Skip to contentSkip to conversation

rust: fix rust-analyzer configuration for generated files

PATCH Applied Tamir Duberstein 3 participants 10 messages Feb 10, 2025 – Mar 11, 2025

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

v2 cover rust: fix rust-analyzer configuration for generated files

The individual patches should be descriptive on their own. They are included in a single series because the second patch uses a function introduced in the first.

I've confirmed this allows me to navigate to symbols defined in generated files as well as to the generated files themselves. I am using an out-of-source build.

---
Changes in v2:
- Add missing core dependency on uapi. Fixes navigation in
rust/kernel/ioctl.rs.
- Link to v1: https://lore.kernel.org/r/20250210-rust-analyzer-bindings-include-v1-0-6cbf420e95b6@gmail.com

---
Tamir Duberstein (2):
scripts: generate_rust_analyzer.py: add missing include_dirs
scripts: generate_rust_analyzer.py: add uapi crate

 scripts/generate_rust_analyzer.py | 41 ++++++++++++++++++++-------------------
 1 file changed, 21 insertions(+), 20 deletions(-)

---
base-commit: a64dcfb451e254085a7daee5fe51bf22959d52d3
change-id: 20250210-rust-analyzer-bindings-include-531e9dacec8d

Best regards,

Signature · 1 line
Tamir Duberstein <[email protected]>
#1 lore raw

PATCH v2 1/2 scripts: generate_rust_analyzer.py: add missing include_dirs

Commit 8c4555ccc55c ("scripts: add generate_rust_analyzer.py") specified OBJTREE for the bindings crate, and source.include_dirs for the kernel crate, likely in an attempt to support out-of-source builds for those crates where the generated files reside in objtree rather than srctree. This was insufficient because both bits of configuration are required for each crate; the result is that rust-analyzer is unable to resolve generated files for either crate in an out-of-source build.

Add the missing bits to improve the developer experience.

Fixes 8c4555ccc55c ("scripts: add `generate_rust_analyzer.py`")
---
 scripts/generate_rust_analyzer.py | 40 +++++++++++++++++++--------------------
 1 file changed, 20 insertions(+), 20 deletions(-)
scripts/generate_rust_analyzer.py +20 -20
index aa8ea1a4dbe5..1f573d19cd99 100755
@@ -85,27 +85,27 @@ def generate_crates(srctree, objtree, sysroot_src, external_src, cfgs):
         ["core", "compiler_builtins"],
     )
 
-    append_crate(
-        "bindings",
-        srctree / "rust"/ "bindings" / "lib.rs",
-        ["core"],
-        cfg=cfg,
-    )
-    crates[-1]["env"]["OBJTREE"] = str(objtree.resolve(True))
+    def append_crate_with_generated(
+        display_name,
+        deps,
+    ):
+        append_crate(
+            display_name,
+            srctree / "rust"/ display_name / "lib.rs",
+            deps,
+            cfg=cfg,
+        )
+        crates[-1]["env"]["OBJTREE"] = str(objtree.resolve(True))
+        crates[-1]["source"] = {
+            "include_dirs": [
+                str(srctree / "rust" / display_name),
+                str(objtree / "rust")
+            ],
+            "exclude_dirs": [],
+        }
 
-    append_crate(
-        "kernel",
-        srctree / "rust" / "kernel" / "lib.rs",
-        ["core", "macros", "build_error", "bindings"],
-        cfg=cfg,
-    )
-    crates[-1]["source"] = {
-        "include_dirs": [
-            str(srctree / "rust" / "kernel"),
-            str(objtree / "rust")
-        ],
-        "exclude_dirs": [],
-    }
+    append_crate_with_generated("bindings", ["core"])
+    append_crate_with_generated("kernel", ["core", "macros", "build_error", "bindings"])
 
     def is_root_crate(build_file, target):
         try:
Signature · 1 line
2.48.1
#2 lore raw

Re: [PATCH v2 1/2] scripts: generate_rust_analyzer.py: add missing include_dirs

Quoted 8 lines · Tamir Duberstein

Commit 8c4555ccc55c ("scripts: add generate_rust_analyzer.py") specified OBJTREE for the bindings crate, and source.include_dirs for the kernel crate, likely in an attempt to support out-of-source builds for those crates where the generated files reside in objtree rather than srctree. This was insufficient because both bits of configuration are required for each crate; the result is that rust-analyzer is unable to resolve generated files for either crate in an out-of-source build.

