Disable g++/libstdc++-v3 on AIX for 3.4?

[Roger Sayle]

On Sat, 17 Apr 2004, David Edelsohn wrote:
> 	I have been testing with the pre-release version of the AIX
> assembler and archiver, with very good testsuite results.  I have been
> mailing those results to the gcc-testresults mailinglist.
>
> 	I am having difficulty understanding what concern you are trying
> to raise.  The PTF will be available soon.  When it is generally
> available, GCC 3.4 will have full functionality on AIX.  There have been
> many AIX assembler PTFs that have improved GCC functionality on AIX.

I'm just a little concerned that although powerpc-ibm-aix5.1 is listed
as a primary evaluation platform from 3.4, the only people that will be
able to build g++ on that platform for some time after its release will
be IBM employees.  If the fix were available now, or prior to the 3.4.0
release, then I agree there'd be less of an issue.  However on almost
all other GCC supported platforms: Solaris, IRIX, HPUX, Tru64, VxWorks,
Linux, Darwin, etc... the inability to build the release candidate by
the general public would normally indicate a broken port.  Your position
within IBM gives you a slightly different perspective from the rest of
the GCC community.  Your postings to gcc-testresults aren't completely
representative of that target's health.


You must also admit that your comment "The PTF will be available soon"
is a forward looking statement, and involves factors outside your
control.  Is there a scheduled release date?  Might it be after 3.4.1
comes out?  This problem report was openned with IBM back in February.
Its currently not even listed as a HIPER (a high priority APAR).


Why isn't a configure test for the broken AIX assembler appropriate?
Even once the fix/patch is out, not everyone will have installed it.
The current failure mode of a "make bootstrap" hanging in an infinite
loop is worse than an error message.  Its not immediately clear
(without -v) that the loop is in the assembler and not GCC's fault.
It might be possible to disable just the 64bit multilib of libstdc++-v3.


Do my concerns really seem that unreasonable?  Admittedly this is a
rare (unique?) corner case, so I'm unsure whether it deserves any
discussion or not.  If the release manager is content with the status
quo, then so am I.

Roger
--