search

tc39 / proposal-ptc-syntax

Discussion and specification for an explicit syntactic opt-in for Tail Calls.
http://tc39.github.io/proposal-ptc-syntax/
169 stars 8 forks source link

Should `continue` be part of an expression? #8

Closed littledan closed 8 years ago

littledan commented 8 years ago

I see two ways that the return continue syntax could be specified:

Option A

There are two fixed special forms, return continue and => continue. These evaluate whether the following expression contains a tail call in the same way as in ES2015.

Option B

continue MemberExpression Arguments is a new way of making a function call be a tail call. It can only be used in expression contexts--in statement context, it is still a labelled continue, with the arguments forming a SyntaxError. Syntactically, each particular tail call has to be marked--so a ternary operator would have its individual branches marked. It would be an Early Error to use continue in a place which is not in tail context.

I haven't figured out how these fit into the grammar or spec of PTC/STC yet, so these are just vague ideas at the moment. Note that they would also both apply for syntax like return ^foo(x), but not for !return foo(x) or function-level syntax.

bterlson commented 8 years ago

I feel like return continue and => continue are more conservative and thus more likely to be acceptable to people. I think the distinction between expr and statement forms can also get confusing.

However, option B does have the benefit of working better with ternary without requiring a complicated grammar.

(Note that option B would need to also include MemberExpression TemplateLiteral).

bterlson commented 8 years ago

Now that I've gone over the options, I don't think the latter option is that onerous, and it has the huge advantage of keeping continue right before the call. Unless I'm missing something obvious, grammatically, then I think ReturnExpression : continue MemberExpression Arguments and friends is the way to go. I've took a stab at defining this grammar here: https://tc39.github.io/proposal-ptc-syntax.

claudepache commented 8 years ago

I find Option B more natural, because tail position is a property of a (sub)expression, not of the entire return statement.

I've took a stab at defining this grammar here: https://tc39.github.io/proposal-ptc-syntax.

In that stab, I think that the ReturnExpression production would be better named TailCallExpression.

bterlson commented 8 years ago

@claudepache Agreed, just pushed an update for that.

littledan commented 8 years ago

Seems like the spec has stabilized on "yes".