desihub / fiberassign / 33925332132

66%

main: 66%

Pull #523

github

Fix merge_results() reliance on mutated global state across worker processes

merge_results() mutated the module-level merged_targets_columns dict to
record *_TARGET columns discovered from the input catalogs, then relied on
worker processes (spawned via multiprocessing.Pool) reading that same
mutated global inside merge_results_tile(). This only worked because
Python <3.14 defaulted to the "fork" start method on Linux, where workers
inherit the parent's already-mutated memory. Python 3.14 changed the
default to "forkserver", where workers no longer see mutations made to the
parent's globals after the module was first imported, so the TARGETS HDU
of merged output files silently dropped columns like DESI_TARGET, and a
later read of those files (test_assign.py::TestAssign::test_io) crashed
with "ValueError: no field of name DESI_TARGET".

Compute the columns locally and pass them explicitly to merge_results_tile
via functools.partial so they are correctly propagated to worker processes
regardless of the multiprocessing start method.

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

4 of 5 new or added lines in 1 file covered. (80.0%)

3978 of 6047 relevant lines covered (65.78%)

3167836.24 hits per line