Regression in error handler since version 5.0.1

[prj-]

When switching from version 5.0.0 to version 5.0.1., we noticed a regression in the error handler which outputs on the terminal some additional (unwanted) characters. See the discussion here: https://gitlab.com/petsc/petsc/-/merge_requests/7143#note_1709258522.

[prj-]

@jsquyres, here is the bisection log.

git bisect start
# bad: [79f675e98ca842c7feb7615b292a202e3c0646ee] Merge pull request #12177 from wenduwan/5.0.1_release
git bisect bad 79f675e98ca842c7feb7615b292a202e3c0646ee
# good: [d0fe8ef8113906ac485c8d4a2e687d6bdcaa8305] update LICENCE file
git bisect good d0fe8ef8113906ac485c8d4a2e687d6bdcaa8305
# good: [d55bb67bd39423f0e4940915b1e906da2279e225] Merge pull request #12069 from jsquyres/pr/v5.0.x/still-moar-docs-updates
git bisect good d55bb67bd39423f0e4940915b1e906da2279e225
# good: [b34742d0131e4ed21c536e147f4efa65ced6deb4] osc_rdma: do pointer match on (const char*)
git bisect good b34742d0131e4ed21c536e147f4efa65ced6deb4
# good: [dcbea8518f78595eded77d8b29eda067e83feec7] Merge pull request #12139 from wenduwan/backport_ompi_info
git bisect good dcbea8518f78595eded77d8b29eda067e83feec7
# bad: [ca56ceb9365bd1f4beabcbf415e7e63b6dee365e] osc/rdma: fix compiler warnings in osc_rdma_accumulate.c
git bisect bad ca56ceb9365bd1f4beabcbf415e7e63b6dee365e
# bad: [15bfea8adefdb4deb3d1f328c9807b128720e27b] Merge pull request #12134 from jsquyres/pr/v5.0.x/update-prrte-and-corresponding-docs
git bisect bad 15bfea8adefdb4deb3d1f328c9807b128720e27b
# bad: [f53e5908c26cc876fc4a8e0e719c960505c15ccd] Update prrte submodule pointer to head of v3.0 branch
git bisect bad f53e5908c26cc876fc4a8e0e719c960505c15ccd
# good: [20e943685f40337c08a005dae289981338bdb5e8] docs: update to match PRRTE v3.0.3 docs
git bisect good 20e943685f40337c08a005dae289981338bdb5e8
# first bad commit: [f53e5908c26cc876fc4a8e0e719c960505c15ccd] Update prrte submodule pointer to head of v3.0 branch

[jsquyres]

We talked about this on the RM call today -- we probably need an easy way to replicate this.

@prj- can you provide a simple recipe for replication?

[prj-]

There is one there: https://gitlab.com/petsc/petsc/-/merge_requests/7143#note_1709308747. Further below the thread, there is the script you'll need to replay the bisection. To just get the example running, the following should be enough.

$ git clone https://gitlab.com/petsc/petsc
$ cd petsc
$ export HASH=f53e5908c26cc876fc4a8e0e719c960505c15ccd PETSC_DIR=`pwd` PETSC_ARCH=arch-debug-openmpi
$ ./configure --with-debugging=1 --download-openmpi=git://https://github.com/open-mpi/ompi --download-openmpi-commit=`echo $HASH` --with-fortran-bindings-inplace
$ make all PETSC_ARCH=`echo ${PETSC_ARCH}` PETSC_DIR=`echo ${PETSC_DIR}`
$ cd src/sys/tests && make ex1f PETSC_ARCH=`echo ${PETSC_ARCH}` PETSC_DIR=`echo ${PETSC_DIR}`
$ ./ex1f >& log 
$ ${PETSC_DIR}/${PETSC_ARCH}/bin/mpiexec -n 1 ./ex1f >& log_bis
$ diff log log_bis

[rhc54]

I'm afraid using something that obtuse doesn't really help much - I need something a little more direct. Fortunately, I already have a test that just calls "abort" and can use that to reproduce the error message, which shows the extra character.

It'll take me a bit to chase it down.

[jsquyres]

@janjust volunteered to take a crack at replicating this issue.

[rhc54]

@janjust volunteered to take a crack at replicating this issue.

As noted above, I already have.

[rhc54]

Took a little bit to chase this down, but eventually figured it out. Fix is in referenced PR for PRRTE.

[jsquyres]

Nice find -- thanks @rhc54!

[janjust]

fixed with ptr update #12237

[prj-]

Our pipeline was clean with 5.0.2rc1, but is again failing with 5.0.2, sadly.

[prj-]

I guess the milestone should be changed from 5.0.2 to something else, 5.0.3 or 5.1.0?

[janjust]

@prj- yes, we expected it, we had a quick release schedule to fix a critical bug for an org so we regressed internal pointers. After today we will bump them again and your issue should be fixed.