Fix indexing bug in _match_psf_stars

Improve test to catch the bug.
Rename variables in the method to make the components more obvious.

python/lsst/pipe/tasks/calibrateImage.py

@@ -717,13 +717,13 @@ def _match_psf_stars(self, psf_stars, stars):
         # Closest matches is a dict of psf_stars source ID to Match record
         # (psf_stars source, sourceCat source, distance in pixels).
         best = {}
-        for match0, match1, d in matches:
-            id0 = match0.getId()
-            match = best.get(id0)
+        for match_psf, match_stars, d in matches:
+            match = best.get(match_psf.getId())
             if match is None or d <= match[2]:
-                best[id0] = (match0, match1, d)
+                best[match_psf.getId()] = (match_psf, match_stars, d)
         matches = list(best.values())
-        ids = np.array([(match0.getId(), match1.getId()) for match0, match1, d in matches]).T
+        # We'll use this to construct index arrays into each catalog.
+        ids = np.array([(match_psf.getId(), match_stars.getId()) for match_psf, match_stars, d in matches]).T
 
         # Check that no stars sources are listed twice; we already know
         # that each match has a unique psf_stars id, due to using as the key
@@ -735,11 +735,11 @@ def _match_psf_stars(self, psf_stars, stars):
                              n_matches, n_unique)
 
         # The indices of the IDs, so we can update the flag fields as arrays.
-        idx0 = np.searchsorted(psf_stars["id"], ids[0])
-        idx1 = np.searchsorted(stars["id"], ids[1])
+        idx_psf_stars = np.searchsorted(psf_stars["id"], ids[0])
+        idx_stars = np.searchsorted(stars["id"], ids[1])
         for field in self.psf_fields:
             result = np.zeros(len(stars), dtype=bool)
-            result[idx0] = psf_stars[field][idx1]
+            result[idx_stars] = psf_stars[field][idx_psf_stars]
             stars[field] = result
 
     def _fit_astrometry(self, exposure, stars):

tests/test_calibrateImage.py

@@ -332,10 +332,19 @@ def test_match_psf_stars(self):
         self.assertEqual(stars["calib_psf_used"].sum(), 0)
         self.assertEqual(stars["calib_psf_reserved"].sum(), 0)
 
+        # Reorder stars to be out of order with psf_stars (putting the sky
+        # sources in front); this tests that I get the indexing right.
+        stars.sort(stars.getCentroidSlot().getMeasKey().getX())
+        stars = stars.copy(deep=True)
+        # Re-number the ids: the matcher requires sorted ids: this is always
+        # true in the code itself, but we've permuted them by sorting on
+        # flux. We don't care what the actual ids themselves are here.
+        stars["id"] = np.arange(len(stars))
+
         calibrate._match_psf_stars(psf_stars, stars)
 
-        # Sort in order of brightness; the psf stars are the 3 brightest, with
-        # two sky sources as the faintest.
+        # Check that the three brightest stars have the psf flags transfered
+        # from the psf_stars catalog by sorting in order of brightness.
         stars.sort(stars.getPsfFluxSlot().getMeasKey())
         # sort() above leaves the catalog non-contiguous.
         stars = stars.copy(deep=True)