2 x86-64 ABI bugs in gcc 3.3 and 3.4

Gabriel Dos Reis gdr@integrable-solutions.net
Tue May 18 13:19:00 GMT 2004

Gerald Pfeifer <gerald@pfeifer.com> writes:

| On Tue, 18 May 2004, Gabriel Dos Reis wrote:
| >| There are 2 x86-64 ABI bugs:
| >|
| >| http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15301
| >| http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15302
| >|
| >| There is one testcase patch in each bug report. Mainline is fixed. I'd
| >| like to see gcc 3.3 and 3.4 get fixed too. I understand they are
| >| ABI changes. But my testcases show gcc 3.3 and 3.4 fail to correctly
| >| compile ABI conforming programs.
| > OK.
| Gaby, I am having a problem with this approval for two reasons:
| 1) We must not change the ABI on an active release branch, especially not
| as late on a release branch, which is supposed to be in deep maintenance
| mode.

I fully considered that this was a release branch and the ABI should
not change there.  And I was going to say "no", mechanically whether
it was approved for 3.4.x or not.  Then, I *looked* at the actual PRs
and after consideeration, it became clear to me that they were cases
where we got the actual ABI wrong.  Then one way would be to leave the
bug there, but then it then became obvious that GCC-3.3.4 could not be
used for the affected plateform. IT is only after more consideration
that I said OK.  It was not a mechanical "yes", it was more close to a
mechanical "no" for conservative reasons.

| 2) Approving such a change for 3.3 without checking the situation on 3.4
| means that we may have ABI breakage from 3.3 to 3.4, and yet again (just
| in the other direction) from 3.4 to 3.5.

Looking at the bug, it meant that 3.4.x could not be used safely on
those plateforms, I was almost certain the reasons that led me to say
yes would lead Mark to approve the patch for 3.4.1 too.  But, yes, it
could have been cheched with Mark.  Did you look at the PRs?

-- Gaby

More information about the Gcc mailing list