Originally we were not using OBJTREE in the kernel crate, but we did pass it anyway, so conceptually it could have been there. So I am not sure if it counts as a fix for that commit, but it shouldn't hurt even if backported.

Regarding include_dirs, it started in kernel before being in mainline because we included the bindings there (i.e. there was not bindings crate), but it should have been probably moved when it was split. Nowadays, I guess we still need it for generated_arch_static_branch_asm.rs, or is it something else that needs it? I assume it shouldn't hurt, in any case, so it looks OK.

Cheers,
Miguel

#3 lore raw

Re: [PATCH v2 1/2] scripts: generate_rust_analyzer.py: add missing include_dirs

Quoted 11 lines · Miguel Ojeda
Quoted 8 lines · Tamir Duberstein

Commit 8c4555ccc55c ("scripts: add generate_rust_analyzer.py") specified OBJTREE for the bindings crate, and source.include_dirs for the kernel crate, likely in an attempt to support out-of-source builds for those crates where the generated files reside in objtree rather than srctree. This was insufficient because both bits of configuration are required for each crate; the result is that rust-analyzer is unable to resolve generated files for either crate in an out-of-source build.

Quoted 4 lines

Originally we were not using OBJTREE in the kernel crate, but we did pass it anyway, so conceptually it could have been there. So I am not sure if it counts as a fix for that commit, but it shouldn't hurt even if backported.

Ah, I see. The reference to OBJTREE in the kernel crate was introduced in commit 169484ab6677 ("rust: add arch_static_branch"). I'm not able to build at commit 8c4555ccc55c ("scripts: add generate_rust_analyzer.py") but I would expect that rust-analyzer didn't work properly there for the bindings crate for out of tree builds because of the missing include_dirs.

Quoted 4 lines

Regarding include_dirs, it started in kernel before being in mainline because we included the bindings there (i.e. there was not bindings crate),

I don't follow this; there's a bindings crate at 8c4555ccc55c ("scripts: add generate_rust_analyzer.py") - at least as far as RA is concerned.

but it should have been probably moved when it was

Quoted 3 lines

split. Nowadays, I guess we still need it for generated_arch_static_branch_asm.rs, or is it something else that needs it? I assume it shouldn't hurt, in any case, so it looks OK.

Yes, without it generated_arch_static_branch_asm.rs is not found by RA.

#4 lore raw

Re: [PATCH v2 1/2] scripts: generate_rust_analyzer.py: add missing include_dirs

Quoted 4 lines · Tamir Duberstein

I don't follow this; there's a bindings crate at 8c4555ccc55c ("scripts: add generate_rust_analyzer.py") - at least as far as RA is concerned.

By "before being in mainline" I mean before Rust was in Linus' tree, i.e. a long time ago in the old rust branch.

Back then, there was a period of time when there was no bindings crate. At some point it was split from the kernel crate, which did have the include_dirs for that reason, and at that moment the include_dirs should have been moved too, but it wasn't.

So we ended up with include_dirs in the kernel crate for no reason (because AFAICS it was not needed until we got generated_arch_static_branch_asm.rs, but I haven't actually tested that in RA) and missing it in the bindings one where it was actually needed.

(I was trying to understand how we ended up here just to double-check why we had things split like that)

Quoted 1 line

Yes, without it generated_arch_static_branch_asm.rs is not found by RA.

Thanks for confirming!

Cheers,
Miguel

#5 lore raw

Re: [PATCH v2 1/2] scripts: generate_rust_analyzer.py: add missing include_dirs

\On Tue, Mar 11, 2025 at 7:01 PM Miguel Ojeda

Quoted 15 lines · <[email protected]> wrote:
Quoted 4 lines · Tamir Duberstein

I don't follow this; there's a bindings crate at 8c4555ccc55c ("scripts: add generate_rust_analyzer.py") - at least as far as RA is concerned.

