GitLab

snglcoinc.py:  another bug fix

See merge request lscsoft/lalsuite!1668

- either the bug fix for the missing triples in the string search, or the bug fix to fix the new bugs that bug fix introduced, introduced another bug that wasn't detected until now:  the union of the "unused singles" sets returned from the doubles generator contained the complete set of unused singles, but the specific set returned on each call didn't correspond to the correct time range and so a few singles were being discarded.  this patch fixes.
- this patch fixes it, and, happily, gives a small performance boost

- feature no longer makes any sense, and no calling code uses

- these tools are now only used for diagnostic purposes.  this patch implements an emulation of the style in which the coincidence engine is used online, which also provides a more thorough test of the code's behaviour, but it is a little slower.

- this library routine is no longer used by the gstlal pipeline for coincidence, so there is no longer a need for on-the-fly ranking stat and FAP/FAR assignment.  some performance improvement can be realized by removing support for those operations (eliminates conditionals from inner loops)

- instead of sorting the result in the calling code.  this change is mostly cosmetic, but this makes the behaviour consistent across different graph node types.

- tiny, *tiny*, improvement in performance, not significant.  main advantage is to just reduce the volume of code to maintain

snglcoinc.py:  fix new bugs in coincidence code

See merge request lscsoft/lalsuite!1667

- replace the "union of all pair-wise intersections" step in the partial coincs construction with a "count occurrences" approach.  breaks even, performance wise, for small networks, but will scale better to 4 and more networks.

- reduce the number of symbols created by the inner loop to reduce Python object create/delete operations

- testing shows that the two-queue approach is slower than a single queue even though that costs redundant sorts.

- after benchmarking tests, the new singles flushing implementation has
  been found to be more efficient than the previous one, so this patch
  makes the conversion permanent by removing the unused code from the old
  implementation

- recent changes to fix bugs in the coincidence engine introduced new ones.  this fixes them.
- the first bug is that subsets of the triggers in a coincident candidate can be reported a second time on occasion, because after their partner is flushed they appear to be a new candidate with a new reference time.
- the second bug is that single triggers are not flushed from the queues for different time slides at the same time, leading to them being recorded more than once in the trigger list

thinca:  allow negative event IDs

See merge request lscsoft/lalsuite!1666

- modify the error handling for get_template_id() so that negative template
  IDs are not interpreted as error codes

Merge release branch

See merge request lscsoft/lalsuite!1665

this is a temporary workaround until the proper fix can be implemented