Re: Patch for i386-darwin: test suite results

[Andreas Tobler <toa at pop dot agri dot ch>]

 Hi Sandro,

Sandro Tolaini wrote:
> On 11/mar/2006, at 12:32, Andrew Haley wrote:
>
> > It isn't, no.  All the gij test failures probably mean that libffi
> > isn't working.
>
> Ok, I spent the last days fixing libfff, that now passes its testsuite. 

All tests do pass? Please send the diffs to me!

> Obviously, the libjava testsuite is now going better:

Please do a patch run for each package on it's own. Means one for gcc, 
one for boehm-gc, one for libffi and one for libjava/gcj. It makes 
review easier. Also, do not include *.in or configure files in the diff. 
Only Makefile.am or configure.ac the rest we can generate on our own. 
Comment files are also not needed.

Regarding boehm-gc, please follow the same rule.

My preference would be that you would do the work on svn head, aka 4.2. 
If we see that it makes sense, we will backport to 4.1.


>                 === libjava Summary ===
>
> # of expected passes            3727
> # of unexpected failures        43
> # of expected failures          14
> # of untested testcases         59

Looks ok, but the mountain is very high :)
> Some tests still fail, but in the next days I don't have time to look 
> into them further. I'm attaching the patch for who wants to dig into it. 
> Note that currently can_unwind_signal=no for this platform. About the 
> patch:

The unwind stuff will cause you some severe work.

> * It's based on gcc 4.0.2, but should apply cleanly to 4.0.3 and maybe 
> 4.1.0. Will not apply as-is to 4.2, because the damned libffi changed 
> quite a bit (there are more parts written in assembly that will require 
> porting/tweaking).
>
> * A large part of the patch is to upgrade boehm to a later version. If 
> you want, you can import boehm a recent boehm 6.7 or 7.0, that supports 
> Darwin/i386. If you apply the patch as-is, you will get the boehm-gc 
> that sits currently on the gcc trunk, plus a couple of patches to 
> support Darwin/i386.
>
> Many thanks goes to Eric Albert for his help with the Darwin/i386 ABI.

Andreas