEAD I've comitted a patch that makes V2 preserve relative path 
of source files, so now we try to put 'test1.o' into $(build_dir)/test1 
subdir, but 'test1' is already an executable.

I tend to believe that UBLAS tests are poorly structured, but it's bad time to 
argue about it, so I've reverted that "preserve relative paths" patch, and 
will look at it again after 1.34.

Thanks for diagnosing and sorry for the breakage!

- Volodya

-- 
Vladimir Prus
http://vladimir_prus.blogspot.com
Boost.Build V2: http://boost.org/boost-build2