Add support for wasm exception handling to Emscripten target #131830

hoodmane · 2024-10-17T12:05:31Z

This is a draft because we need some additional setting for the Emscripten target to select between the old exception handling and the new exception handling. I don't know how to add a setting like that, would appreciate advice from Rust folks. We could maybe choose to use the new exception handling if Ctarget-feature=+exception-handling is passed? I tried this but I get errors from llvm so I'm not doing it right.

cc @kleisauke @kripken @sbc100 @workingjubilee @alexcrichton

rustbot · 2024-10-17T12:05:41Z

r? @compiler-errors

rustbot has assigned @compiler-errors.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

rust-log-analyzer · 2024-10-17T12:34:54Z

The job x86_64-gnu-llvm-18 failed! Check out the build log: (web) (plain)

Click to see the possible cause of the failure (guessed by this bot)

------
 > importing cache manifest from ghcr.io/rust-lang/rust-ci-cache:c32c805632780b5c1de330e3f44561b336c2efe163bc0990acb392390157a8e1d9f855d75914a239aa40c49d77f4a837247d05d2f8d46f554b98e1f46712a3e3:
------
##[endgroup]
Setting extra environment values for docker:  --env ENABLE_GCC_CODEGEN=1 --env GCC_EXEC_PREFIX=/usr/lib/gcc/
[CI_JOB_NAME=x86_64-gnu-llvm-18]
debug: `DISABLE_CI_RUSTC_IF_INCOMPATIBLE` configured.
---
sccache: Starting the server...
##[group]Configure the build
configure: processing command line
configure: 
configure: build.configure-args := ['--build=x86_64-unknown-linux-gnu', '--llvm-root=/usr/lib/llvm-18', '--enable-llvm-link-shared', '--set', 'rust.randomize-layout=true', '--set', 'rust.thin-lto-import-instr-limit=10', '--set', 'change-id=99999999', '--enable-verbose-configure', '--enable-sccache', '--disable-manage-submodules', '--enable-locked-deps', '--enable-cargo-native-static', '--set', 'rust.codegen-units-std=1', '--set', 'dist.compression-profile=balanced', '--dist-compression-formats=xz', '--set', 'rust.lld=false', '--disable-dist-src', '--release-channel=nightly', '--enable-debug-assertions', '--enable-overflow-checks', '--enable-llvm-assertions', '--set', 'rust.verify-llvm-ir', '--set', 'rust.codegen-backends=llvm,cranelift,gcc', '--set', 'llvm.static-libstdcpp', '--enable-new-symbol-mangling']
configure: target.x86_64-unknown-linux-gnu.llvm-config := /usr/lib/llvm-18/bin/llvm-config
configure: llvm.link-shared     := True
configure: rust.randomize-layout := True
configure: rust.thin-lto-import-instr-limit := 10
---
failures:

---- [codegen] tests/codegen/emcripten-catch-unwind.rs stdout ----

error: verification with 'FileCheck' failed
status: exit status: 1
command: "/usr/lib/llvm-18/bin/FileCheck" "--input-file" "/checkout/obj/build/x86_64-unknown-linux-gnu/test/codegen/emcripten-catch-unwind/emcripten-catch-unwind.ll" "/checkout/tests/codegen/emcripten-catch-unwind.rs" "--check-prefix=CHECK" "--check-prefix" "NONMSVC" "--allow-unused-prefixes" "--dump-input-context" "100"
--- stderr -------------------------------
/checkout/tests/codegen/emcripten-catch-unwind.rs:49:12: error: CHECK: expected string not found in input
/checkout/tests/codegen/emcripten-catch-unwind.rs:49:12: error: CHECK: expected string not found in input
 // CHECK: [[ALLOCA:%.*]] = alloca
           ^
/checkout/obj/build/x86_64-unknown-linux-gnu/test/codegen/emcripten-catch-unwind/emcripten-catch-unwind.ll:14:7: note: scanning from here
      ^
      ^
/checkout/obj/build/x86_64-unknown-linux-gnu/test/codegen/emcripten-catch-unwind/emcripten-catch-unwind.ll:22:10: note: possible intended match here
 %catchpad2.i = catchpad within %catchswitch1.i [ptr null]

Input file: /checkout/obj/build/x86_64-unknown-linux-gnu/test/codegen/emcripten-catch-unwind/emcripten-catch-unwind.ll
Check file: /checkout/tests/codegen/emcripten-catch-unwind.rs


