Rescan commands segfaults due to missing model pointer

[Juergen Reuter]

Using the attached SINDARIN and LHE file, i.e. doing a rescan without an integration, leads to a segfault/runtime error with the following error message:

Runtime Error: hep_events.f90, line 117: Reference to disassociated POINTER MODEL
Program terminated by fatal error
hep_events.f90, line 117: Error occurred in HEP_EVENTS:HEPEUP_TO_EVENT
eio_lhef.f90, line 594: Called by EIO_LHEF:EIO_LHEF_INPUT_EVENT
event_streams.f90, line 271: Called by EVENT_STREAMS:EVENT_STREAM_ARRAY_INPUT_EVENT
simulations.f90, line 1927: Called by SIMULATIONS:SIMULATION_READ_EVENT_ES_ARRAY
simulations.f90, line 1775: Called by SIMULATIONS:SIMULATION_RESCAN
commands.f90, line 4668: Called by COMMANDS:CMD_RESCAN_EXECUTE
commands.f90, line 5835: Called by COMMANDS:COMMAND_LIST_EXECUTE
whizard.f90, line 348: Called by WHIZARD:WHIZARD_PROCESS_STREAM
whizard.f90, line 323: Called by WHIZARD:WHIZARD_PROCESS_FILE
main.f90, line 415: Called by MAIN

[Juergen Reuter]

SINDARIN file triggering the runtime error

[Juergen Reuter]

LHE file used for producing the error.

[Juergen Reuter]

There is more behind the lines. In principle, a user could think of just using something like:

n_events = 100
sample_format = lhef
?write_raw = false
?read_raw = false
$sample = "new_rescanned_file"
rescan "file_with_generated_events" (proc) { $rescan_input_format = "lhef" }

just to convert an event file. This leads to undefined behaviour, like variables not being set. Investigating. So defining the process(es) for which that event file has been generated is the minimum requirement for this to work. Question is what the intention is for this ticket: just to catch runtime errors and point the user to improper/unintended usage or make it possible to use WHIZARD as a converter for event files? Comments welcome.

[Bijan Chokoufe Nejad]

Well first of all Whizard should not segfault.

Secondly, why can't I use Whizard as a converter for event files? I mean maybe setup a dummy process but in principle for the supported read in and write out formats, this should work, right?

[Juergen Reuter]

WHIZARD should not segfault, true. The way it is implemented at the moment, doesn't allow what you want. One had to bypass several things. I'm not sure that this is possible with the infrastructure at the moment. At the moment, the rescan is really attached to the simulation. Maybe one would have to write a converter independently from the rest of the infrastructure. Let me try to understand the missing model pointer first.

[Juergen Reuter]

Crucial observation is that if the parameter use_process in simulation_rescan is set to true, then things work out, and the event file is (partially) successfully rescanned. However, sqrt(s) is not set properly. Main question is, shall we trigger a use_process = .true. here or not.

[Juergen Reuter]

Just doing an init-only version of the integrate_process does not do the job, as WHIZARD really needs the process pointer of all entries of simulation_t to be set. Where can this be done?

[Juergen Reuter]

Setting the process and process_instance pointer with dummy arguments still gives the following RT error:

Runtime Error: beams.f90, line 496: ALLOCATABLE BEAM_DATA%PMATRIX is not currently allocated
Program terminated by fatal error
beams.f90, line 496: Error occurred in BEAMS:BEAM_INIT
sf_base.f90, line 881: Called by SF_BASE:SF_CHAIN_INIT
instances.f90, line 1836: Called by INSTANCES:PROCESS_INSTANCE_SETUP_SF_CHAIN
instances.f90, line 1537: Called by INSTANCES:PROCESS_INSTANCE_INIT
simulations.f90, line 695: Called by SIMULATIONS:ENTRY_SET_PROCESS_PTR
simulations.f90, line 1483: Called by SIMULATIONS:SIMULATION_SET_PROCESS_PTR
commands.f90, line 4701: Called by COMMANDS:CMD_RESCAN_EXECUTE
commands.f90, line 5877: Called by COMMANDS:COMMAND_LIST_EXECUTE
whizard.f90, line 348: Called by WHIZARD:WHIZARD_PROCESS_STREAM
whizard.f90, line 323: Called by WHIZARD:WHIZARD_PROCESS_FILE
main.f90, line 415: Called by MAIN
Aborted (core dumped)

[Juergen Reuter]

Proposed solution: if rescan in the sense of just converting event files is used, then I do a dry integration run, i.e. call integrate_process with the init_only flag and put it on the global stack under its process id. Then the conversion works. Any objections? [of course, I still have to safeguard against no process being defined at all]

[Juergen Reuter]

Resolved in r7871. We actually do a dry run (init_only) call of integrate_process in order to have a valid process pointer. Some of the data like cross section, error, and sqrts are not correctly set. This might still be improved if met by a perfectionist, but the intent of avoiding the segmentation fault is fulfilled. Note that now indeed a process has to be defined,because otherwise WHIZARD will throw an error that it cannot find the process in the process library. Also sqrts has to be set for the sake of integrate_process.

[Juergen Reuter]

Milestone renamed