Syntax errors for jumps within `eval`

[andrykonchin]

Syntax errors for jumps within `eval`

[kddnewton]

This is by design. When you pass scopes, we assume you are parsing an eval. When parsing an eval, a syntax error is not raised for those jumps. Instead, it relies on the compiler to evaluate whether or not they are valid. This matches CRuby semantics.

[andrykonchin]

Hm, but SyntaxError is raised in CRuby:

$ ruby -v test.rb
ruby 3.3.0 (2023-12-25 revision 5124f9ac75) [x86_64-darwin23]
(eval at test.rb:1): (eval at test.rb:1):1: Invalid break (SyntaxError)

test.rb:

eval("def m; break; end")

---

Ah, you mean it's a responsibility of compiler. Got it 👍

[kddnewton]

Yeah, unfortunately. Consider for instance the difference between 0 rescue next and eval("0 rescue next"). The second raises Can't escape from eval with next.

[eregon]

Is it mostly about different error messages? I'm not sure if that difference is important enough to have this behavior differ with scopes.
For the def m; break; end case it seems to behave the same within or outside eval in CRuby, so it feels like Prism could always give a syntax error for that one, no?
It's not very important, but it would avoid duplicating this logic in the compiler/translator, where it feels wrong/annoying to do so.

[eregon]

Some other examples:
def m; retry; end Prism seems to always error for this, eval or not, good.
module M; yield; end Prism seems to only handle (= error) it if not in eval. This feels very much like a syntactic restriction, so would be nicer if Prism handled it always.

[kddnewton]

I'll try to find which tests were failing, but when I implemented this I got a couple more test-all tests passing. There are some things that are relying on this behavior.

[kddnewton]

Ahh, it was the ERB tests. We need this to not error out in case you're parsing an ERB file

[eregon]

We need this to not error out in case you're parsing an ERB file

I'm not following, could you give a concrete example?
Which specific jump should not error when parsing an ERB file?
And how comes there is no error from compile.c then, i.e. how come one can distinguish between parser and compile.c in that case? (eval does both).

[eregon]

Ah maybe it's related to Ripper? I've seen https://bugs.ruby-lang.org/issues/20460 now