Fwd: Proposal: Roundtrip serialization of Cmm (parser-compatible pretty-printer output)
Diego Antonio Rosario Palomino
diegorosario2013 at gmail.com
Mon Jul 28 18:26:31 UTC 2025
---------- Forwarded message ---------
De: Diego Antonio Rosario Palomino <diegorosario2013 at gmail.com>
Date: lun, 28 jul 2025 a la(s) 12:56 p.m.
Subject: Re: Proposal: Roundtrip serialization of Cmm (parser-compatible
pretty-printer output)
To: Hécate <hecate at glitchbra.in>
Hello all,
Thank you for the thoughtful responses so far, and thank you Simon for
summarizing Andreas's comments.
*"Do you have any use-cases in mind? Suppose you were 100% successful —
would anyone use it?"*
Yes — my mentor, *Csaba Hruska*, would. He's currently working on a custom
STG optimizer that uses experimental techniques to enable whole-program
optimizations for Haskell code. The intended pipeline is:
*GHC STG → custom optimizer → textual Cmm → code generation*
However, the current *parseable* Cmm is not sufficient for his use case,
because it *cannot represent everything the Cmm AST can express*.
Beyond this specific use case, achieving *roundtrip serializability* for
Cmm could make it a *viable alternative to LLVM* for Haskell projects.
Native code generation via Cmm is much faster than through LLVM. And while
outputting LLVM from Cmm currently produces *less performant* code than
directly targetting LLVM, I believe the inefficiencies could be fixed
relatively easily. Enabling such improvements is part of the motivation for
my documentation work — to help developers understand and work with Cmm and
its infrastructure.
*"You need a compelling reason to change the input language (understood by
the parser) since libraries may include .cmm files, which will break. (It'd
be interesting to audit Hackage to see how many libraries do include such
.cmm files.)"*
To clarify, this proposal would *not* break backwards compatibility. There
are two implementation paths:
1.
Introduce a *second parser* that accepts a syntax 100% identical to the
pretty printer output.
2.
Extend the *current parser* by adding a mode (or block) that uses a
distinct keyword (e.g., low_level_unwrapped) to indicate: "expect exact
syntax, no convenience fills."
In either case, existing .cmm files would continue to be supported as-is.
The current parser wouldn't need features removed or changed — the new
syntax would *only add capabilities*.
*"It’s unclear from your example how those blocks would work exactly. Is
low_level_unwrapped a label? If so can we goto it? Is it a keyword?
Something else entirely?"* — Andreas
Apologies for the confusion — I’m not well-versed in the formal terminology.
To clarify: low_level_unwrapped (or very_low_level, or another name) would
be a *keyword or syntactic construct* that tells the parser to interpret
the contents of the block { ... } using a syntax *identical to what the
pretty printer emits*.
For example:
function1 { } // existing low-level syntax
function2() { } // existing high-level syntax
very_low_level { ... } // new mode: code with exact pretty-printed
syntax inside the block
*"Rather than change the language understood by the parser, would it not be
easier to change the language spat out by the pretty-printer to be
compatible with the parser?"*
Unfortunately, that’s not a practical path forward.
At the start of the project, Csaba (my mentor) recommended leaving the
parser mostly untouched and focusing instead on extending the pretty
printer. However, we’ve realized that the differences between the parser
and the pretty printer are not trivial. The parser — even in its current
“low-level” mode — *inserts inferred data* via convenience functions.
It *abstracts
part of the structure*, meaning we cannot fully recover the original Cmm
ADT just by parsing.
In other words, *modifying the pretty printer to match the parser would
require it to lose information* — which I strongly oppose. If Cmm is
generated programmatically, the pretty-printed version would lack
structural information present in the internal data structure. And
parseable Cmm would still be *incapable of expressing all features of the
AST*.
I hope that also addresses your concern, Hécate.
This GSoC project runs until *November 10th*. I was granted extra time
since, unlike most participants, I’m not working through summer vacation —
I’m in the Southern Hemisphere.
(Also, I realize I previously used the wrong project name in this thread —
the correct title of my GSoC project is *“Documenting and improving Cmm.”*)
Regarding the risk of *bitrot* in a new parser or new syntax mode: one
possible mitigation would be to add *regression tests* that check whether
parsing a file and pretty-printing it results in compatible output.
On a related note, I’ve noticed that *some Cmm examples in the
documentation and even in source code comments are incorrect or outdated*.
Part of my work includes identifying and correcting these inconsistencies.
Thanks again to everyone for your time and input — I greatly appreciate the
discussion and feedback.
Best regards,
*Diego Antonio Rosario Palomino*
GSoC 2025 – Documenting and improving Cmm
El lun, 28 jul 2025 a la(s) 11:04 a.m., Hécate via ghc-devs (
ghc-devs at haskell.org) escribió:
> Hi Diego,
>
> Thank you very much for your work in this direction, it's sorely needed.
