Free of syn by janhohenheim · Pull Request #66 · bevyengine/variadics_please

janhohenheim · 2026-04-29T22:07:23Z

Objective

syn is hella slow: https://fasterthanli.me/articles/the-virtue-of-unsynn

Solution

switch to unsynn, which is what facet uses

Testing

cargo test runs
running a cargo build on my machine with a heavily customized setup reports the following timing:
- main: 2.16 s
- unsynn: 0.56 s

Disclaimer: I'm not a metaprogramming expert or anything, so please take a critical look at this PR before merging :)

Followup

Add to https://github.com/fasterthanlime/free-of-syn!

github-actions · 2026-04-29T22:07:38Z

Welcome, new contributor!

Please make sure you've read our contributing guide and we look forward to reviewing your pull request shortly ✨

janhohenheim · 2026-04-29T23:56:47Z

  "MIT-0",
  "Unlicense",
  "Zlib",
+  "Unicode-3.0",


This just allows unicode symbols to be redistributed freely. Seems alright :)

janhohenheim · 2026-04-29T23:57:02Z

 [dependencies]
-syn = "2.0"
 quote = "1.0"
-proc-macro2 = "1.0"


proc-macro2 is re-exported by unsynn

janhohenheim · 2026-04-29T23:57:38Z

 license = "MIT OR Apache-2.0"
 keywords = ["bevy", "variadics", "docs"]
-rust-version = "1.81.0"
+rust-version = "1.83.0"


Looked it up, this is the minimal version that supports rust-lang/rust#128183, which unsynn uses

alice-i-cecile · 2026-04-30T00:02:14Z

 #![allow(missing_docs, reason = "Not all docs are written yet, see #3492.")]
-#![cfg_attr(any(docsrs, docsrs_dep), feature(doc_cfg, rustdoc_internals))]
+#![cfg_attr(any(docsrs, docsrs_dep), feature(doc_cfg))]
+#![expect(


Surely we should move this to the actual error enum, not the crate itself.

it triggers for something inside the unsynn! macro. I don't know how to move the expect there :/

Comment to this effect please :)

And issue upstream ideally.

Done... I think? I'm very confused by Radicle (their git forge). Locally it's showing me the issue, but not on the webpage. Since the page is self-hosted, I assume they just didn't run a sync yet.

Can't link the issue since I don't know the hyperlink to it yet.

janhohenheim · 2026-04-30T00:16:34Z

Note: this regresses the error messages a bit
Before:

After:

I'm sure this can be fixed by emitting a custom error message with quote_spanned! + compile_error!, but I'm not sure how to do that exactly

janhohenheim · 2026-04-30T00:45:12Z

Update: diagnostics should now look the same :)

janhohenheim · 2026-04-30T00:51:28Z

+    })
+}
+
+fn pretty_print_error(err: Error) -> TokenStream {


Unfortunately there no upstream pretty printing yet: https://seed.pipapo.org/nodes/seed.pipapo.org/rad:z39WbeupErKS8TwbDS5yU8eZSa3C/issues/960feccb89aef6452b5ec0a6ce6a00604c8d8d21

Link this in that issue please to help them out, and leave this link as a comment in the code please!

Done... I think? I'm very confused by Radicle (their git forge). Locally it's showing me the issue, but not on the webpage. Since the page is self-hosted, I assume they just didn't run a sync yet.

Co-authored-by: atlv <email@atlasdostal.com>

atlv24

nice work Jan!!

alice-i-cecile

Remember to write release notes!

alice-i-cecile · 2026-04-30T01:10:13Z

-syn = "2.0"
 quote = "1.0"
-proc-macro2 = "1.0"
+unsynn = "0.3"


The primary complaint here is that we're increasing supply-chain risk by doing this: single maintainer, relatively small name, new-ish crate.

That said, given the counterfactual and meaningfully improved compile time benchmarks I'm happy to take this trade!

Yeah I'm primarily basing my own trust on Amos (fasterthanlime), since he is a big name and is heavily pushing this crate.

alice-i-cecile

Broadly on board, but needs a bit of cleanup now :)

janhohenheim · 2026-04-30T01:35:41Z

@alice-i-cecile done!

cart

This is likely to actually regress Bevy compile times, as we are compiling syn for a million other reasons already, meaning this actually adds an additional crate to compile.

The only way this could be faster is if the cost of compiling a whole additional AST parsing library is offset by the unsynn macro logic being faster across all use cases. And for all I know unsynn could be slower to execute!

Lets benchmark

janhohenheim · 2026-04-30T11:16:09Z

I use variadics_please in non-Bevy projects and don’t really plan on unsynning all of Bevy. I can check the compile time differences on Bevy later, but I personally want to look at this PR in isolation, since that is how variadics_please markets itself.

