[red-knot] Add initial set of tests for unreachable code #17159
Conversation
sharkdp
force-pushed
the
david/unreachable-code-tests
branch
from
aae6786 to
ee6e92b
Compare
|
| inside the unreachable section would not cause problems at runtime. And type checking the | ||
| unreachable code under the assumption that it *is* reachable might lead to false positives: |
There was a problem hiding this comment.
I like that reasoning. I assume that only means we won't emit diagnostics. Would we still infer types for those sections?
It will be interesting to see how we integrate this into the linter... but that's a problem for another day
There was a problem hiding this comment.
I think we can still infer types for unreachable sections. Whether it's useful is less clear; you are likely to see a lot of Never types, since no binding from outside the unreachable code section can reach the unreachable code section.
| inside the unreachable section would not cause problems at runtime. And type checking the | ||
| unreachable code under the assumption that it *is* reachable might lead to false positives: |
There was a problem hiding this comment.
I think we can still infer types for unreachable sections. Whether it's useful is less clear; you are likely to see a lot of Never types, since no binding from outside the unreachable code section can reach the unreachable code section.
|
I suspect that a lot of "don't emit diagnostics in unreachable code" will fall out naturally from the prevalence of In fact it may be better to let it fall out like this than to implement a blanket policy of "no diagnostics in unreachable code". An argument could be made that in the cases where there is a binding inside the unreachable code section, we should still emit diagnostics on invalid uses of that binding. For example, if Unresolved-reference is an exception here; we definitely have to silence this in unreachable code, as a special case. |
|
@carljm we may need special handling at least for |
|
Yeah, "unused type-ignore" is another case, like unresolved-reference, that might need special handling. |
I'm not at all sure we can predict which error codes are going to have false positives in unreachable code. This PR also has examples of |
|
That may be true, but I don't think adding |
|
Hmm, I suppose so. Are we sure we won't add more error codes in the future that are similarly issues around boundness, though? It feels somewhat fragile to special-case specific error codes. Anyway, we can iterate on it! |
* origin/main: (35 commits) [red-knot] Callable types are disjoint from literals (#17160) [red-knot] Fix inference for `pow` between two literal integers (#17161) [red-knot] Add GitHub PR annotations when mdtests fail in CI (#17150) [red-knot] Fix equivalence of differently ordered unions that contain `Callable` types (#17145) [red-knot] Add initial set of tests for unreachable code (#17159) [`airflow`] Move `AIR302` to `AIR301` and `AIR303` to `AIR302` (#17151) ruff_db: simplify lifetimes on `DiagnosticDisplay` [red-knot] Detect division-by-zero in unions and intersections (#17157) [`airflow`] Add autofix infrastructure to `AIR302` name checks (#16965) [`flake8-bandit`] Mark `str` and `list[str]` literals as trusted input (`S603`) (#17136) [`airflow`] Add autofix for `AIR302` attribute checks (#16977) [`airflow`] Extend `AIR302` with additional symbols (#17085) [`airflow`] Move `AIR301` to `AIR002` (#16978) [`airflow`] Add autofix for `AIR302` method checks (#16976) ruff_db: switch diagnostic rendering over to `std::fmt::Display` [red-knot] Add 'Goto type definition' to the playground (#17055) red_knot_ide: update snapshots red_knot_python_semantic: remove comment about `TypeCheckDiagnostic` ruff_db: delete most of the old diagnostic code red_knot: use `Diagnostic` inside of red knot ...
* origin/main: (82 commits) [red-knot] Fix more [redundant-cast] false positives (#17170) [red-knot] Three-argument type-calls take 'str' as the first argument (#17168) Control flow: `return` and `raise` (#17121) Bump 0.11.3 (#17173) [red-knot] Improve `Debug` implementation for `semantic_index::SymbolTable` (#17172) [red-knot] Fix `str(…)` calls (#17163) [red-knot] visibility_constraint analysis for match cases (#17077) [red-knot] Fix playground crashes when diagnostics are stale (#17165) [red-knot] Callable types are disjoint from literals (#17160) [red-knot] Fix inference for `pow` between two literal integers (#17161) [red-knot] Add GitHub PR annotations when mdtests fail in CI (#17150) [red-knot] Fix equivalence of differently ordered unions that contain `Callable` types (#17145) [red-knot] Add initial set of tests for unreachable code (#17159) [`airflow`] Move `AIR302` to `AIR301` and `AIR303` to `AIR302` (#17151) ruff_db: simplify lifetimes on `DiagnosticDisplay` [red-knot] Detect division-by-zero in unions and intersections (#17157) [`airflow`] Add autofix infrastructure to `AIR302` name checks (#16965) [`flake8-bandit`] Mark `str` and `list[str]` literals as trusted input (`S603`) (#17136) [`airflow`] Add autofix for `AIR302` attribute checks (#16977) [`airflow`] Extend `AIR302` with additional symbols (#17085) ...
…7159) ## Summary Add an initial set of tests that will eventually document our behavior around unreachable code. In the last section of this suite, I argue why we should never type check unreachable sections and never emit any diagnostics in these sections.
Summary
Add an initial set of tests that will eventually document our behavior around unreachable code. In the last section of this suite, I argue why we should never type check unreachable sections and never emit any diagnostics in these sections.