On Mon, May 23, 2011 at 11:33:42AM +0200, Eric Botcazou wrote:
> > Do you mean something like the attached one? We don't have guality
> > testsuite for ada and I think ada testcases are quite unreadable for most
> > people. Definitely to me. Plus it is always better to have a single
> > file testcase over 5 files.
>
> Sure. Last time I mentioned the guality testsuite, I was told that its
> results
> were random. Thanks for the testcase in any case.
They are far from random. There is just an issue when using new debug info
features your debugger doesn't understand yet (if you are using recentish
gdb (like 7.2 or 7.3 that should be only a problem on the trunk), Jeff Law
posted a patch for it. Other than that, the thing is just that not all
tests pass on all architectures or at all optimization levels, various
targets are very different and so while some var might be still available
in some place on one target somewhere, on another target there is no way
to compute its value without pessimizing generated code etc. So, it is
always preferrable to compate test_summary output with earlier output from
the same architecture/gdb version.
> > Anyway, IMHO the desired outcome is that the
> > parameters will have DW_OP_fbreg {0,4} as their location, so perhaps the
> > DRAP register needs to be remapped somehow during adjust_insn.
>
> So you don't care about their location during the prologue?
I do, but DW_OP_fbreg should be precise all the time, on e.g. i386 it is
the address of the first parameter stack slot, that is a constant address
through the whole function. And, .eh_frame is responsible for making sure
the CFA is correct at every instruction.
Jakub
