running with different n_events gives segfault

[sschmidt]

Running e.g. test/qedtest.sin with n_events = 10 and afterwards increasing n_events and running it again produces a segfault when reading the events file:

Reading events from file 'whizard.evx' ...
Segmentation fault

[Juergen Reuter]

The problem appears only if the user changes the number of events from n1 /= n2, but does not rerun with the "-r" option. If the user correctly uses that option then everything is fine. But, there should be a MD5 checksum mismatch for the first case, or something similar.

[Juergen Reuter]

The error appears only when increasing, not when decreasing the number of events.

[Juergen Reuter]

Here is the gdb backtrace for the case of increasing from 15 to 16 events:

#0  data_transfer_init (dtp=0xbf1c20ec, read_flag=1) at ../../../libgfortran/io/transfer.c:2070
#1  0xb7d8c2b5 in __flavors_MOD_flavor_read_raw () at flavors.f90:222
#2  0xb7d9040c in __quantum_numbers_MOD_quantum_numbers_read_raw () at quantum_numbers.f90:380
#3  0xb7d9555d in __state_matrices_MOD_node_read_raw_rec () at state_matrices.f90:303
#4  0xb7d95707 in __state_matrices_MOD_node_read_raw_rec () at state_matrices.f90:311
#5  0xb7d95707 in __state_matrices_MOD_node_read_raw_rec () at state_matrices.f90:311
.... (going up to #57)
....

[kilian]

I can't reproduce this, neither with nagfor, neither with gfortran. Could you double-check and/or be more specific on the input?

(Maybe it's gone with a recent update?)

[sschmidt]

I still have the error with both nagfor and gfortran and r1634

I assembled the three steps into one sin file (see attached), the output is

n_events =  10
| No analysis setup has been provided.
| Reading events from file 'whizard.evx' ...
| ... 10 events read.
n_events =  100
| No analysis setup has been provided.
| Reading events from file 'whizard.evx' ...
Segmentation fault

[sschmidt]

single input file

[sschmidt]

just an idea: Could it be possible that the cause for this behaviour is the same as in #152 ?

Might be because #152 also doesn't appear on all machines. Could anybody check if these bugs appear always together?

[Christian Speckner]

Fixed in r1641. WHIZARD checked iostat only after trying to read in the event including the density matrices, resulting ultimately in read being called on invalid memory (a null pointer on my system). WK might want to check if he likes the place I put the check :)

[Juergen Reuter]

Milestone v2.0-rc2 deleted