Disallow `export_name` starting with "llvm." #140837

moxian · 2025-05-09T03:37:46Z

Fixes #140822
Motivation in the issue, and in code comments.

This implementation adds the check to right to #[export_name] parsing code. I considered moving it closer to the place where the new name gets emitted to llvm, but that looked harder to get right, and easier to accidentally break going forward.

I don't believe any other attributes need this handling. The only one that comes close is link_name but that one uses the name, and does not define it, and in fact #[link_name = "llvm.stuff"] is entirely justified and actively used in real code.

Two small drive-by changes in separate commits (removed an obsolete and misleading //FIXME comment, and tighten the diagnostics span for the null-byte-in-export_name check). Happy to split them into dedicated PR(s) if this is deemed valuable.

This comment from 2017 says "every error should have an associated error code" However, a more recent rustc-dev-guide doc from 2020 suggests that this is not the case, and that errorcode-less errors are fine and sometimes even preferable to errorcode-full ones: rust-lang/rustc-dev-guide#967

rustbot · 2025-05-09T03:37:51Z

r? @lcnr

rustbot has assigned @lcnr.
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

rustbot · 2025-05-09T03:37:52Z

Some changes occurred in compiler/rustc_codegen_ssa/src/codegen_attrs.rs

cc @jdonszelmann

Some changes occurred in compiler/rustc_codegen_ssa

cc @WaffleLapkin

rustbot · 2025-05-09T03:51:18Z

Reminder, once the PR becomes ready for a review, use @rustbot ready.

moxian · 2025-05-09T04:59:13Z

@rustbot ready

Simply because it looks a little bit prettier that way, IMO

Urgau · 2025-05-09T09:11:19Z

compiler/rustc_codegen_ssa/messages.ftl

@@ -52,6 +52,9 @@ codegen_ssa_expected_one_argument = expected one argument

 codegen_ssa_expected_used_symbol = expected `used`, `used(compiler)` or `used(linker)`

+codegen_ssa_export_name_llvm_intrinsic = exported symbol name must not start with "llvm."
+    .note = symbols starting with "llvm." are reserved for LLVM intrinsics and cannot be defined in user code


We don't have a concept of "user code" so let's not confuse users into believing there is one.

Suggested change

.note = symbols starting with "llvm." are reserved for LLVM intrinsics and cannot be defined in user code

.note = symbols starting with `llvm.` are reserved for LLVM intrinsics

Urgau · 2025-05-09T09:11:40Z

compiler/rustc_codegen_ssa/messages.ftl

@@ -52,6 +52,9 @@ codegen_ssa_expected_one_argument = expected one argument

 codegen_ssa_expected_used_symbol = expected `used`, `used(compiler)` or `used(linker)`

+codegen_ssa_export_name_llvm_intrinsic = exported symbol name must not start with "llvm."


Suggested change

codegen_ssa_export_name_llvm_intrinsic = exported symbol name must not start with "llvm."

codegen_ssa_export_name_llvm_intrinsic = exported symbol name must not start with `llvm.`

(same below)

jieyouxu · 2025-05-09T09:22:22Z

Note

Please ignore this nomination, as it's probably more of a bug-fix. Left unresolved for reference only.

Background

Rejecting #[export_name] whose value starts with llvm. seems very reasonable, but then it also feels at the same time somewhat arbitrary, because we now hard-error on "certain" #[export_name] values (namely, those that begin with llvm.). A few considerations:

This was rejected / permitted in an oscillating fashion previously, as the rejection depended on verify-llvm-ir (i.e. the rust frontend was not the one "actively" (or rather, explicitly) performing the rejection).
We almost certainly need to be rejecting this, because otherwise

Creating a definition for an intrinsic is invalid LLVM IR, the rest is not really relevant. The Rust frontend must generate an error when trying to use an export_name starting with llvm..
-- Target features getting erased when using #[export_name] with names of LLVM intrinsics #140822 (comment)
We should make an explicit decision whether the rustc frontend should be explicitly rejecting this.

Possible breakage

I don't really expect any breakages (someone is likely doing some... weird things if this causes their code to no longer compile), but this is technically constraining the set of programs that we accept.

Nomination questions Some questions

~~(T-compiler, T-lang)~~ Should we perform this explicit rejection for export_names that begin with llvm.?
~~(T-compiler, T-lang)~~ Does the llvm. rejection also happen when the cg backend is e.g. cg_gcc? Or do we only reject llvm. when cg backend is cg_llvm?
- Probably universally reject (across any and all cg backends, that is)?
~~(T-lang)~~ If we universally reject, do we document this at the language level (Reference) as sth like "IR intrinsic name limitations"?

stale

jieyouxu · 2025-05-09T10:01:08Z

EDIT: I removed the lang nomination, probably more of a "bug fix" / "impl detail".

Noratrieb · 2025-05-09T10:28:07Z

GitHub code search yields no results; https://github.com/search?q=%22export_name+%3D+%5C%22llvm.%22&type=code
(it finds stuff without the period so it's probably not wrong: https://github.com/search?q=%22export_name+%3D+%5C%22llvm%22&type=code)

this is such a weird and broken thing to do that I think we can just ship this without crater or FCW or anything

lcnr · 2025-05-09T14:56:31Z

r? compiler

WaffleLapkin · 2025-05-09T15:01:41Z

Is there a reason this is handled in rustc_codegen_ssa and not rustc_codegen_llvm? I assume while weird, #[export_name = "llvm...."] should not break any other backends.

