Restore compatibility with Python 3.9's legacy LL(1) parser yet again #4426
Conversation
|
I don't want to have to do this again, so we need a CI config which will test on the LL1 parser before I'll be happy to merge this. |
I don't think there's much more to this other than setting up a normal Python 3.9 environment, and setting That said, I only tackled the with-statements in |
|
Yeah, I think this is just running our |
As a follow-up to 422425f, the legacy LL(1) parser also parses `with (expr1, expr2):` as simple with-statements, with a single tuple as the context manager expression. Such an expression can easily arise accidentally due to the use of an auto-formatter. Guard against this by rewriting these statements as nested with-statements and adding an appropriate comment about LL(1) parsability.
the-13th-letter
force-pushed
the
restore-py39-compatibility-LL1-parser
branch
from
b83c52e to
6285f22
Compare
|
Rebased on The failed |
Liam-DeVoe
left a comment
Liam-DeVoe
left a comment
There was a problem hiding this comment.
Looks good, thanks! (@Zac-HD I'll give you a bit to check my tox configuration if you want to)
|
[I was going to merge this now, but realized it requires a code-owner review] |
3 participants
Fixes #4421, yet again, this time taking care of compound with-statements with parentheses, which Python 3.9/LL(1) would parse as simple with-statements with a tuple as the single context manager expression.
I don't know why this didn't trigger when I originally tested my fix in #4423, but it does now: on pretty much every
hypothesis.given(...), I get anAttributeErrorinhypothesis.internal.conjecture.engine.ConjectureRunner.run.This PR now splits up all compound with-statements into nested simple ones and adds a comment on why they shouldn't be combined again.
As before, I only tackle theI double-checked every with-statementsrcdirectory, not thetestsor otherwise.inI could find for false positives/negatives and forgotten constellations.src