[Bug translator/19500] .callee test failures

[dsmith at redhat dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=19500

--- Comment #1 from David Smith <dsmith at redhat dot com> ---
This appears to be the result of s390x's odd debuginfo. On x86_64, you'll see
something like this (from 'eu-readelf --debug-dump libreloc.so'):

====
...
 [    2d]    subprogram
             external             (flag_present) Yes
             name                 (string) "bar"
...
 [    79]    subprogram
             external             (flag_present) Yes
             name                 (string) "foo"
...
             GNU_all_call_sites   (flag_present) Yes
...
 [    bb]      GNU_call_site
               low_pc               (addr) +0x00000000000006e2 <foo+0x12>
               abstract_origin      (ref4) [    2d]
...
====

This means that foo() has one callee, bar(), which we can see since foo() has 1
GNU_call_site attribute, whose abstract_origin is 2d (which maps to bar()).

On s390x, we see the following (from 'eu-readelf --debug-dump libreloc.so'):

====
...
 [    2d]    subprogram
             external             (flag_present) Yes
             name                 (string) "bar"
...
 [    7b]    subprogram
             external             (flag_present) Yes
             name                 (string) "foo"
...
             GNU_all_call_sites   (flag_present) Yes
 [    bd]      GNU_call_site
               low_pc               (addr) +0x000000000000073a <foo+0x22>
...
====

On s390x, we've got a GNU_call_site attribute, but it doesn't have an
abstract_origin attribute to tell us which function was called.

I've sent an email to a gcc developer and we'll see if he can shine any light
on this.

-- 
You are receiving this mail because:
You are the assignee for the bug.