CC @gpshead .

We may need to understand the following decision that cache was removed for this case.
@vstinner, Is this different behavior from what @iritkatriel intended?

41a64a5

I have doubts about this issue as well. It came up in #94410 (comment), where it seemed to make sense, but I've seen other issues where we do/don't want various things cached, and I don't know if there's an obviously correct thing to do here.

We may need to understand the following decision that cache was removed for this case.

Sorry, I don't recall the rationale for this change in Python 3.

updatecache() changes:

• commit 921c824 (1992, initial commit adding Lib/linecache.py): except IOError: (...) return []
• commit c341c62 (1992) iterates on sys.path and return [] if the file is not found
• commit aada7b8 (2010) adds except IOError: pass -- introduce a BUG! lines doesn't exist in this case (NameError)
• commit 78f43a7 (2010) replaces pass with lines = [] -- fix the NameError
• commit 41a64a5 (2010) replaces lines = [] with return []

We should instead cache the negative lookup result and bypass the search in future queries.

It sounds reasonable to me.