desihub / desispec / 31842737677

37%

main: 39%

Build Type

Pull #2797

github

Committed by sbailey

Commit Message

Remove redundant sorts from find_target_priority

find_target_priority previously sorted the full TARGETID column up to
three times: the necessary priority sort, a np.unique() call that
re-sorted data already sorted by the previous step, and an
ordered_unique() call that independently re-sorted the original array
just to recover first-appearance order. Since tsel is already sorted
by TARGETID after the priority sort, group boundaries (and thus the
unique TARGETIDs and their winning rows) can be found with a single
O(n) scan instead.

This changes find_target_priority's output order from first-appearance
to ascending TARGETID order. create_summary_catalog and
get_exp2uniqpix_map don't care about this order (both already
effectively expand/join back onto the full TARGETID column by key), so
they simplify along with it. coadd_fibermap does depend on
first-appearance order, matching the separate ordered_unique() call
that coadd() uses to build the flux/ivar/mask/resolution-data arrays
it must stay aligned with, so it now explicitly restores that order
via ordered_unique() + searchsorted rather than relying on
find_target_priority to provide it.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Pull Request Pull Request #2797: Handle discrepant RA,DEC,etc per TARGETID

Coverage Stats

56 of 58 new or added lines in 3 files covered. (96.55%)

687 existing lines in 3 files now uncovered.

13259 of 35490 relevant lines covered (37.36%)

0.37 hits per line