[3.7] bpo-40807: Show warnings once from codeop._maybe_compile (GH-20486)

[miss-islington]

• bpo-40807: Show warnings once from codeop._maybe_compile

• Move catch_warnings

• news

Co-authored-by: Terry Jan Reedy [email protected]
(cherry picked from commit 052d3fc)

Co-authored-by: Cheryl Sabella [email protected]

https://bugs.python.org/issue40807

[miss-islington]

@csabella and @terryjreedy: Status check is done, and it's a failure ❌ .

[miss-islington]

@csabella and @terryjreedy: Status check is done, and it's a failure ❌ .

[terryjreedy]

With the test.support change backported, the new test runs on 3.7 and fails on
self.assertEqual(len(w.warnings), 1) because '0 is 0' does not cause SyntaxWarning in 3.7. I believe the same might be true of '\e' and DeprecationWarning. We could make the test pass by changing 1 to 0, but that would be equivalent to deleting the test, and the blurb would still be wrong.

I think we should hold off on this until we learn of code that will make the current test (and blurb) relevant, or until we are sure that there is no such code, or until the next and final 3.7 bugfix release (in about a month).

[terryjreedy]

See previous comment. Either change test with relevant code, so it passes, or close upon deciding there is no such code or 3.7 goes into security mode.

[terryjreedy]

You forgot that the outer quotes are not part of the compiled syntax. (I have done this too.) \e is a syntax error, '\e' is valid and gives no warning in 3.7.

[csabella]

Sorry about that. I was actually having trouble getting the '\e' to show the Deprecation Warning on any release until I started the REPL or IDLE with the ./python -Wall. I did this change quickly just to see how the tests would be run and it looks like I did it too quickly. :-(

[terryjreedy]

The important point is '\e' warns when starting IDLE (legitimately) with py -3.7 -Wall -m idlelib. So the backport is needed.

Should the tests always pass, not just with CI? At least if one does not explicitly turn off warnings for tests (which seems like a bad idea)? I made a local 3.7 branch from this PR and ran python -m test test_codeop and it passed normally. python -m test.test_codeop reports 5 tests rather than the previous 4 and prints the DeprecationWarning.

So I will merge and we will see how this fares on buildbots.

[terryjreedy]

20 seconds later I realize that warnings <= 1 is sufficient to prove not 3 or even 2. But 0 could also result from a bad patch that suppressed too much. So if a fix is needed, better to add a context forcing the warning on.