Install libgccjit into the compiler's sysroot when cg_gcc is enabled by Kobzol · Pull Request #145341 · rust-lang/rust

Kobzol · 2025-08-13T08:40:47Z

This PR installs the libgccjit.so library (which is essentially GCC) into the rustc sysroot (stageN/lib on Linux) when the GCC codegen backend is enabled. This allows using the GCC codegen backend "out of the box" with the resulting rustc. It would be nice to get rid of the libgccjit.so.0 alias (https://rust-lang.zulipchat.com/#narrow/channel/131828-t-compiler/topic/Forcing.20unversioned.20dylib.20dependencies.20on.20Linux/with/534180740), but it's not blocking for this change.

You can try running x build std --set 'rust.codegen-backends=["llvm", "gcc"]' and then compiling a hello world with rustc +stage1 -Zcodegen-backend=gcc main.rs.

It is now also possible to build stage2 rustc when the GCC is configured to be the default codegen backend, without any further LD_LIBRARY_PATH hacks: ./x build compiler --stage 2 --set 'rust.codegen-backends=["gcc"]'.

After this change, it should be pretty simple to add a dist/rustup step for actually shipping cg_gcc to end users.

CC @GuillaumeGomez

r? @jieyouxu

…ckend is enabled

rustbot · 2025-08-13T08:40:52Z

jieyouxu is currently at their maximum review capacity.
They may take a while to respond.

rustbot · 2025-08-13T08:40:54Z

This PR changes how GCC is built. Consider updating src/bootstrap/download-ci-gcc-stamp.

…mponent

…ry when installing `libgccjit.so`

jieyouxu

Thanks

jieyouxu · 2025-08-13T12:30:31Z

@bors r+ rollup=never

bors · 2025-08-13T12:30:34Z

📌 Commit f40a784 has been approved by jieyouxu

It is now in the queue for this repository.

Kobzol · 2025-08-13T14:38:42Z

I don't think this needs rollup=never, tbh, as this code shouldn't even be exercised on CI, perhaps outside of one PR CI job (famous last words?).

@bors r+ rollup

bors · 2025-08-13T14:39:27Z

📌 Commit f40a784 has been approved by jieyouxu

It is now in the queue for this repository.

…ouxu Install libgccjit into the compiler's sysroot when cg_gcc is enabled This PR installs the `libgccjit.so` library (which is essentially GCC) into the rustc sysroot (`stageN/lib` on Linux) when the GCC codegen backend is enabled. This allows using the GCC codegen backend "out of the box" with the resulting rustc. It would be nice to get rid of the `libgccjit.so.0` alias (https://rust-lang.zulipchat.com/#narrow/channel/131828-t-compiler/topic/Forcing.20unversioned.20dylib.20dependencies.20on.20Linux/with/534180740), but it's not blocking for this change. You can try running `x build std --set 'rust.codegen-backends=["llvm", "gcc"]'` and then compiling a hello world with `rustc +stage1 -Zcodegen-backend=gcc main.rs`. It is now also possible to build stage2 rustc when the GCC is configured to be the default codegen backend, without any further `LD_LIBRARY_PATH` hacks: `./x build compiler --stage 2 --set 'rust.codegen-backends=["gcc"]'`. After this change, it should be pretty simple to add a dist/rustup step for actually shipping cg_gcc to end users. CC ``@GuillaumeGomez`` r? ``@jieyouxu``

Rollup of 9 pull requests Successful merges: - #144761 (aarch64: Make `outline-atomics` a known target feature) - #144949 (More `Printer` cleanups) - #144955 (search graph: lazily update parent goals) - #144962 (Add aarch64_be-unknown-none-softfloat target) - #145153 (Handle macros with multiple kinds, and improve errors) - #145241 ([AVR] Changed data_layout) - #145341 (Install libgccjit into the compiler's sysroot when cg_gcc is enabled) - #145349 (Correctly handle when there are no unstable items in the documented crate) - #145356 (Add another example for escaped `#` character in doctest in rustdoc book) r? `@ghost` `@rustbot` modify labels: rollup

Rollup merge of #145341 - Kobzol:codegen-backend-gcc, r=jieyouxu Install libgccjit into the compiler's sysroot when cg_gcc is enabled This PR installs the `libgccjit.so` library (which is essentially GCC) into the rustc sysroot (`stageN/lib` on Linux) when the GCC codegen backend is enabled. This allows using the GCC codegen backend "out of the box" with the resulting rustc. It would be nice to get rid of the `libgccjit.so.0` alias (https://rust-lang.zulipchat.com/#narrow/channel/131828-t-compiler/topic/Forcing.20unversioned.20dylib.20dependencies.20on.20Linux/with/534180740), but it's not blocking for this change. You can try running `x build std --set 'rust.codegen-backends=["llvm", "gcc"]'` and then compiling a hello world with `rustc +stage1 -Zcodegen-backend=gcc main.rs`. It is now also possible to build stage2 rustc when the GCC is configured to be the default codegen backend, without any further `LD_LIBRARY_PATH` hacks: `./x build compiler --stage 2 --set 'rust.codegen-backends=["gcc"]'`. After this change, it should be pretty simple to add a dist/rustup step for actually shipping cg_gcc to end users. CC ```@GuillaumeGomez``` r? ```@jieyouxu```

antoyo · 2025-08-17T23:41:28Z

@Kobzol: With this PR, is it still possible to overwrite the libgccjit.so with LD_LIBRARY_PATH or otherwise for development purposes?

