Crash with quick script that changes log values or instrument parameters

[Martyn Gigg]

Demonstrated by the script below.

import dgreduce

iliad_setup=dgreduce.setup
iliad=dgreduce.arb_units
iliad_abs=dgreduce.abs_units
iliad_help=dgreduce.help

inst='mar'
iliad_setup(inst)
ext='.raw'
mapfile='mari_res'
cal_file='MAR11060.RAW'
#load vanadium file


for i in range(200):
        print '############' ,i,' ##################'
        run='11001'
        WB='11060'
        ei = 12.5
        rebin_params = "-6.25,0.03125,11.875"
        BkgdRange = [11705.726218019521, 19500]

        w1=iliad(WB,run,ei,rebin_params,mapfile,det_cal_file=cal_file,norm_method='current')
        #w1test=iliad(WB,run,ei,rebin_params,mapfile,det_cal_file=cal_file,norm_method='current',save_format='',background=True,bkgd_range=BkgdRange)
        DeleteWorkspace('w1')
        #DeleteWorkspace('w1test')

that never reaches the end successfully.

In release it either shows the terminate dialog saying about m_run being empty or something about a null pointer in the ParameterMap

[Alex Buts]

Yes it is the problem I am experiencing with the Reductuion GUI

[Martyn Gigg]

Lock the DataItem RWLock while generating the ADS tree for the dock.

The ADS is now generating a node tree that describes the state of the ADS when queried. This in turn queries the workspace itself and while running in MantidPlot an executing script can cause things wrapped in a cow_ptr to be copied and at the same time access the pointer while it is being copied, causing a crash.

Refs #7583

Changeset: e6de55b899a74120c7da75773190024f12cb0fbc

[Martyn Gigg]

This is not ideal. The dock used to lazy-load the information about a workspace which in-practice avoided this problem because 99% of the time the information was never generated. At this stage in the release it's not possible to put back the lazy loading but the crash needsto be avoided.

[Martyn Gigg]

Branch: bugfix/7583_crash_on_dock_updating

Tester: The script above should reproduce the error. It will require the systemtests/Data directory to be in the Mantid search path and to be run from MantidPlot.

I would advise running the script BEFORE starting to test to ensure that the segfault actually occurs and then merge to the code and test the script again.

[Russell Taylor]

OK, it made it through the script 3 times without crashing with that change. I think you could have achieved the same thing (and saved me compilation time!) with a one line change in Workspace.cpp: Kernel::ReadLock _lock(*this);

We certainly need to get back to laziness in this area.

[Stuart Campbell]

This ticket has been transferred to github issue ​8428