Simple decays sindarin segfaults

[Bijan Chokoufe Nejad]

With the following, we get a segmentation fault

n_events = 1
process dec_T = t => b, Wp

!process eeWW = e1, E1 => t, T     !!! At this location -> all fine
integrate (dec_T)
process eeWW = e1, E1 => t, T     !!! At this location -> SIGSEGV

unstable t (dec_T)
sqrts = 450 GeV
simulate (eeWW)

As indicated above, it works when you integrate the decay after the definition of the production process. Using whizard --debug transforms, we see that it fails in the decay event transformation:

 Current event transform:
 Event transform: trivial (hard process)
 Current event transform:
 Event transform: partonic decays

Program received signal SIGSEGV: Segmentation fault - invalid memory reference.

[Juergen Reuter]

Milestone renamed

[kilian]

The actual issue is that event generation has to call the matrix element of the decay. After the decay process has been generated and integrated, a new process is added, and the whole process library is recompiled. The old matrix element (compiled) code is lost. However, the matrix element call occurs via a procedure pointer to the old library. This has to fail, and it is outside of the control of the Fortran runtime, therefore no backtrace (Nagfor).

Possible cure: recompiling the library has to invalidate all previous references to it. I'll have to think about the level where this can be detected.

[kilian]

Fixed. Recompiling the process library invalidates all pointers to it. I inserted a counter that is incremented whenever this occurs. When a process instance is created, it will check the process whether the counter still agrees with a stored value.

In effect, the program aborts with a fatal error if an invalid library is accessed.

(NB: The bugfix branch contains a leftover bug from the fix of #792, so make check will fail at that point. To be resolved when the #792 branch has been successfully merged.)

[Juergen Reuter]

Fixed in trunk in r7853.

[Juergen Reuter]

Milestone renamed