More consistently report problems with non-linear patterns. #2231
Conversation
Also document limitation, and suggest a fix. Closes racket#1386 Closes racket#2152
|
I like these documentation changes. The code changes seem to have broken a test, however. |
|
Yes, the code changes are a bit too aggressive. I have to figure out how to make it less so, but it's not totally obvious. |
|
@AlexKnauth this is relevant to #4304 |
|
While working on #4304 I found that just erroring when nonlinear patterns are detected via the For example, I would like (match '(1 2 3 4)
[(list a ... a) a])to be an error, but the renaming of |
| @@ -175,16 +177,21 @@ | |||
| (and (bound-identifier=? v v*) (or id (list v v*))))) | |||
| => | |||
| (lambda (id) | |||
| (if (identifier? id) | |||
| (if (and (identifier? id) (not (inside-GSeq?))) | |||
There was a problem hiding this comment.
This (not (inside-GSeq?)) check also doesn't allow nonlinear patterns inside the "tail" of GSeq patterns, such as in (list a _ ... a) and (cons a (list pre ... a post ...)).
That seems too restrictive to me. In my opinion it should be restricted based on the pattern itself being under ellipses, not based on it being in the tail after something else that's under ellipses.
There was a problem hiding this comment.
I agree. This PR is too aggressive as I said at the time, but I didn't settle on a better fix.
Also document limitation, and suggest a fix.
Closes #1386 Closes #2152