gold testsuite: line numbers in debug_msg test

[Marcin Kościelnicki]

I'm having problems with the gold debug_msg test on s390, specifically 
with the line numbers printed. debug_msg.sh already allows some leeway 
for them, but it seems the code emitted for s390 is more unusual than it 
expects.

The first problem is with undefined symbol references in testfn. The 
testsuite assumes the relocation for the undefined symbol will be inside 
an instruction described by .debug_line to be on the exact line of the 
undefined reference. However, this doesn't work for s390, which uses 
literal pools - the addresses (which are relocated) are stored after the 
function code, and .debug_line considers them to belong to the closing 
brace of the function.

The second problem is with SometimesInlineFunction in odr_violation2.cc 
- while the test checks for first instruction belonging to either the 
opening brace or the return statement, gcc on RHEL does one better by 
marking the first instruction (which is a multiply instruction) as 
belonging to line 29 (the closing brace).

I'm attaching a patch to make the testsuite more open-minded about the 
correct line numbers.  Does that look OK?

Marcin Kościelnicki
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-gold-testsuite-Make-debug_msg.sh-more-open-minded-ab.patch
Type: text/x-patch
Size: 6144 bytes
Desc: not available
URL: <https://sourceware.org/pipermail/binutils/attachments/20150923/080265ed/attachment.bin>