Doxygen 1.8.13 segfaults on dolfin source

[Tormod Landet]

I have tested several doxygen versions, including the three latest

1.8.12 - works 1.8.13 - segfault 1.8.14-dev (current master) - works

I believe it is this (fixed) doxygen issue: https://github.com/doxygen/doxygen/pull/555

We know the doxygen version in the Python script that builds the SWIG files and in the script that builds the documentation, so we can potentially print a warning and fallback to empty docstrings if the buggy version is installed

[Jan Blechta]

Sure, warning for buggy version would be good. But generally error handling could be bettter. If doxygen fails maybe cmake should stop and say what happened.

[Tormod Landet]

Fix in master

The docstring generation will now check if docstrings were actually generated and produce dummy docstrings if doxygen crashed or something else went wrong somehow. This should make the whole thing more robust at the expense of some people not seeing the warning during configure and ending up with dummy docstrings

[Jan Blechta]

Isn't it feasible to stop on the error? The errors should be explicit.

[Tormod Landet]

I guess it should be possible somehow in CMake, but is it really an error or just a warning? It is only the correct docstrings that are missing, the code runs fine with dummy docstrings.

[Jan Blechta]

Yes, but it hides a problem from a user when she could be expecting that it would work as doxygen was installed.

[Tormod Landet]

Now it gives the reason for the dummy docstrings in the actual docstrings. Should be more user friendly.

The conflict between discoverable error messages and not making doxygen a requirement for building will always be there. If doxygen is not a requirement we should not fail due to a buggy version being installed

[Jan Blechta]

Ok. Thanks for the improvements anyway.