@dpgeorge This PR solves the problem and aioble works now, but it's worth noting that this actually triggered a seg fault, because the nlr raise in the middle of poll_set_add_obj left the element in poll_set->map partially initialised (specifically elem->value is NULL, because that happens right at the end).

This might be a case for the new nlr jump callback, to cleanup if the ioctl failed? i.e. it seems wrong that poller.register can raise and leave the poller in an invalid state if the pollable's ioctl raises.

Code size report:

   bare-arm:    +0 +0.000% 
minimal x86:    +0 +0.000% 
   unix x64:    +0 +0.000% standard
      stm32:    +0 +0.000% PYBV10
     mimxrt:    +0 +0.000% TEENSY40
        rp2:    +0 +0.000% RPI_PICO
       samd:    +0 +0.000% ADAFRUIT_ITSYBITSY_M4_EXPRESS

Added a change to the ThreadSafeFlag test to make it no longer sensitive to the order that the scheduler runs relative to the main code. e.g. the ioctl(MP_STREAM_GET_FILENO) call could trigger the scheduler.

Codecov Report

Merging #12443 (3e204b8) into master (c854d0e) will decrease coverage by 0.01%.
The diff coverage is n/a.

❗ Current head 3e204b8 differs from pull request most recent head fae83a6. Consider uploading reports for the commit fae83a6 to get more accurate results

@@            Coverage Diff             @@
##           master   #12443      +/-   ##
==========================================
- Coverage   98.39%   98.38%   -0.01%     
==========================================
  Files         158      158              
  Lines       20939    20940       +1     
==========================================
  Hits        20602    20602              
- Misses        337      338       +1     

see 4 files with indirect coverage changes

it's worth noting that this actually triggered a seg fault, because the nlr raise in the middle of poll_set_add_obj left the element in poll_set->map partially initialised (specifically elem->value is NULL, because that happens right at the end).

Yes I was aware of this issue when writing the code. It was really hard to solve so I just left it and instead added the comment in poll_set_add_obj:

            // If an exception is raised below when adding the new object then the map entry for that
            // object remains unpopulated, and methods like poll() may crash.  This case is not handled.

This might be a case for the new nlr jump callback, to cleanup if the ioctl failed?

The poll change was made before nlr jump callbacks were added, so, yes, maybe we can use nlr jump callbacks to fix this issue. It's low priority though, because a well-behaved program shouldn't crash in this way.