Change symbolLiteral to match actual implementation, clean up reserved words formatting #1588
Conversation
That was because |
|
OK, thanks! |
specification/dartLangSpec.tex
Outdated
| that does not begin with an underscore (`\code{\_}'), | ||
| evaluates to an instance of \code{Symbol} representing the identifier \id. | ||
| All occurrences of \code{\#\id} evaluate to the same instance | ||
| that does not begin with an underscore (`\code{\_}') or \VOID{} |
There was a problem hiding this comment.
Maybe rewrite as "is \VOID{} or an identifier that does not ..." to avoid the ambiguity of whether "or \VOID{}" binds to the "is" or the "not begin with". Also potentially ambiguous whether we're saying that void is an identifier.
Maybe it can be done with commas too, or by adding "either"/"or".
"where \id{} is either an identifier that does not begin with an underscore (`\code{_}'), or is the reserved word \VOID{},"
Comma after \VOID{} at least.
There was a problem hiding this comment.
The wording was actually rather sloppy. I revised the section as a whole, such that we only specify canonicalization twice (once for the private identifier, once for all other cases). I also moved the static analysis (just one sentence) from the end to the beginning, to bring about the ordering of topics that we've used in all other sections that were substantially revised.
3 participants
This CL changes symbol literals to allow
#void, because we have allowed that for several years.We have had the test
tests/corelib/symbol_reserved_word_test.dartsince 2014, but it is badly flawed. In particular, it is a multi-test, but it tests many cases where a compile-time error is expected using//# ...: continued, which means that the test will pass as long as at least one of the desired compile-time errors occur. However, some parts of the test have been executed in a meaningful manner, so we do have a history of behavior from this test. The analyzer has no current failures on this test, so it's a non-breaking change to specify the behavior that it actually tests.I've changed the test to express the intent stated in comments in that file (https://dart-review.googlesource.com/c/sdk/+/195503). In particular, it is no longer a multi-test, which fixes the logical error.
However, the test claims that
#voidandconst Symbol('void')should not be a compile-time error, and thatnew Symbol('void')should not be a run-time error. This is the change which is reflected by this PR.#void.foois flagged as an error by the analyzer and expected to be an error (so it's not a breaking change to keep considering this to be an error). It does change the reason:#voidis accepted, and.foois a member access on that symbol, so the error is now thatSymboldoes not have afoo. However, that's not a breaking change.#foo.voidwas a syntax error and remains a syntax error.The remaining part of
symbol_reserved_word_test.darttests that#wordis an error wheneverwordis a reserved word, and that is also true after this PR.I have no idea why
#voidwas allowed (even though the spec never said so), but I suspect that it could be because that symbol occurs somewhere in the result of an invocation of adart:mirrorsmethod. In any case, it seems safe and non-breaking to add support for#voidand keep all the other reserved words an error.Finally, this PR cleans up the formatting of the reserved words and moves a couple of lexical rules away from section 'Reserved Words' and into the section where the only reference to those rules occurs.