Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Option for whether lib.rs should be rebuilt after build.rs is rerun #1422

Closed
daboross opened this issue Mar 15, 2015 · 3 comments
Closed

Option for whether lib.rs should be rebuilt after build.rs is rerun #1422

daboross opened this issue Mar 15, 2015 · 3 comments

Comments

@daboross
Copy link
Contributor

Currently, whenever build.rs is re-run, both lib.rs and main.rs get rebuilt. It would be useful if there were some option to only rebuild main.rs after build.rs is rerun, not both. This wouldn't want to be the default behaviour, as most build scripts will affect content in lib.rs, but some will only affect main.rs, and an option for that would be nice.

This isn't that urgent, and it may not actually want to be built, but it would probably be useful.

This would probably depend on #1162.
@alexcrichton
Copy link
Member

I believe this is a duplicate of #1162 because Cargo does not currently ever assume that a binary is the location where a native library should be linked if a library is also present. Cargo will instruct the compiler to link the native library to the Rust library and then purposefully not link the native library to the binary generated.

Unfortunately this means that this is not possible for Cargo to know that a build script (or native dependency) is in fact for just a binary, not a library.

@daboross
Copy link
Contributor Author

Would it be possible for this to be a configuration though? I realize it isn't going to be what most people want, but it seems like there would be some projects where build.rs only affects main.rs, not lib.rs (where it is used for something other than native library linking).

@alexcrichton
Copy link
Member

That sounds reasonable to me, I've opened #1430 on the topic.

github-merge-queue bot pushed a commit that referenced this issue Feb 28, 2025
@epage
Bump cc to 1.2.16 to fix x86 windows jobs in rust-lang/rust CI (#…
2622e84 
…15245)

### What does this PR try to resolve?

GitHub Runner Images 20250224.5.0+ ship Windows 11 SDK 10.0.26100+
compared to the previous Windows 11 SDK 10.0.22621, which bumped the
UCRT headers. The new UCRT headers use SSE2 types. However, `cc`
versions <= 1.2.15 emit `/arch:IA32` for `x86` Windows targets for
`clang-cl`, which causes compilation errors since `clang-cl` can't find
SSE2 types without `/arch:SSE2` specified (or defaulted). (Note that
MSVC at the time of writing silently accepts and emits instruments for
code using SSE2 types, as opposed to `clang-cl` hard error-ing).

`cc` 1.2.16 contains a fix for this problem,
rust-lang/cc-rs#1425, to correctly emit
`/arch:SSE2` instead of `/arch:IA32` to enable `clang-cl` to find the
SSE2 types. However, cargo's `cc` currently is still on 1.2.13.

To fix this for rust-lang/rust CI, we need to bump anything that
transitively relies on `cc` and tries to use `clang-cl` on `x86` Windows
targets to compile any C/C++ code that transitively use functions or
types that require SSE2 types, such as `<wchar.h>`.

### How should we test and review this PR?

The fix was initially intended for `rustc_{codegen_ssa,llvm}` `cc`, and
based on testing in rust-lang/rust#137724, I was
able to successfully build `rustc_{codegen_ssa,llvm}` with a forked `cc`
based on 1.2.15 which contains the fix from
rust-lang/cc-rs#1425. Note that in the same PR,
while the compiler build succeeded, the build of cargo itself failed
since it transitively used a `cc` *without* the fix to build
`curl-sys`[^dep-chain], which failed as one might expect (`curl-sys`
tries to build C code that uses `<wchar.h>` which runs into the same
problem). Hence, this PR is opened to bump cargo's `cc` to a `cc`
version containing the fix.

### Additional information

This `x86` Windows CI problem is:

- Discussed in
https://rust-lang.zulipchat.com/#narrow/channel/242791-t-infra/topic/spurious.20.28.3F.29.20i686.20msvc.20errors.
- Tracked by rust-lang/rust#137733.

#### `cc` changelog between 1.2.13 and 1.2.16

<details>
<summary>`cc` changes since 1.2.13 up to and including 1.2.16</summary>

#####
[1.2.16](rust-lang/cc-rs@cc-v1.2.15...cc-v1.2.16)
- 2025-02-28

###### Fixed

- force windows compiler to run in `out_dir` to prevent artifacts in cwd
(#1415)

###### Other

- use `/arch:SSE2` for `x86` target arch (#1425)
- Regenerate windows-sys binding
([#1422](rust-lang/cc-rs#1422))
- Regenerate target info
([#1418](rust-lang/cc-rs#1418))
- Add LIB var when compiling flag_check (#1417)
- Change flag ordering
([#1403](rust-lang/cc-rs#1403))
- Fix archiver detection for musl cross compilation
([#1404](rust-lang/cc-rs#1404))

#####
[1.2.15](rust-lang/cc-rs@cc-v1.2.14...cc-v1.2.15)
- 2025-02-21

###### Other

- Regenerate target info
([#1406](rust-lang/cc-rs#1406))
- Always read from all `CFLAGS`-style flags
([#1401](rust-lang/cc-rs#1401))
- Simplify the error output on failed `Command` invocation
([#1397](rust-lang/cc-rs#1397))

#####
[1.2.14](rust-lang/cc-rs@cc-v1.2.13...cc-v1.2.14)
- 2025-02-14

###### Other

- Regenerate target info
([#1398](rust-lang/cc-rs#1398))
- Add support for setting `-gdwarf-{version}` based on RUSTFLAGS
([#1395](rust-lang/cc-rs#1395))
- Add support for alternative network stack io-sock on QNX 7.1 aarch64
and x86_64 ([#1312](rust-lang/cc-rs#1312))

</details>

[^dep-chain]: I think the dep chain is something like git2 ->
libgit2-sys -> curl -> curl-sys?
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@alexcrichton @daboross