cc @mcara

Puh, these are those kind of consequences that make io.fits so "interesting" 😕

I don't have the time atm to do a proper in-depth review. However if it fixes the reported problem and the test suite doesn't complain it seems like a clear improvement.

I could also revert the change to _BaseHDU.__new__, but given that no test was failing because of this, I think it's a good point to avoid the double detection from the header, and introduce this classmethod for specific cases like this one. Hopefully not having the detection in _BaseHDU.__new__ should not have other unintended consequences 🤞

@jaytmiller , can you please confirm that works for you?

Well, that's weird. There is a html-docs error that I don't see on latest master. Maybe rebasing would help?

/home/circleci/project/docs/whatsnew/3.2.rst:98:Inline literal start-string without end-string.

I believe that failure got fixed by #8669

@saimn - can you rebase? Thanks!

Codecov Report

Merging #8666 into master will increase coverage by <.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #8666      +/-   ##
==========================================
+ Coverage   86.95%   86.95%   +<.01%     
==========================================
  Files         399      399              
  Lines       59410    59413       +3     
  Branches     1100     1100              
==========================================
+ Hits        51658    51662       +4     
+ Misses       7111     7110       -1     
  Partials      641      641

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 81a7939...92d2293. Read the comment docs.

@astrofrog - Done.

Anyone to review/approve? 🙏

Sorry, I looked this over and was relatively convinced it should work based on the regression test case.

Doc build is still failing after rebase and its log indicates that it went over the max 10 minutes limit.

...
reading sources... [ 95%] timeseries/times
reading sources... [ 95%] uncertainty/index
Too long with no output (exceeded 10m0s)

Does this PR somehow make the doc build too long? @astrofrog or @Cadair , how do we increase the limit on CircleCI?

This PR should not increase the build time, and the error looks similar to the ones with the recent mpl issue. I can try a new rebase to be sure?

I can try a new rebase to be sure?

Feel free to try again. Thanks! (I was playing with the CircleCI buttons and accidentally restarted everything... ops.)

All builds pass! Merging since it was approved :)