By "before being in mainline" I mean before Rust was in Linus' tree, i.e. a long time ago in the old rust branch.

Back then, there was a period of time when there was no bindings crate. At some point it was split from the kernel crate, which did have the include_dirs for that reason, and at that moment the include_dirs should have been moved too, but it wasn't.

Quoted 8 lines

So we ended up with include_dirs in the kernel crate for no reason (because AFAICS it was not needed until we got generated_arch_static_branch_asm.rs, but I haven't actually tested that in RA) and missing it in the bindings one where it was actually needed.

(I was trying to understand how we ended up here just to double-check why we had things split like that)

Ack, thanks for explaining.

Quoted 3 lines
Quoted 1 line

Yes, without it generated_arch_static_branch_asm.rs is not found by RA.

Thanks for confirming!

You're welcome!

#6 lore raw

PATCH v2 2/2 scripts: generate_rust_analyzer.py: add uapi crate

In reply to Tamir Duberstein

Commit 4e1746656839 ("rust: uapi: Add UAPI crate") did not update rust-analyzer to include the new crate.

Add the missing definition to improve the developer experience.

Fixes 4e1746656839 ("rust: uapi: Add UAPI crate")
---
 scripts/generate_rust_analyzer.py | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
scripts/generate_rust_analyzer.py +2 -1
index 1f573d19cd99..d3fe091a55dd 100755
@@ -105,7 +105,8 @@ def generate_crates(srctree, objtree, sysroot_src, external_src, cfgs):
         }
 
     append_crate_with_generated("bindings", ["core"])
-    append_crate_with_generated("kernel", ["core", "macros", "build_error", "bindings"])
+    append_crate_with_generated("uapi", ["core"])
+    append_crate_with_generated("kernel", ["core", "macros", "build_error", "bindings", "uapi"])
 
     def is_root_crate(build_file, target):
         try:
Signature · 1 line
2.48.1
reviewer Tested-by on the v2 series
#7 lore raw
In reply to Tamir Duberstein
Quoted 9 lines · Tamir Duberstein

The individual patches should be descriptive on their own. They are included in a single series because the second patch uses a function introduced in the first.

I've confirmed this allows me to navigate to symbols defined in generated files as well as to the generated files themselves. I am using an out-of-source build.

This is amazing, I have been missing this for so long 😆 With this patch I get auto completion of symbols in bindings and uapi, without it, no completions. I build out of tree.

For the whole series:

Best regards,
Andreas Hindborg

#8 lore raw
Quoted 7 lines · Andreas Hindborg
Quoted 4 lines · Tamir Duberstein

The individual patches should be descriptive on their own. They are included in a single series because the second patch uses a function introduced in the first.

Quoted 5 lines
Quoted 5 lines

I've confirmed this allows me to navigate to symbols defined in generated files as well as to the generated files themselves. I am using an out-of-source build.

Hi folks, gentle ping here.

Quoted 7 lines

This is amazing, I have been missing this for so long 😆 With this patch I get auto completion of symbols in bindings and uapi, without it, no completions. I build out of tree.

For the whole series:

Thanks for testing!

Applied
#9 lore raw
In reply to Tamir Duberstein
Quoted 10 lines · Tamir Duberstein

The individual patches should be descriptive on their own. They are included in a single series because the second patch uses a function introduced in the first.

I've confirmed this allows me to navigate to symbols defined in generated files as well as to the generated files themselves. I am using an out-of-source build.

Applied to rust-fixes -- thanks everyone!

[ Originally we were not using OBJTREE in the kernel crate, but
we did pass the variable anyway, so conceptually it could have been there since then.

Regarding include_dirs, it started in kernel before being in mainline because we included the bindings directly there (i.e. there was no bindings crate). However, when that crate got created, we moved the OBJTREE there but not the include_dirs. Nowadays, though, we happen to need the include_dirs also in the kernel crate for generated_arch_static_branch_asm.rs which was not there back then -- Tamir confirms it is indeed required for that reason. - Miguel ]

[ Slightly reworded title. - Miguel ]

[ Slightly reworded title. - Miguel ]

Cheers,
Miguel