Investigate: Check we are using memory responsibly in SANS reduction

[Nick Draper]

Richard Heenan: Combine “convert to wavelength” and “rebin” in the main detector data reduction to avoid the memory requirements of the huge intermediate ragged workspace. This ought to speed up data reduction and allow more workspace storage before users run out of memory.

Nick Draper: Check we are using the same input and output workspaces for convertunits and rebin so we work in place or overwrite. If we are storing the workspace for any length of time we nee to know why and present the implications.

This is an investigative task in the first instance.

[Steve Williams]

(In [8601]) The SANS errors workspace sometimes wasn't getting deleted. Deal with the situation where the full wave range transmission workspace has a different rebinning from the sample re #2153

[Steve Williams]

(In [8620]) Temporally reverse some changes until the Windows installers are back in action re #2153

[Steve Williams]

(In [8638]) Deal with the situation where the full wave range transmission workspace has a different rebinning from the sample re #2153

[Steve Williams]

Throughout the SANS analysis the input and output workspaces from ConvertUnits() are the same. This causes Mantid to write the data in place which is the most memory efficient that Mantid can currently be. In all those instances Rebin() or InterpolatingRebin() is called with the same in/output workspaces reducing memory usage and this step can't be done more efficiently.

[Martyn Gigg]

In 1.1.9493, the output seems reasonable throughout the analysis given the required data that users wish us to leave at the end. The changes around the errors workspace seem sensible and they haven't affected the results.

[Stuart Campbell]

This ticket has been transferred to github issue ​3000