gas: elf/bad-group testcase

gas: elf/bad-group testcase

[Jan Beulich]

Hello,

I'm trying to understand what this testcase is actually testing.
Originally it was looking for just _some_ error. In commit
85024cd8bcb9 it was changed to expect an error on line 10. But
line 10 is a .uleb128, where the difference of symbols between
two sections cannot be resolved. How's that group related? Was it
perhaps some other error which was originally meant to be raised,
but which has meanwhile disappeared (on x86 at least the one
error mentioned above is the only one).

For context: A pending change of mine broke this testcase for two
targets, and understanding the goal of the testcase would
certainly help me figure out the breakage.

Thanks, Jan

Re: gas: elf/bad-group testcase

[Alan Modra]

On Thu, May 05, 2022 at 10:48:47AM +0200, Jan Beulich wrote:
> Hello,
> 
> I'm trying to understand what this testcase is actually testing.

https://sourceware.org/pipermail/binutils/2011-August/073809.html

-- 
Alan Modra
Australia Development Lab, IBM

Re: gas: elf/bad-group testcase

[Jan Beulich]

On 05.05.2022 12:48, Alan Modra wrote:
> On Thu, May 05, 2022 at 10:48:47AM +0200, Jan Beulich wrote:
>> Hello,
>>
>> I'm trying to understand what this testcase is actually testing.
> 
> https://sourceware.org/pipermail/binutils/2011-August/073809.html

Thanks. I read this to mean it's entirely uninteresting what kind of
error there was. And I've meanwhile figured that the specific error
here, which presumably most targets get to see (x86 does), disappeared
for targets setting TC_FINALIZE_SYMS_BEFORE_SIZE_SEG to zero. This is
because that error would only be seen if S_GET_VALUE() in
cvt_frag_to_fill()'s processing of rs_leb128 runs with finalize_syms
already set to 1; the same expression wouldn't be evaluated another
time later anymore (and hence the error has no chance of surfacing).
I have to admit that I find this quite strange. But I could also see
this being an interaction between leb128 processing (which has some
further anomalies anyway) and TC_FINALIZE_SYMS_BEFORE_SIZE_SEG == 0
which so far simply has been overlooked.

Jan

Re: gas: elf/bad-group testcase

[Alan Modra]

On Thu, May 05, 2022 at 01:07:48PM +0200, Jan Beulich wrote:
> On 05.05.2022 12:48, Alan Modra wrote:
> > On Thu, May 05, 2022 at 10:48:47AM +0200, Jan Beulich wrote:
> >> Hello,
> >>
> >> I'm trying to understand what this testcase is actually testing.
> > 
> > https://sourceware.org/pipermail/binutils/2011-August/073809.html
> 
> Thanks. I read this to mean it's entirely uninteresting what kind of
> error there was.

Yes.  I guess all we are really interested in is not getting a gas
segfault.

-- 
Alan Modra
Australia Development Lab, IBM