Some negative TARGETIDs have OBJTYPE == "" in fiberassign* output files.

[dylanagreen]

Noticed originally by @abrodze, the output of fba_launch generates fiberassign output files from the raw fba files using

fiberassign/py/fiberassign/assign.py

Line 1474 in d93c2e3

def merge_results(targetfiles, skyfiles, tiles, result_dir=".",

which calls the per tile version

fiberassign/py/fiberassign/assign.py

Line 1102 in d93c2e3

def merge_results_tile(out_dtype, copy_fba, params):

The merging code only sets values for "fiberassign valid" TARGETIDs, defined as positive TARGETIDS:

fiberassign/py/fiberassign/assign.py

Lines 1265 to 1266 in d93c2e3

	# Rows containing assigned fibers
	fassign_valid = np.where(fiber_data["TARGETID"] >= 0)[0]

Then it only sets the OBJTYPE for these TARGETIDs:

fiberassign/py/fiberassign/assign.py

Lines 1325 to 1342 in d93c2e3

	objtype = np.zeros(len(fassign_valid), dtype="S3")
	# AR relaxing the condition on the presence of any
	# AR *_TARGET, excluding FA_TARGET
	# AR hence accepting CMX_.., SV1_..., SV2_...
	#if "DESI_TARGET" in out_dtype.names: # AR commented out
	if "_TARGET" in [name[-7:] for name in out_dtype.names if name!="FA_TARGET"]:
	# This is a main survey file.
	objtype[:] = "TGT"
	is_sky = (tile_targets["DESI_TARGET"][target_rows]
	& desi_mask.SKY) != 0
	is_sky |= (tile_targets["DESI_TARGET"][target_rows]
	& desi_mask.SUPP_SKY) != 0
	objtype[is_sky] = "SKY"
	badmask = \
	desi_mask.mask("BAD_SKY|NO_TARGET|IN_BRIGHT_OBJECT")
	is_bad = (tile_targets["DESI_TARGET"][target_rows]
	& badmask) != 0
	objtype[is_bad] = "BAD"

Which means that negative TARGETIDs have the default fill value (an empty string) for OBJTYPE.

Nominally I think the negative TARGETIDs are supposed to be handled by this line:

fiberassign/py/fiberassign/assign.py

Lines 1297 to 1302 in d93c2e3

	# Special handling for STUCK positioners that land on good SKY positions.
	# In the FA files these get FA_TYPE & 4 (SKY) and a negative TARGETID.
	stucksky_rows = np.where((fiber_data['TARGETID'] < 0) *
	((fiber_data['FA_TYPE'] & TARGET_TYPE_SKY) != 0))[0]
	if len(stucksky_rows):
	outdata['OBJTYPE'][stucksky_rows] = 'SKY'

but evidently something in the boolean condition here is not flagging for the entire set of negative targetids (is fiberassign not propogating the TARGET_TYPE_SKY bit for some of these targets?). That is, only a subset of the negative targetids are captured by these two lines.

Picking a pseudorandom fiberassign file (fiberassign-101689.fits.gz) and reading the "FIBERASSIGN" header:
643 TARGETIDs are negative
400 of those are OBJTYPE == "SKY" set correctly
243 of these have no OBJTYPE == "".

This seems like an oversight but I don't know what the correct value should be set for these targets, but I certainly don't think it should be an empty string.

It's also not clear if the issue is actually in assign.py or if something upstream isn't propagating a bit correctly which is then mangling the results in this file.

[dylanagreen]

Stephen and I discussed this and after a spot check confirming that all blank string OBJTYPE values specifically fall only on negative TARGETIDs, we've decided the best course of action to avoid touching fiberassign is to intercept these blank strings in desispec assemble_fibermap and correct those blank strings to BAD at that step. This avoids changing the fiberassign output ~4.5 years into the survey, and regardless of whether it's updated in fiberassign it would need to be updated in desispec anyway to handle the previously generated fiberassign files..

[sbailey]

Another detail from the chat: the negative TARGETIDs with blank OBJTYPE are stuck positioners that are not pointing at blank sky locations. If stuck positioners are serendipitously pointing at blank sky, they are upgraded to OBJTYPE='SKY'.

[araichoor]

last message from Stephen is correct.
additional remark: the 21 broken positioners will have OBJTYPE="".

without checking files/codes, from my memory:

• unassigned fibers will have a negative TARGETID, with OBJTYPE="";
• if it is a stuck positioner falling on a blank usable sky position, it will be promoted to OBJTYPE="SKY".

I think that unassigned fibers only are stuck or broken fibers (as the default sky targets are a dense grid on the sky, a valid fiber should always have at least a sky target to be assigned to; for regions like bright star neighborhood, I think there will be targets with OBJTYPE="BAD", so some assignment, but which is not use for science).
see e.g. tileid=9489 with the Mira star.