ENH: mechanism for failing when warnings are issued without conversion to exceptions? #11518

Activity

1. RonnyPfannschmidt commented on Oct 18, 2023

   

   RonnyPfannschmidt

   on Oct 18, 2023

   Member

   More actions

   This is a dupe, let me find the one I made
2. RonnyPfannschmidt commented on Oct 18, 2023

   

   RonnyPfannschmidt

   on Oct 18, 2023

   Member

   More actions

   #10098 is not quite a dupe but similar enough

   I propose post test warnings failure as a feature
3. Zac-HD

   added

   plugin: warningsrelated to the warnings builtin pluginrelated to the warnings builtin plugin

   on Oct 21, 2023
4. Zac-HD commented on Oct 21, 2023

   

   Zac-HD

   on Oct 21, 2023

   Member

   More actions

   I'd consider this a bug in Numpy: the code described in scipy/scipy#18781 (comment) is going to give the wrong answer if users have warnings-as-errors, and for that matter if they try to ctrl-c it'll misbehave badly too. Instead of

   try:
      return np.float64(float(obj))
   except:
      return np.array(obj, dtype=np.float64)[()]

   you should do something like

   try:
      return np.float64(float(obj))
   except Warning:
       raise
   except Exception:
      return np.array(obj, dtype=np.float64)[()]
5. Zac-HD commented on Jul 22, 2026

   

   Zac-HD

   on Jul 22, 2026

   Member

   More actions

   Closing as resolved: pytest now has a --max-warnings/max_warnings option (#14372, closing #14371) that fails the run based on the post-test warnings summary count without requiring filterwarnings = error, which is exactly the feature discussed here and avoids the try/except-swallowing problem described above.

   (I do also think that hiding warnings when end-users set -Werror is regrettable though)
6. Zac-HD

   closed this as completedon Jul 22, 2026