Use 'cargo check' to build the sysroot and target crate by Aaron1011 · Pull Request #1136 · rust-lang/miri

Aaron1011 · 2020-01-01T09:17:17Z

I'm using my original approach from PR #1048. Ideally, we would
distinguish between build-deps/dependencies/'final crate' via a
different approach (e.g. the target directory). However, I
haven't been able to get that to work just yet.

However, everything should be working with the approach I'm using. At a
minimum, we can use this PR to verify that everything works as expected
when we don't actually produce native build outputs.

Aaron1011 · 2020-01-01T09:19:08Z

src/bin/cargo-miri.rs

        Command::new(val)
    } else {
-        Command::new("xargo")
+        Command::new("xargo-check")


It really is that easy 😄

Well, longer-term we'd ideally also stop creating a Cargo.toml. ;)

RalfJung · 2020-01-02T16:30:28Z

I remain pretty skeptical of the enormous hacks this introduces, and would prefer something relying on --target, as laid out in #1057 (comment).

src/bin/cargo-miri.rs

Aaron1011 · 2020-01-09T11:19:56Z

@RalfJung: I've modified this PR to distinguish between crate types using only the command-line arguments passed by Cargo. Using the target directory turned out to be unnecessary - --emit=link, --crate-type bin, and --test is sufficient.

RalfJung · 2020-01-15T19:36:23Z

What about --target? Cargo can already distinguish these things for us, that sounds much less fragile.

Aaron1011 · 2020-01-15T21:52:15Z

@RalfJung: We need to check for --crate-type bin and --test regardless of whether or not we rely on --target - both normal dependencies and the final crate (binary or test artifact) go in the same directory.

I think using --target is actually much more fragile. We need to keep track of (or determine) the root directory as well as the current target triple, so that we can avoid getting tricked by something like /some/path/with/target/x86_64-unknown-linux/gnu/then/more (e.g. a root directory path that shares names with the subdirs that Cargo generates).

I think checking for the presence of one flag --emit=link is much simpler. If it ever breaks, we can just change the implementation of is_build_dep.

RalfJung · 2020-01-17T08:48:21Z

Looks like I'll need to look closer into this and do some experiments myself to fully appreciate what is happening here. That might take a few weeks until I find the time, sorry.

RalfJung · 2020-02-24T13:22:44Z

I think using --target is actually much more fragile. We need to keep track of (or determine) the root directory as well as the current target triple, so that we can avoid getting tricked by something like /some/path/with/target/x86_64-unknown-linux/gnu/then/more (e.g. a root directory path that shares names with the subdirs that Cargo generates).

I don't think we need to look at the root directory at all, we can just call rustc to ask it for the default target.

But you are right that this is not really much less hacky than your variant, if at all. So since this works, we can just use it. I did some slight refactoring like moving the test methods down inside inside_cargo_rustc to make that code more local; this should be good to go. Sorry it took me so long!

@bors r+

bors · 2020-02-24T13:22:46Z

📌 Commit 7bcf51a has been approved by RalfJung

Fixes rust-lang#1057 Since we are no longer using "cargo rustc", we now use the rustc arguments passed by Cargo to determine whether we are building a build dependency, normal dependency, or "target" (final binary or test) crate.

RalfJung · 2020-02-24T13:43:48Z

@bors r+

bors · 2020-02-24T13:43:49Z

📌 Commit 47f2b12 has been approved by RalfJung

Use 'cargo check' to build the sysroot and target crate Fixes #1057 I'm using my original approach from PR #1048. Ideally, we would distinguish between build-deps/dependencies/'final crate' via a different approach (e.g. the target directory). However, I haven't been able to get that to work just yet. However, everything should be working with the approach I'm using. At a minimum, we can use this PR to verify that everything works as expected when we don't actually produce native build outputs.

bors · 2020-02-24T13:43:56Z

⌛ Testing commit 47f2b12 with merge fc08cce...

bors · 2020-02-24T13:44:55Z

💔 Test failed - status-appveyor

RalfJung · 2020-02-24T13:45:47Z

@bors r+

bors · 2020-02-24T13:45:48Z

📌 Commit faf7bf5 has been approved by RalfJung

bors · 2020-02-24T13:45:55Z

⌛ Testing commit faf7bf5 with merge 36305e3...

bors · 2020-02-24T14:09:18Z

☀️ Test successful - checks-travis, status-appveyor
Approved by: RalfJung
Pushing 36305e3 to master...

