Add diagnostic coverage, remove possibly-unreachable unary op diagnostic #4519
Conversation
toolchain/parse/context.cpp
Outdated
| } else { | ||
| CARBON_CHECK(!IsLexicallyValidInfixOperator(), | ||
| "whitespace should have been required for an infix parse"); |
There was a problem hiding this comment.
I spent some time digging into this. It's almost not possible to reach this diagnostic, because almost every single one of our symbolic prefix or postfix operators happens to also be valid as an infix operator -- and so if a use were lexically valid as an infix operator, it would have been parsed as one. The exception is that the -- and ++ operators have no infix form. And even though they can only appear at the start of a statement, it is possible to reach this diagnostic for them.
Here's a testcase that produces the "whitespace is required before this unary operator" diagnostic:
fn F() { var n: i32 = 0; if (false) {}--n; }
There was a problem hiding this comment.
Thanks! I've added a parse test, with some minimization.
Note, I also added a similar test that doesn't repro, but produces a very similar parse tree so probably should, but leaving it as a todo.
I'm working to make sure remaining diagnostics have coverage, at least the ones I'd previously added a TODO for. Note in particular that I couldn't figure out a repro for UnaryOperatorRequiresWhitespace; if you have one, I can add a test, but otherwise maybe it's actually unreachable due to being diagnosed through infix logic (or, maybe this'll let fuzzing tell me an example).