Urgau · 2025-05-09T15:03:56Z

Is there a reason this is handled in rustc_codegen_ssa and not rustc_codegen_llvm? I assume while weird, #[export_name = "llvm...."] should not break any other backends.

Wouldn't it be weird if your code is working on one backend but not on another one?

WaffleLapkin · 2025-05-09T15:07:46Z

@Urgau generally yes, but

this might already happen
the name overlaps with the backend's internal symbol, so I don't think in this case it's weird

jieyouxu · 2025-05-09T15:12:13Z

@Urgau generally yes, but

1. this might already happen

2. the name overlaps with the backend's internal symbol, so I don't think in this case it's weird

Yeah, I should've left the nomination comment open, also asked

~~(T-compiler, T-lang)~~ Should we perform this explicit rejection for export_names that begin with llvm.?

~~(T-compiler, T-lang)~~ Does the llvm. rejection also happen when the cg backend is e.g. cg_gcc? Or do we only reject llvm. when cg backend is cg_llvm?

Probably universally reject (across any and all cg backends, that is)?

~~(T-lang)~~ If we universally reject, do we document this at the language level (Reference) as sth like "IR intrinsic name limitations"?

traviscross · 2025-05-09T16:41:53Z

If we universally reject, do we document this at the language level (Reference) as sth like "IR intrinsic name limitations"?

Yes, this should be documented in the Reference.

@rustbot labels +S-waiting-on-documentation

cc @ehuss

rustbot assigned lcnr May 9, 2025

rustbot added A-attributes Area: Attributes (`#[…]`, `#![…]`) 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. labels May 9, 2025

This comment has been minimized.

Sign in to view

This comment was marked as outdated.

Sign in to view

rustbot 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 May 9, 2025

moxian force-pushed the llvm-export branch from 30b458e to 67f47ca Compare May 9, 2025 04:53

rustbot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. and removed S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. labels May 9, 2025

This comment has been minimized.

Sign in to view

moxian added 2 commits May 8, 2025 22:15

disallow export_name starting with "llvm."

453fcbc

Tighten the error span for explort_name null byte check

d443666

Simply because it looks a little bit prettier that way, IMO

moxian force-pushed the llvm-export branch from 67f47ca to d443666 Compare May 9, 2025 05:19

Urgau reviewed May 9, 2025

View reviewed changes

This comment was marked as resolved.

Sign in to view

jieyouxu removed I-lang-nominated Nominated for discussion during a lang team meeting. I-compiler-nominated Nominated for discussion during a compiler team meeting. labels May 9, 2025

jieyouxu removed I-lang-easy-decision Issue: The decision needed by the team is conjectured to be easy; this does not imply nomination T-lang Relevant to the language team, which will review and decide on the PR/issue. labels May 9, 2025

rustbot assigned Nadrieril and unassigned lcnr May 9, 2025

rustbot added the S-waiting-on-documentation Status: Waiting on approved PRs to documentation before merging label May 9, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Disallow `export_name` starting with "llvm." #140837

Disallow `export_name` starting with "llvm." #140837

moxian commented May 9, 2025 •

edited

Loading

rustbot commented May 9, 2025

rustbot commented May 9, 2025

This comment has been minimized.

This comment was marked as outdated.

rustbot commented May 9, 2025

moxian commented May 9, 2025

This comment has been minimized.

Urgau May 9, 2025

Urgau May 9, 2025

jieyouxu commented May 9, 2025 •

edited

Loading

This comment was marked as resolved.

jieyouxu commented May 9, 2025

Noratrieb commented May 9, 2025

lcnr commented May 9, 2025

WaffleLapkin commented May 9, 2025

Urgau commented May 9, 2025

WaffleLapkin commented May 9, 2025

jieyouxu commented May 9, 2025

traviscross commented May 9, 2025

	.note = symbols starting with "llvm." are reserved for LLVM intrinsics and cannot be defined in user code
	.note = symbols starting with `llvm.` are reserved for LLVM intrinsics

		@@ -52,6 +52,9 @@ codegen_ssa_expected_one_argument = expected one argument

		codegen_ssa_expected_used_symbol = expected `used`, `used(compiler)` or `used(linker)`

		codegen_ssa_export_name_llvm_intrinsic = exported symbol name must not start with "llvm."

Disallow export_name starting with "llvm." #140837

Are you sure you want to change the base?

Disallow export_name starting with "llvm." #140837

Conversation

moxian commented May 9, 2025 • edited Loading

rustbot commented May 9, 2025

rustbot commented May 9, 2025

This comment has been minimized.

This comment was marked as outdated.

rustbot commented May 9, 2025

moxian commented May 9, 2025

This comment has been minimized.

Urgau May 9, 2025

Choose a reason for hiding this comment

Urgau May 9, 2025

Choose a reason for hiding this comment

jieyouxu commented May 9, 2025 • edited Loading

Background

Possible breakage

Nomination questions Some questions

This comment was marked as resolved.

jieyouxu commented May 9, 2025

Noratrieb commented May 9, 2025

lcnr commented May 9, 2025

WaffleLapkin commented May 9, 2025

Urgau commented May 9, 2025

WaffleLapkin commented May 9, 2025

jieyouxu commented May 9, 2025

traviscross commented May 9, 2025

Disallow `export_name` starting with "llvm." #140837

Disallow `export_name` starting with "llvm." #140837

moxian commented May 9, 2025 •

edited

Loading

jieyouxu commented May 9, 2025 •

edited

Loading