Trunk completely broken with gfortran 4.9.2

[Juergen Reuter]

This is COMPLETE BLOCKER as this is the official gfortran version at the moment. I get completely unreadable backtraces, basically just some freed c pointers that have never been allocated. As guess is that this has to do with the refactoring of interactions. Interestingly, looking at the sf_int self test, only sf_epa and sf_ewa fail, the rest goes through. Could someone reconfirm? I used r6771.

[Juergen Reuter]

There are two more knowledge points: 1) thanks to CW: confirmation that it is the compiler, 4.9.2 segfaults also on Linux at the same places as on the MAC. 2) r6758 is ok, i.e. works. Now we have to break it down.

[Juergen Reuter]

I do think that making interaction_get_state_matrix_ptr into a TBP was the problem. Going to check this now. Maybe it's only a problem when this gets inherited to the sf_int family, but if not, we need to leave interaction_get_state_matrix_ptr as an explicit function (not too bad, though). Believe, that this issue can be settled until dawn.

[Juergen Reuter]

unfortunately, the f*ed up compiler version 4.9.2 is screwing me: it's not working. r6768 was introducing the problems with EPA and EWA leading to the failure of 14 different tests involving EPA and EWA. Commit 6769 introduced then the errors leading to the failure of another 94 tests, all of which seem to involve event simulation. My suspicion was that it was interaction_get_flv_content because that is called from term_instance_setup_flv (or similar), but this didn't solve the issue.

As I mentioned in the email, also 5.0 is affected, at least the svn version from 20 days ago. What shall we do? Completely step back from refactoring the interaction module and close #707 as wontfix?

[Juergen Reuter]

This has been revoked by reversing commit 6769-6773. More or less working again. Thanks to BACN for technical help.