@Aaron1011

…c-morse no more codegen for miri_start_panic With rust-lang/miri#1136 landed, we don't generate code any more for crates that will be run by Miri. So the LLVM backend does not have to implement the `miri_start_panic` intrinsic any more. Cc @Aaron1011

tesuji · 2020-02-25T16:01:25Z

src/bin/cargo-miri.rs

+}
+
+fn cargo() -> Command {
+    if let Ok(val) = std::env::var("CARGO") {


I would like to use env;:var_os when ever it is possible.

That probably makes sense, if we don't need it to really be UTF-8... do you want to submit a PR?

Aaron1011 force-pushed the feature/check-libstd branch from b07f3ac to 29b2fad Compare January 1, 2020 09:18

Aaron1011 commented Jan 1, 2020

View reviewed changes

RalfJung reviewed Jan 2, 2020

View reviewed changes

src/bin/cargo-miri.rs Outdated Show resolved Hide resolved

Aaron1011 force-pushed the feature/check-libstd branch 2 times, most recently from df667ff to d980c36 Compare January 9, 2020 11:17

RalfJung mentioned this pull request Feb 13, 2020

Unable to run individual binaries with cargo-miri when multiple binaries exist #1173

Closed

RalfJung force-pushed the feature/check-libstd branch from d980c36 to 7bcf51a Compare February 24, 2020 13:20

This comment has been minimized.

Sign in to view

Aaron1011 and others added 2 commits February 24, 2020 14:41

Use 'cargo check' to build the sysroot and target crate

e530829

Fixes rust-lang#1057 Since we are no longer using "cargo rustc", we now use the rustc arguments passed by Cargo to determine whether we are building a build dependency, normal dependency, or "target" (final binary or test) crate.

refactor cargo-miri a bit

443163f

RalfJung force-pushed the feature/check-libstd branch from 7bcf51a to 443163f Compare February 24, 2020 13:41

fix Cargo.toml

47f2b12

update lockfile

faf7bf5

bors merged commit 36305e3 into rust-lang:master Feb 24, 2020

RalfJung mentioned this pull request Feb 24, 2020

no more codegen for miri_start_panic rust-lang/rust#69437

Merged

tesuji reviewed Feb 25, 2020

View reviewed changes

Conversation

Aaron1011 commented Jan 1, 2020

Uh oh!

Aaron1011 Jan 1, 2020

Choose a reason for hiding this comment

Uh oh!

RalfJung Jan 2, 2020

Choose a reason for hiding this comment

Uh oh!

RalfJung commented Jan 2, 2020

Uh oh!

Uh oh!

Aaron1011 commented Jan 9, 2020

Uh oh!

RalfJung commented Jan 15, 2020

Uh oh!

Aaron1011 commented Jan 15, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

RalfJung commented Jan 17, 2020

Uh oh!

RalfJung commented Feb 24, 2020

Uh oh!

bors commented Feb 24, 2020

Uh oh!

This comment has been minimized.

RalfJung commented Feb 24, 2020

Uh oh!

bors commented Feb 24, 2020

Uh oh!

bors commented Feb 24, 2020

Uh oh!

bors commented Feb 24, 2020

Uh oh!

RalfJung commented Feb 24, 2020

Uh oh!

bors commented Feb 24, 2020

Uh oh!

bors commented Feb 24, 2020

Uh oh!

bors commented Feb 24, 2020

Uh oh!

tesuji Feb 25, 2020

Choose a reason for hiding this comment

Uh oh!

RalfJung Feb 25, 2020

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Aaron1011 commented Jan 15, 2020 •

edited

Loading