Update: I checked out Bevy.
This is using my trusty config.

syn:

hhh@hhh-fedora ~/g/bevy (main)> hyperfine 'cargo build' --prepare 'cargo clean'
Benchmark 1: cargo build
  Time (mean ± σ):     45.751 s ±  1.311 s    [User: 525.875 s, System: 81.569 s]
  Range (min … max):   44.392 s … 48.276 s    10 runs

unsynn:

hhh@hhh-fedora ~/g/bevy (main)> hyperfine 'cargo build' --prepare 'cargo clean'
Benchmark 1: cargo build
  Time (mean ± σ):     44.810 s ±  0.677 s    [User: 521.251 s, System: 80.650 s]
  Range (min … max):   44.058 s … 45.896 s    10 runs

At first glance it looks like the mean is down for unsynn, but the standard deviation is high enough that one should not take that as a giveaway.
The stastical read is "we have no evidence for the compile times being different"

On the other hand, given that there's no obvious downside, I'm happy with this result, since it matters for non-syn consumers

cart · 2026-04-30T19:36:21Z

Thank you for the diligence! I would expect this PR in particular to be a regression somewhere in the range of 0 to 0.5 seconds (as all_tuples! is a relatively "low cost" macro ... not a lot of performance to claw back there) , and we know that we're compiling an additional crate that costs about ~0.5. In my mind these numbers make perfect sense: a potential slight regression hidden in the noise of 0.677 second standard deviation.

That being said, I do find the promise of unsynn and incremental improvements to the (rather sorry) state of proc macros to be compelling. I'm content to start running this experiment. It would be most interesting to port one of the "higher parsed token volume" Bevy macros that read all of the struct fields, across many structs in the Bevy codebase, to actually justify the cost of unsynn in the tree / actually prove the theory here.

That being said, given that macro_rules derives and "const-reflect driven proc macros" are both on the horizon, investing in an ecosystem-wide porting effort could end up being very wasteful, as those each have pretty significant advantages (and I expect us to want to take advantage if / when they land).

Free of syn

8b016a5

janhohenheim added 12 commits April 30, 2026 00:13

Clippy

2a53c02

Remove some hand parsing

5f78d36

Remove manual parsing

8103d92

Add validation back in

af0c2ef

Improve comment

c2a41f4

Improve parser

649da91

Improve error

b36a53d

Allow clippy

a81b90a

Bump msrv

01b2de7

Allow unicode license

9bc828e

Deny syn

21312e9

Lint

56a666f

janhohenheim commented Apr 29, 2026

View reviewed changes

alice-i-cecile requested a review from atlv24 April 30, 2026 00:00

alice-i-cecile reviewed Apr 30, 2026

View reviewed changes

Add at field

3bbf5e1

alice-i-cecile reviewed Apr 30, 2026

View reviewed changes

Comment thread src/lib.rs

janhohenheim added 2 commits April 30, 2026 02:40

Improve diagnostics

42bb6de

Improve diagnostics

90b2d94

janhohenheim requested a review from alice-i-cecile April 30, 2026 00:45

janhohenheim commented Apr 30, 2026

View reviewed changes

janhohenheim added 3 commits April 30, 2026 02:52

Lint

269aa04

Shrimple

00a9556

Comments

83eca7d

atlv24 reviewed Apr 30, 2026

View reviewed changes

Comment thread src/lib.rs Outdated

Update src/lib.rs

05af7c6

Co-authored-by: atlv <email@atlasdostal.com>

atlv24 approved these changes Apr 30, 2026

View reviewed changes

alice-i-cecile requested changes Apr 30, 2026

View reviewed changes

alice-i-cecile reviewed Apr 30, 2026

View reviewed changes

janhohenheim requested a review from alice-i-cecile April 30, 2026 01:35

janhohenheim added 2 commits April 30, 2026 03:50

Feedback

2cf29ae

lint

9dc660b

alice-i-cecile approved these changes Apr 30, 2026

View reviewed changes

cart requested changes Apr 30, 2026

View reviewed changes

alice-i-cecile merged commit 77b1fae into bevyengine:main Apr 30, 2026
27 checks passed

janhohenheim mentioned this pull request Apr 30, 2026

Add variadics_please fasterthanlime/free-of-syn#15

Open

Uh oh!

Conversation

janhohenheim commented Apr 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Objective

Solution

Testing

Followup

Uh oh!

github-actions Bot commented Apr 29, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

janhohenheim commented Apr 30, 2026

Uh oh!

janhohenheim commented Apr 30, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

atlv24 left a comment

Choose a reason for hiding this comment

Uh oh!

alice-i-cecile left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

alice-i-cecile left a comment

Choose a reason for hiding this comment

Uh oh!

janhohenheim commented Apr 30, 2026

Uh oh!

cart left a comment

Choose a reason for hiding this comment

Uh oh!

janhohenheim commented Apr 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cart commented Apr 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

janhohenheim commented Apr 29, 2026 •

edited

Loading

janhohenheim commented Apr 30, 2026 •

edited

Loading

cart commented Apr 30, 2026 •

edited

Loading