Sprint review 8.1.25: Alex presentation of development design

[cmarooney-stfc]

FORMULATION IN PROGRESS - NOT COMPLETE

Alex presented how his current developments are integrated. The tickets under consideration are

• Serializable equality method (eq) should be replaced by overloaded equal_to_tol #1147
• hashable class  #1788 enabling
  • make instruments, samples and detectors hashable #1790
  • odd behaviour of multifit and covariance matrix #1798 arising from tests of above. Noting that the issue might be a change in use of RNGs
  • Problems with unique references singleton implementation #1791 arising from problems with these tests
  • (no ticket) garbage collection of removed objects in unique_references_container
• (no ticket) CoW pointer issue

Discussion details:

Problems identified with hashing

• performance (NB some fixes related to this have been identified fix fetch of glc #1801)
• need for hashes in class (to be identified below, associated class merged hashable #1789)

From #1789, hashing now only deals with the contents of serializable, and the hash object is now given a tolerance as in equal_to_tol. This practice was approved in the merging of #1789 but was not liked by the other participants in this meeting.

ACTION: Chris to review the approval process for #1789 and create a ticket to review if this practice should continue.

It was also noted that run_id is not hashable in this ticket as not contributing to the uniqueness of the object. Again this was thought suprising and in need of review.

ACTION: Alex to find existing ticket number on run_id, and if not available, make a new one.

The equal_to_tol issue #1147 appeared at points in this discussion, but at the close it seemed to be agreed that its substance was not relevant to the other items discussed. It should however be clarified. Hence

ACTION: Alex to make a ticket to write devdocs for #1147.

ACTION: Alex to make a separate ticket to replace eq_to_tol type names with equal_to_tol conforming with the existing usage.

Toby and Alex agreed to discuss the multifit/covariance issue #1798 offline.

Time prevented moving on to discuss #1791 and Alex has subsequently been asked to arrange a follow on meeting to deal with that.