That is, if you want to have distinct tokens for multiplication
and pointer type asterisks. I don't see why you would need to do
that distinction in the lexer though. You can just lex * as an
asterisk, and later in parser figure out what that asterisk meant.
The same way various parenthesis and brackets, or comma are often
overloaded in the programming languages, but that doesn't
complicate their lexing in any way.
The Haskell indentation is much more complicated.
Your example illustrates that parser cannot operate (decide
between variable definition or an expression) without also
processing typedef statements. So C forces part of renaming to be
done in parsing. That is unfortunate coupling, but it's different
coupling.
- Oleg
The fun (?) thing about C syntax is that you _cannot_ defer this. Consider the following (invalid) C program:
int main(void) {
t * x;
int t;
t * x;
}
When I pass this through gcc, what I get is:
file.c: In function ‘main’:
file.c:2:3: error: unknown type name ‘t’
2 | t * x;
| ^
file.c:4:5: error: invalid operands to binary * (have ‘int’ and ‘int *’)
4 | t * x;
| ^
The first 't * x' statement was parsed as a declaration of the variable 'x' with as type 't*'. The second such statement was parsed as a multiplication. The difference in behaviour is the declaration of 't' as a variable in between.
When starting this email I thought that the default was the other way round, i.e. 't * x' is parsed as a multiplication unless 't' is defined as a type; this would be accomplished by e.g. 'typedef int t;'. However it seems that the default, at least in gcc 13.2.1, is a variable declaration. Luckily (?), the point stands that to lex C, if you want to distinguish multiplication from the pointer type symbol, you need communication from the parser.
- Tom
On 01/11/2023 01:51, Oleg Grenrus wrote:
In C, AFAIU you can (and probably should) defer `typedef` usage recognition to a separate "renamer/ name resolution" pass. In Haskell we are forced to do name resolution after parsing, as we don't need to declare stuff before use. Even so, separate pass is usually a good idea anyway, you are better equipped to produce good error messages. In fact GHC does even more: it defers the unbound names reporting to the type checking phase, so it can give the types to unbound variables, like:
Prelude> x : "foo"
<interactive>:2:1: error: Variable not in scope: x :: Char
- Oleg
On 1.11.2023 2.32, Brandon Allbery wrote:
Feedback between lexer and parser isn't exactly unusual. Consider that parsing a C `typedef` generally needs to feed back to the lexer so uses will be recognized properly.
On Wed, Nov 1, 2023 at 12:28 AM Oleg Grenrus <oleg.grenrus@iki.fi> wrote:
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.
On 1.11.2023 0.31, Travis Athougies wrote:
[1] https://www.mail-archive.com/haskell-prime@haskell.org/msg01938.html which can be imp
- Oleg
According to the Haskell report [1] (See Note 5), a virtual `}` token> _______________________________________________ > 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.
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.
--
brandon s allbery kf8nh
_______________________________________________ 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.
_______________________________________________ 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.