>
> I'm all for having proper roundtrip correctness for Cmm, but I am not sure
> altering the parser is the way to go.
> In my opinion, GHC should produce valid textual Cmm, that can be ingested
> by the parser at it is today.
>
> Have a nice day,
> Hécate
> Le 28/07/2025 à 02:16, Diego Antonio Rosario Palomino a écrit :
>
> Hello GHC devs,
>
> I'm currently working on Cmm documentation and tooling improvements as
> part of my Google Summer of Code project. One of my core goals is to make
> Cmm roundtrip serializable.
>
> Right now, the in-memory Cmm data structure—generated programmatically
> (e.g., from STG via GHC)—can be pretty-printed, and Cmm can also be parsed.
> However, the pretty-printed version is not compatible with the parser. That
> is, we cannot take the output of the pretty printer and feed it directly
> back into the parser.
>
> Example:
>
> Parseable version:
>
> sum {
> cr:
> bits64 x;
> x = R1 + R2;
> R1 = x;
> jump %ENTRY_CODE(Sp(0))[R1];
> }
>
> Pretty-printed version:
>
> sum() { // []
> { info_tbls: []
> stack_info: arg_space: 8
> }
> {offset
> cf: // global
> _ce::I64 = R1 + R2;
> R1 = _ce::I64;
> call (I64[Sp + 0 * 8])(R1) args: 8, res: 0, upd: 8;
> }
> }
>
> Another example:
>
> Parseable version:
>
> simple_sum_4 { // [R2, R1]
> cr: // global
> bits64 _cq;
> _cq = R2;
> bits64 _cp;
> _cp = R1;
> R1 = _cq + _cp;
> jump (bits64[Sp])[R1];
> }
>
> Pretty-printed version:
>
> simple_sum_4() { // []
> { info_tbls: []
> stack_info: arg_space: 8
> }
> {offset
> cs: // global
> _cq::I64 = R2;
> _cr::I64 = R1;
> R1 = _cq::I64 + _cr::I64;
> call (I64[Sp])(R1) args: 8, res: 0, upd: 8;
> }
> }
>
> While it’s possible to write parseable Cmm that resembles the
> pretty-printed version (and hence the internal ADT), they don’t fully
> match—mainly because the parser inserts inferred fields using convenience
> functions.
>
> Proposal:
>
> To make roundtrip serialization possible, I propose supporting a new
> syntax that matches the pretty printer output exactly.
>
> There are a couple of design options:
>
> 1.
>
> Create a separate parser that accepts the pretty-printed syntax. Files
> could then use either the current parser or the new strict one.
> 2.
>
> Extend the current parser with a dedicated block syntax like:
>
> low_level_unwrapped {
> ...
> }
>
> This second option is the one my mentor recommends, as it may better
> reflect GHC developers' preferences. In this mode, the parser would not
> insert any inferred data and would expect the input to match the
> pretty-printed form exactly.
>
> This would enable a true roundtrip:
>
> -
>
> Compile Haskell to Cmm (in-memory AST)
> -
>
> Pretty-print and write it to disk (wrapped in low_level_unwrapped {
> ... })
> -
>
> Later read it back using the parser and continue with codegen
>
> Optional future direction:
>
> As a side note: currently the parser has both a “high-level” and a
> “low-level” mode. The low-level mode resembles the AST more closely but
> still inserts some inferred data.
>
> If we introduce this new “exact” low-level form, it's possible the
> existing low-level mode could become redundant. We might then have:
>
> -
>
> High-level syntax
> -
>
> New low-level (exact)
> -
>
> And possibly deprecate the current low-level variant
>
> I’d be interested in your thoughts on whether that direction makes sense.
>
> Serialization libraries?
>
> One technically possible—but likely unacceptable—alternative would be to
> derive serialization via a library like aeson. That would enable
> serializing and deserializing the Cmm AST directly. However, I understand
> that aeson adds a large dependency footprint, and likely wouldn't be
> suitable for inclusion in GHC.
>
> Final question:
>
> Lastly—I’ve heard that parts of the Cmm pipeline may currently be under
> refactoring. If that’s the case, could you point me to which parts (parser,
> pretty printer, internal representation, etc.) are being modified? I’d like
> to align my efforts accordingly and avoid conflicts.
>
> Thanks very much for your time and input! I'm happy to iterate on this
> based on your feedback.
>
> Best regards,
> Diego Antonio Rosario Palomino
> GSoC 2025 – Cmm Documentation & Tooling
>
> _______________________________________________
> ghc-devs mailing listghc-devs at haskell.orghttp://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs
>
> --
> Hécate ✨
> 🐦: @TechnoEmpress
> IRC: Hecate
> WWW: https://glitchbra.in
> RUN: BSD
>
> _______________________________________________
> ghc-devs mailing list
> ghc-devs at haskell.org
> http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/ghc-devs/attachments/20250728/e35d6a8b/attachment.html>
More information about the ghc-devs
mailing list