ISLE: improve syntax BNF

[eagr]

• correct <pattern-arg> syntax(?)
• re-org items in order to improve readability

[eagr]

for sake of completeness. I suppose they're also ignored by the parser?

[cfallin]

Actually no -- the full list of whitespace characters is here. The original list was correct.

[eagr]

Should have double checked first, sorry about that.

[eagr]

let's just use the dumbest mapping :)

[cfallin]

Eh, there is a reason I had chosen typedecl -- type is too close to ty (the latter reads as an abbreviation of the former with no semantic difference), the existing nonterminal for type expressions, and there is a semantic difference between uses of a type and the form that declares them. Let's stick with typedecl.

[eagr]

technically not a value, could be confusing

[cfallin]

Well, here I would disagree again. You can call it a "type expression" if you want, but it's a value in the universe of types. It is definitely not a typedef: typedef is, in my mind, another name for the whole form that defines a new type (like a typedef in C/C++ for example). Perhaps "type body" if you want to distinguish the semantic place this occupies -- an expanded form that can be named in a type declaration form's right-hand side but not in-line wherever types are given.

[eagr]

Yeah, typedef is worse than typevalue.. I did mean something like "type body". How about

<typedef> ::= <ident> [ "extern" | "nodebug" ] <type-body>

to set it apart from <decl>?

[eagr]

technically it's a field of the variant

[eagr]

I could be very wrong about this. But having this kind of infix operator defined in S-expression feels weird and I couldn't find it in the extractor parser, so I think this is a typo?

[cfallin]

This is outdated syntax, sorry -- we removed argument polarity a while ago. The whole <expr> case needs to go away and you can inline <pattern> for <pattern-arg> since we don't have the distinction anymore.

[github-actions]

Subscribe to Label Action

cc @cfallin, @fitzgen

This issue or pull request has been labeled: "cranelift", "isle"

Thus the following users have been cc'd because of the following labels:

• cfallin: isle
• fitzgen: isle

To subscribe or unsubscribe from this label, edit the .github/subscribe-to-label.json configuration file.

Learn more.

[cfallin]

Thanks for the PR! Unfortunately this introduces some inaccuracies and I have a lot of comments -- see below.

[cfallin]

Actually no -- the full list of whitespace characters is here. The original list was correct.

[cfallin]

Eh, there is a reason I had chosen typedecl -- type is too close to ty (the latter reads as an abbreviation of the former with no semantic difference), the existing nonterminal for type expressions, and there is a semantic difference between uses of a type and the form that declares them. Let's stick with typedecl.

[cfallin]

Well, here I would disagree again. You can call it a "type expression" if you want, but it's a value in the universe of types. It is definitely not a typedef: typedef is, in my mind, another name for the whole form that defines a new type (like a typedef in C/C++ for example). Perhaps "type body" if you want to distinguish the semantic place this occupies -- an expanded form that can be named in a type declaration form's right-hand side but not in-line wherever types are given.

[cfallin]

This is outdated syntax, sorry -- we removed argument polarity a while ago. The whole <expr> case needs to go away and you can inline <pattern> for <pattern-arg> since we don't have the distinction anymore.

[eagr]

Out of curiosity, is there a reason for convert instead of convertor, given the extractor?

[cfallin]

Out of curiosity, is there a reason for convert instead of convertor, given the extractor?

Locally it made sense as "convert T to U using foo"; one could debate the many options further but we already had that discussion three(?) years ago and it's unlikely to change now absent good reason.