Details
-
Type:
Bug
-
Status: Invalid
-
Resolution: Done
-
Fix Version/s: None
-
Component/s: spectractor
-
Labels:None
-
Team:Commissioning
-
Urgent?:No
Description
The nightly clean build failed on 2023-04-04.
FAIL: run_tests.test_extractor_ctio
|
----------------------------------------------------------------------
|
Traceback (most recent call last):
|
[...]
|
File "/j/ws/scipipe/lsst_distrib/centos-7-conda/lsstsw/build/spectractor/tests/test_extractor.py", line 64, in test_extractor_ctio
|
assert np.isclose(spectrum.x0[1] * parameters.CCD_REBIN, 683.0577836601408, atol=1 * parameters.CCD_REBIN)
|
AssertionError:
|
I'll note that the use of a simple assert works with pytest but not so well with nose (which is being used here) as the actual value is not printed.
Attachments
Activity
It appears that scikit-image 0.20.0 was responsible for this. It reverted back to scikit-image 0.19.3 on the next run, with the same results as the run previous to this failed one. The scikit-image failure was fixed on DM-38582. Closing this as a duplicate (hence Invalid).
Oh, I guess it was a nan that wasn't close then - I'm just surprised that it made it that far though, I'd have expected the actual running code to barf before then, but that's fine, as long as it's fixed, thanks!
It's not obvious to me, from looking through that output file, what went wrong here. And we've not seen this before or since either, right?
I appreciate that a) the test could be better formed to help see how far off we are, and b) it means we're probably a little susceptible to some small numerical changes here, but for now should we just chalk it up to being "one of those things" and keep an eye? Or do you want to see something done about this?
(I suspect it must have been something transient in the env solve bringing in a different package somewhere, right?)