Tatsh / deltona / 24947783792 / 5

100%

master: 100%

push

github

refactor: preserve commas in single-element LHS tuple unpacks

When classifying an opening `(`, `_prev_sig` walked backwards across
`NEWLINE` tokens and could reach a `NAME` from the previous logical
statement (for example the `int` in `count: int\n(count,) = ...`).
That mis-classified the opener as a function call, so
`_is_required_single_element` returned `False` and the rewriter
stripped the comma from `(count,)`. The result, `(count) = ...`, is a
parenthesised target rather than a 1-tuple unpack, silently changing
the assignment's semantics.

Add a `_STATEMENT_BREAKS` set covering `NEWLINE`, `INDENT`, `DEDENT`,
and `ENCODING` and have `_prev_sig` return `None` on hitting one, so
backward classification stops at the start of the current statement.
`NL` (in-expression continuation) is unchanged, keeping multi-line
forward scans for trailing commas working as before.

Add regression tests covering the original shape (after a type
annotation in a function), at module top, and after an arbitrary
prior statement.

Signed-off-by: Andrew Udvare <audvare@gmail.com>

395 of 395 branches covered (100.0%)

Branch coverage included in aggregate %.

3289 of 3289 relevant lines covered (100.0%)

1.0 hits per line