Yes, the "communication between lexer and parser" is exactly what GHC does.

Amelia has a nice post about it https://amelia.how/posts/parsing-layout.html which made it click it for me.

Note, you don't actually need to use alex and happy, you can do hand-written lexer and parsec (or alex and parsec, ...). The key insight is to have stateful lexer, and control it from the parser.

Amelia's post grammar is a bit too strict, e.g. GHC accepts real semis in virtual layout, and also empty "statements" in between, so we can write

   \x y z -> case x of True -> y;;;;;; False -> z

but that's easy (at least in parsec) to adjust the parser grammar to accept those.

Or, you can *approximate* the parse-error rule with "alternative layout rule" [1], which can be implemented as a pass between lexing and parsing, or as a stateful lexer (but in this case parser won't need to adjust lexer's state). GHC has an undocumented AlternativeLayoutRule extension, so you can experiment with it to see what it accepts (look for tests in GHC source for examples). It handles let-in bindings well enough.

[1] https://www.mail-archive.com/haskell-prime@haskell.org/msg01938.html which can be imp

- Oleg

On 1.11.2023 0.31, Travis Athougies wrote:
According to the Haskell report [1] (See Note 5), a virtual `}` token
is inserted if parsing the next token would cause a parse error and the
indentation stack is non-empty.

I'm trying to lex and parse Haskell source and this sort of interplay
(which requires two-way communication between lexer and parser) makes
it very difficult to write a conformant implementation.

I can't change the standard (obviously), but I'm wondering if this is
actually what GHC (de facto the only Haskell compiler) does, or if it
applies some other rule. If so, does anyone know the exact mechanism of
its implementation?

I've been programming Haskell for more than a decade, and while I have
an intuitive understanding of the indentation rules, I would have
assumed the source could be lexed without also having a parser. In
particular, the note seems to imply that the main purpose of this is to
properly lex `let`/`in` bindings. Perhaps there's an alternate
equivalent rule?

Curious to hear other's thoughts.

Travis

[1]
https://www.haskell.org/onlinereport/haskell2010/haskellch10.html#x17-17800010.3
> _______________________________________________ > Haskell-Cafe mailing list > To (un)subscribe, modify options or view archives go to: > http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe > Only members subscribed via the mailman list are allowed to post.