Albert Chin wrote:
Why add -qHALT=E to CFLAGS on AIX (build/apr_hints.m4)? That seems
like an undue burden.
         It forces the client application to work with
-qHALT=E. I don't see an equivalent restriction for Solaris
(-errwarn), HP-UX (+We), Tru64 UNIX (-msg_error), etc. Subversion, for
example, breaks with this.

    *-ibm-aix*)
        ...
        dnl If using xlc, remember it, and give it the right options.
        if $CC 2>&1 | grep 'xlc' > /dev/null; then
          APR_SETIFNULL(AIX_XLC, [yes])
>           APR_ADDTO(CFLAGS, [-qHALT=E])

For some reason I suspect that the Subversion folks would be pleased to know that, and eager to resolve the warning/error/whatever. But I understand how somebody taking some standard package may not be so keen on resolving language nuances which could possibly be ignored.

I guess your experience differs, but in my experience, the same sort of thing that won't compile with xlc will often have compile failures with vendor compiler for HP-UX, Tru64, or z/OS as well (all using APR-specified options). (I had cron jobs reporting build issues on various platforms for a lot of the busier stage of APR development; somebody would check in something not quite right and AIX, HP-UX, and Tru64 would very often report the same problem. I don't recall any onerous failures specific to AIX.)

Removing this could possibly affect autoconfiguration. See the original change log:
--
Clean up the detection of various versions of AIX. This has one glaring
annoyance; xlC must be detected early on, and passed a certain flag to
quit on needed error conditions.
Submitted by: Victor J. Orlikowski <[EMAIL PROTECTED]>
Reviewed by: Ryan Bloom
--
No idea here what detection failed.


Try turning it off and see if anything autodetected changes. If no change, then it is worth trying to see exactly what class of problems are fatal with HALT=E with the goal of making it maintainer mode only.

Reply via email to