-dump-input=help explains the following input dump.
Input was:
<<<<<<
<<<<<<
            1: ; ModuleID = 'emcripten_catch_unwind.df8ae0c697b5edf5-cgu.0' 
            2: source_filename = "emcripten_catch_unwind.df8ae0c697b5edf5-cgu.0" 
            3: target datalayout = "e-m:e-p:32:32-p10:8:8-p20:8:8-i64:64-f128:64-n32:64-S128-ni:1:10:20" 
            4: target triple = "wasm32-unknown-emscripten" 
            5:  
            6: ; Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(none) uwtable 
            7: define noundef i32 @ptr_size() unnamed_addr #0 { 
            9:  ret i32 4 
           10: } 
           11:  
           12: ; Function Attrs: uwtable 
           12: ; Function Attrs: uwtable 
           13: define noundef i32 @test_catch_unwind(ptr nocapture noundef nonnull readonly %try_fn, ptr noundef %data, ptr nocapture noundef nonnull readonly %catch_fn) unnamed_addr #1 personality ptr @__gxx_wasm_personality_v0 { 
           14: start: 
check:49'0           X error: no match found
           15:  invoke void %try_fn(ptr %data) 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           16:  to label %__rust_try.exit unwind label %catchswitch.i 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
check:49'0     ~
check:49'0     ~
           18: catchswitch.i: ; preds = %start 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           19:  %catchswitch1.i = catchswitch within none [label %catchpad.i] unwind to caller 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
check:49'0     ~
check:49'0     ~
           21: catchpad.i: ; preds = %catchswitch.i 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           22:  %catchpad2.i = catchpad within %catchswitch1.i [ptr null] 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
check:49'1              ?                                                  possible intended match
           23:  %0 = tail call ptr @llvm.wasm.get.exception(token %catchpad2.i) 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           24:  %1 = tail call i32 @llvm.wasm.get.ehselector(token %catchpad2.i) 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           25:  call void %catch_fn(ptr %data, ptr %0) [ "funclet"(token %catchpad2.i) ] 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           26:  catchret from %catchpad2.i to label %__rust_try.exit 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
check:49'0     ~
check:49'0     ~
           28: __rust_try.exit: ; preds = %start, %catchpad.i 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           29:  %common.ret.op.i = phi i32 [ 0, %start ], [ 1, %catchpad.i ] 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           30:  ret i32 %common.ret.op.i 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~
           31: } 
check:49'0     ~~
check:49'0     ~
check:49'0     ~
           33: declare i32 @__gxx_wasm_personality_v0(...) unnamed_addr #2 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
check:49'0     ~
check:49'0     ~
           35: ; Function Attrs: mustprogress nocallback nofree nosync nounwind willreturn 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           36: declare ptr @llvm.wasm.get.exception(token) #3 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
check:49'0     ~
check:49'0     ~
           38: ; Function Attrs: mustprogress nocallback nofree nosync nounwind willreturn 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           39: declare i32 @llvm.wasm.get.ehselector(token) #3 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
check:49'0     ~
check:49'0     ~
           41: attributes #0 = { mustprogress nofree norecurse nosync nounwind willreturn memory(none) uwtable "target-cpu"="generic" "target-features"="+exception-handling" } 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           42: attributes #1 = { uwtable "target-cpu"="generic" "target-features"="+exception-handling" } 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           43: attributes #2 = { "target-cpu"="generic" } 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           44: attributes #3 = { mustprogress nocallback nofree nosync nounwind willreturn } 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
check:49'0     ~
check:49'0     ~
           46: !llvm.module.flags = !{!0} 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~
           47: !llvm.ident = !{!1} 
check:49'0     ~~~~~~~~~~~~~~~~~~~~
check:49'0     ~
check:49'0     ~
           49: !0 = !{i32 8, !"PIC Level", i32 2} 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           50: !1 = !{!"rustc version 1.84.0-nightly (c92ba44cb 2024-10-17)"} 
check:49'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
------------------------------------------

workingjubilee · 2024-10-18T23:51:22Z

This is a draft because we need some additional setting for the Emscripten target to select between the old exception handling and the new exception handling.

Hm. I feel a little out-of-context here. Could you explain

what makes the "old" exception-handling old?
what makes the "new" exception-handling new?
what parts of an emscripten program must agree on which exception handling we're using?
- in the minimal case: linkage of a Rust main.rs, std, and Emscripten?
- in other configurations of linkage that Emscripten supports?
why we must have a toggle for this?

