Bug in Load SQW file-backed

[Owen Arnold]

I'm pretty sure this used to work, but loading a large sqw file now with the back-end set to on is not producing a working nxs file.

Tried to load the 29Gb iron file into mantid plot using the file-backend. This worked, but intermediate MDEW nxs file was not properly read by Paraview, said to hold zero cells. Also the file was amazingly small ~5.2Gb on disk.

Explicitly making a MDEW nexus file using SaveMD worked. And produced a ~50Gb file, which could be fully read by Paraview.

[Janik Zikovsky]

I just tried it with a 3 GB file with no problem - will try with the larger file later... What OS were you on?

SaveMD (with UpdateFileBackEnd) is supposed to be called at the end of LoadSQW anyway so it should be unnecessary...

[Owen Arnold]

This issue was found on Windows. Just to clarify:

• Loaded a 29Gb SQW dataset using LoadSQW
• Loaded MDEW looked fine, but Nexus file produced was ~5.2Gb (unuseable when loaded into PV)
• Explicity called SaveMD with MakeFileBackend resulted in 50Gb file, which was useable by PV.

[Janik Zikovsky]

OK I just tested the same file under Linux and had no trouble. A 51 GB file was created. I will need to find a way to get that file to my home Windows machine to try on Windows.

[Janik Zikovsky]

I have found the very same file to work for me under Windows 7. The .nxs file was created (eventually - it took 100 minutes but my drive is kinda slow) at 50.2 GB in size. Was able to re-load the .nxs file with LoadMD and view it in SliceViewer just fine.

[Janik Zikovsky]

It is possible that more recent changes (i.e. I fixed the virtual destructor of MDEventWorkspace so that it properly closes the file now) are responsible for the fix.

[Michael Reuter]

After waiting for 6 hours for the SQW to NXS conversion to finish and NOT trying to load the NXS file into memory, this all seems to work just fine.

[Stuart Campbell]

This ticket has been transferred to github issue ​4742