Kobzol · 2025-08-18T05:11:15Z

I would assume so. Can you please try? Add gcc to codegen backends in bootstrap.toml, then x build library, and then run the resulting rustc and try to replace libgccjit.

antoyo · 2025-10-24T22:12:07Z

I would assume so. Can you please try? Add gcc to codegen backends in bootstrap.toml, then x build library, and then run the resulting rustc and try to replace libgccjit.

I've just tried setting LD_LIBRARY_PATH before running ./x.py build --stage 2 and it did not seem to affect the build: I set it to upstream libgccjit (so missing the many patches of our fork) and it built the compiler successfully.

I don't remember exactly why I thought this PR could break this. I don't see anything related to the environment variables handling that changed.
Perhaps I had this thought because of this sentence in your first post:

It is now also possible to build stage2 rustc when the GCC is configured to be the default codegen backend, without any further LD_LIBRARY_PATH hacks: ./x build compiler --stage 2 --set 'rust.codegen-backends=["gcc"]'.

I remember some peculiarities about rustup not taking LD_LIBRARY_PATH into account.

Do you have any ideas of what could have caused this? (It could be in a different PR, though.)

And more generally, does setting LD_LIBRARY_PATH work with cg_llvm? How do people test a different LLVM version when hacking on rustc?

Kobzol · 2025-10-25T06:49:05Z

I would first try it manually, just to see if LD_LIBRARY_PATH is applied when loading transitive dylib dependencies of a codegen backend. So build a stage 1 rustc that has cg_gcc as its default codegen backend, and then build something with it, and set the env. varm when invoking rustc.

antoyo · 2025-10-25T14:07:41Z

I would first try it manually, just to see if LD_LIBRARY_PATH is applied when loading transitive dylib dependencies of a codegen backend. So build a stage 1 rustc that has cg_gcc as its default codegen backend, and then build something with it, and set the env. varm when invoking rustc.

If manually running ./stage2/bin/rustc and setting this env var, it does seem to have an effect.
I can get different errors if using different versions:

/opt/gcc/lib/libgccjit.so.0: version `LIBGCCJIT_ABI_36' not found

or:

undefined symbol: gcc_jit_set_global_personality_function_name, version LIBGCCJIT_ABI_37

Kobzol · 2025-10-25T15:41:09Z

Then I guess that bootstrap overrides this env. var. (which I indeed think is happening).

antoyo · 2025-10-25T15:42:35Z

Then I guess that bootstrap overrides this env. var. (which I indeed think is happening).

My impression is that it doesn't use the environment in which it is called. But it could also override this specific var.

Install libgccjit into the compiler's sysroot when the GCC codegen ba…
6e5a6e0
…ckend is enabled

rustbot assigned jieyouxu Aug 13, 2025

rustbot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. T-bootstrap Relevant to the bootstrap subteam: Rust's build system (x.py and src/bootstrap) labels Aug 13, 2025

GuillaumeGomez approved these changes Aug 13, 2025
View reviewed changes

Kobzol added 2 commits August 13, 2025 10:50

Make sure that we won't accidentally ship libgccjit.so in the dist co…
76bc558
…mponent

Remove the libgccjit.so.0 alias and only create the versioned libra…
f40a784
…ry when installing `libgccjit.so`

jieyouxu approved these changes Aug 13, 2025
View reviewed changes

bors added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Aug 13, 2025

This comment was marked as outdated.
Sign in to view

bors 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-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. labels Aug 13, 2025

bors added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. labels Aug 13, 2025

GuillaumeGomez mentioned this pull request Aug 13, 2025
Rollup of 11 pull requests #145363
Closed

This was referenced Aug 13, 2025
Rollup of 11 pull requests #145365
Closed
Rollup of 9 pull requests #145366
Merged

bors merged commit 6f0746d into rust-lang:masterAug 13, 2025
10 checks passed

rustbot added this to the 1.91.0 milestone Aug 13, 2025

Kobzol deleted the codegen-backend-gcc branch August 14, 2025 05:42

GuillaumeGomez mentioned this pull request Aug 14, 2025
Add new --test-codegen-backend bootstrap option #145256
Merged

antoyo mentioned this pull request Oct 26, 2025
Bootstrap should allow overriding a dynamic library #148150
Open

Uh oh!

Uh oh!

Conversation

Kobzol commented Aug 13, 2025• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rustbot commented Aug 13, 2025

Uh oh!

rustbot commented Aug 13, 2025

Uh oh!

jieyouxu left a comment

Choose a reason for hiding this comment

Uh oh!

jieyouxu commented Aug 13, 2025

Uh oh!

bors commented Aug 13, 2025

Uh oh!

Kobzol commented Aug 13, 2025

Uh oh!

This comment was marked as outdated.

This comment was marked as outdated.

This comment was marked as outdated.

This comment was marked as outdated.

This comment was marked as outdated.

This comment was marked as outdated.

bors commented Aug 13, 2025

Uh oh!

Uh oh!

antoyo commented Aug 17, 2025

Uh oh!

Kobzol commented Aug 18, 2025

Uh oh!

antoyo commented Oct 24, 2025

Uh oh!

Kobzol commented Oct 25, 2025

Uh oh!

antoyo commented Oct 25, 2025

Uh oh!

Kobzol commented Oct 25, 2025

Uh oh!

antoyo commented Oct 25, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

Kobzol commented Aug 13, 2025•
edited
Loading