Is "new" meaning the wasm exception-handling proposal with exnref and wasm-level try blocks?

hoodmane · 2024-10-21T08:59:59Z

Thanks for the questions.

The short answer is these are two different ABIs. Perhaps it would be most accurate to think of wasm32-emscripten-old-eh and wasm32-emscripten-new-eh as distinct targets. This is similar to -sWASM_BIGINT. Unfortunately, Emscripten has tons of different ABIs. But some of the ABIs are better than others in terms of code speed, size, and bugginess.

What makes the "old" / "new" exception-handling old / new?

Well the older system has existed for a lot longer and worked with the first version of WebAssembly deployed in browsers. The new exception handling uses native support and has only been universally available in engines since ~March 2022.

what parts of an emscripten program must agree on which exception handling we're using?

We need different IR, different object files, and different linker arguments. But it's transparent to Rust/C++/C code.

in the minimal case: linkage of a Rust main.rs, std, and Emscripten?

Yeah so we need a distinct build of the standard library because the object files are different. The compilation of both main.rs and std need to change, namely by changing the implementation of the try intrinsic. Perhaps only libpanicunwind is affected and other libs that don't use the try intrinsic are unaffected, but it's unclear. Also, the linker arguments change.

why we must have a toggle for this?

Currently, Pyodide is stuck using the old ABI because Rust only supports the old ABI. If Rust switched to only supporting the new ABI, it would make the transition difficult.

hoodmane · 2024-10-21T09:00:09Z

To make matters worse, the new exception handling support was changed in October of 2023 so there are now two new exception handling variants:

the old new exception handling, which is universally supported but also deprecated, and
the new new exception handling, which is currently only supported in the most recent release of Firefox.

I think there is no difference in the IR the frontend generates between the old new and new new variants, though the lowering pass has distinct code for the two cases, the object file format is different, and so the new variant will require a different setting code generation time. It's a fairly simple transform to adjust an old object file to support the new variant and they plan "to provide a converter tool from the old binary to the new format to facilitate the use of the new instructions without the need for recompilation."

So we may end up with three ABIs.

sbc100 · 2024-10-21T16:57:53Z

I think there is no difference in the IR the frontend generates between the old new and new new variants

That is true today but the plan is to make llvm support the new variant natively ASAP. However at the at point I imagine there will be a lowering pass added to binaryen to support the previous version, so you should still only need a single object file format / libc build.

sbc100 · 2024-10-21T17:00:23Z

Unfortunately, Emscripten has tons of different ABIs.

While its true emscripten supports a lot of different options almost all of them are link-time only settings and don't effect the object file ABI. There are only subset of settings that effect the object format (i.e. that are vaslid at compile time): https://github.com/emscripten-core/emscripten/blob/10cb9d46cdd17e7a96de68137c9649d9a630fbc7/tools/settings.py#L81-L112. Note that WASM_BIGINT is not one of them.

hoodmane · 2024-10-21T19:40:24Z

Are MAIN_MODULE/SIDE_MODULE really compile time settings? I only pass them in LDFLAGS...

sbc100 · 2024-10-21T19:43:53Z

Are MAIN_MODULE/SIDE_MODULE really compile time settings? I only pass them in LDFLAGS...

Only in the sense that they enable -fPIC.. you can use -fPIC itself instead if you prefer.

workingjubilee · 2024-10-21T21:44:30Z

In general, "more_compiler_abi_flags == more_compatibility" is actually "more_compiler_flags == incompatibility", because you now have to have even more flags set to the correct values to compile code correctly.

So I would prefer that we not expose, nor even pretend that we wish to offer compatibility with, "tons of different ABIs". We already have quite enough problems with ABI-affecting flags. It does not really take many booleans to be able to count to a million, and a few million untested variants of Rust is not a thing we should be trying to support.

Having said that, supporting the new ABI behind an unstable flag that will then be ditched at some point has precedent, especially for wasm targets, see #117919

workingjubilee · 2024-10-21T22:01:37Z

If a -Z flag for such a transition period is acceptable, @hoodmane, then like that flag did, opening an MCP, would be the next step: https://github.com/rust-lang/compiler-team/issues/new?template=major_change.md

While we Just Merged the soundness-fixing change of #131533 we usually do MCPs for Significant Changes To Preexisting Targets, especially ones that require this... shindig. A transition period and involvement of rustc's CLI.

