Re: [GNU Autoconf 2.67] testsuite: 193 199 233 253 254 failed on Haiku

[Ralf Wildenhues]

Hi Scott,

thanks for the test feedback.

* scott mc wrote on Tue, Aug 03, 2010 at 10:22:55PM CEST:
> [Moderator Note: The original message was very large at 203803 bytes.
> I gzip'd the log files down to 16K and 18K, re-attached the files and
> re-mailed the message. --Bob Proulx]

thanks Bob!

> Haiku currently uses autoconf-2.65, I have attached the test results
> from running make check with 2.65 and 2.67.  With the regression in
> 2.66 we rolled back to 2.65.  iku.
> This was run on Haiku r37150 which was built on June 15th, 2010.

These results do not actually point at a regression in Autoconf; here's
why:

|  193: autotest.at:1250   parallel test execution
|       autotest

This looks like the stdout output from the shell redirected to a file in
append mode: >>"$at_stdout" did not actually catch all output.  To me,
that seems like a race condition hidden in file handling on Haiku
somewhere.  Or in the shell.

If you run this test 20 times, how often do you get failures?
  for run in `seq 10`; do
    make check TESTSUITEFLAGS=193
  done

|  199: autotest.at:1391   parallel autotest and signal handling
|       autotest

Also seen in 2.65, and actually fails at the same place.  Was this
system very loaded when testing?  Are there known issues in job/process
group handling in Haiku, or in interaction of jobs with signals?

|  233: torture.at:854     Substitute and define special characters
|       ac_define ac_define_unquoted

Also in 2.65.  I've seen this elsewhere, need to look up whether sh,
sed, or awk is likely to blame.

|  253: compile.at:170     AC_LANG_SOURCE example
|  254: compile.at:216     AC_LANG_PROGRAM example

These are actual bugs in the Autoconf testsuite (but not in the rest of
Autoconf).  Will fix.

Cheers,
Ralf