They're usually sensible enough that it's a bit of a formality, so don't overthink it, the answers to my questions are more-or-less what you would paste in there, with a bit of additional stitching up to explain why e.g. Pyodide needs a flag for this transition and we don't want to do just a hard cut over.

hoodmane · 2024-10-22T12:16:44Z

do MCPs

Sounds good. Thanks again @workingjubilee for all of your help. I really appreciate it.

It does not really take many booleans to be able to count to a million, and a few million untested variants of Rust is not a thing we should be trying to support.

Yeah this is why I'm raising this as a concern. It seems clearly out of scope to support all of them. It's important to have a solution to this in mind ahead of time. I would prefer that Rust only supports the 2 configurations I care about and has real test coverage for them, but different projects will standardize on different options. It's a thorny issue coordination problem. It would be nice if Emscripten could formalize specific triples...

hoodmane · 2024-10-22T12:17:11Z

The fact that the ABIs can be linearly sorted by runtime support date is helpful. Instead of 20 different booleans leading to 2^20 different options toggling each independently, there are 21 options when we sort. Then if you only support:

the most recent date with a -Z flag
some option that you think is "old enough to be universally available"

maybe it's good enough. But having the in-between steps available as -Z flags would be helpful for debugging the target.

hoodmane · 2024-10-31T14:33:17Z

Okay I opened a MCP:
rust-lang/compiler-team#801

Add support for wasm exception handling to Emscripten target

d315a30

rustbot assigned compiler-errors Oct 17, 2024

rustbot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. T-libs Relevant to the library team, which will review and decide on the PR/issue. labels Oct 17, 2024

hoodmane marked this pull request as draft October 17, 2024 12:05

workingjubilee added the O-emscripten Target: 50% off wasm32-unknown-musl. the savings come out of stdio.h, but hey, you get SDL! label Oct 21, 2024

workingjubilee added the A-ABI Area: Concerning the application binary interface (ABI) label Oct 21, 2024

hoodmane mentioned this pull request Oct 31, 2024

Unstable flag to choose Emscripten unwinding abi rust-lang/compiler-team#801

Open

3 tasks

compiler-errors added S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Nov 4, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add support for wasm exception handling to Emscripten target #131830

Add support for wasm exception handling to Emscripten target #131830

hoodmane commented Oct 17, 2024 •

edited

Loading

rustbot commented Oct 17, 2024

rust-log-analyzer commented Oct 17, 2024

workingjubilee commented Oct 18, 2024 •

edited

Loading

hoodmane commented Oct 21, 2024

hoodmane commented Oct 21, 2024 •

edited

Loading

sbc100 commented Oct 21, 2024

sbc100 commented Oct 21, 2024

hoodmane commented Oct 21, 2024

sbc100 commented Oct 21, 2024

workingjubilee commented Oct 21, 2024

workingjubilee commented Oct 21, 2024 •

edited

Loading

hoodmane commented Oct 22, 2024

hoodmane commented Oct 22, 2024 •

edited

Loading

hoodmane commented Oct 31, 2024

Add support for wasm exception handling to Emscripten target #131830

Are you sure you want to change the base?

Add support for wasm exception handling to Emscripten target #131830

Conversation

hoodmane commented Oct 17, 2024 • edited Loading

rustbot commented Oct 17, 2024

rust-log-analyzer commented Oct 17, 2024

workingjubilee commented Oct 18, 2024 • edited Loading

hoodmane commented Oct 21, 2024

hoodmane commented Oct 21, 2024 • edited Loading

sbc100 commented Oct 21, 2024

sbc100 commented Oct 21, 2024

hoodmane commented Oct 21, 2024

sbc100 commented Oct 21, 2024

workingjubilee commented Oct 21, 2024

workingjubilee commented Oct 21, 2024 • edited Loading

hoodmane commented Oct 22, 2024

hoodmane commented Oct 22, 2024 • edited Loading

hoodmane commented Oct 31, 2024

hoodmane commented Oct 17, 2024 •

edited

Loading

workingjubilee commented Oct 18, 2024 •

edited

Loading

hoodmane commented Oct 21, 2024 •

edited

Loading

workingjubilee commented Oct 21, 2024 •

edited

Loading

hoodmane commented Oct 22, 2024 •

edited

Loading