trans: Move rust_try into the compiler by alexcrichton · Pull Request #27176 · rust-lang/rust

alexcrichton · 2015-07-21T02:32:40Z

This commit moves the IR files in the distribution, rust_try.ll,
rust_try_msvc_64.ll, and rust_try_msvc_32.ll into the compiler from the main
distribution. There's a few reasons for this change:

LLVM changes its IR syntax from time to time, so it's very difficult to
have these files build across many LLVM versions simultaneously. We'll likely
want to retain this ability for quite some time into the future.
The implementation of these files is closely tied to the compiler and runtime
itself, so it makes sense to fold it into a location which can do more
platform-specific checks for various implementation details (such as MSVC 32
vs 64-bit).
This removes LLVM as a build-time dependency of the standard library. This may
end up becoming very useful if we move towards building the standard library
with Cargo.

In the immediate future, however, this commit should restore compatibility with
LLVM 3.5 and 3.6.

rust-highfive · 2015-07-21T02:32:45Z

r? @pcwalton

(rust_highfive has picked a reviewer for you, use r? to override)

alexcrichton · 2015-07-21T02:33:06Z

r? @brson

cc @rust-lang/compiler (my knowledge of trans is pretty bad and I may have done egregious things)

Aatch · 2015-07-21T04:09:17Z

"ony" -> "only"

brson · 2015-07-21T17:49:06Z

@bors r+

bors · 2015-07-21T17:49:07Z

📌 Commit 6046fb7 has been approved by brson

brson · 2015-07-21T17:49:19Z

Awesome patch.

alexcrichton · 2015-07-21T22:31:54Z

@bors: r=brson

bors · 2015-07-21T22:31:55Z

📌 Commit 970a3ca has been approved by brson

nagisa · 2015-07-21T22:48:01Z

If you are “defining” the shim, declare::define_internal_rust_fn should be used.

This commit moves the IR files in the distribution, rust_try.ll, rust_try_msvc_64.ll, and rust_try_msvc_32.ll into the compiler from the main distribution. There's a few reasons for this change: * LLVM changes its IR syntax from time to time, so it's very difficult to have these files build across many LLVM versions simultaneously. We'll likely want to retain this ability for quite some time into the future. * The implementation of these files is closely tied to the compiler and runtime itself, so it makes sense to fold it into a location which can do more platform-specific checks for various implementation details (such as MSVC 32 vs 64-bit). * This removes LLVM as a build-time dependency of the standard library. This may end up becoming very useful if we move towards building the standard library with Cargo. In the immediate future, however, this commit should restore compatibility with LLVM 3.5 and 3.6.

alexcrichton · 2015-07-21T23:09:45Z

@bors: r=brson c35b2bd

bors · 2015-07-22T04:16:19Z

⌛ Testing commit c35b2bd with merge 3475c5b...

bors · 2015-07-22T04:19:12Z

💔 Test failed - auto-win-gnu-64-nopt-t

alexcrichton · 2015-07-22T04:22:27Z

@bors: retry

On Tue, Jul 21, 2015 at 9:19 PM, bors notifications@github.com wrote:

[image: 💔] Test failed - auto-win-gnu-64-nopt-t
http://buildbot.rust-lang.org/builders/auto-win-gnu-64-nopt-t/builds/766
—
Reply to this email directly or view it on GitHub
#27176 (comment).

bors · 2015-07-22T09:13:11Z

⌛ Testing commit c35b2bd with merge 25281b1...

This commit moves the IR files in the distribution, rust_try.ll, rust_try_msvc_64.ll, and rust_try_msvc_32.ll into the compiler from the main distribution. There's a few reasons for this change: * LLVM changes its IR syntax from time to time, so it's very difficult to have these files build across many LLVM versions simultaneously. We'll likely want to retain this ability for quite some time into the future. * The implementation of these files is closely tied to the compiler and runtime itself, so it makes sense to fold it into a location which can do more platform-specific checks for various implementation details (such as MSVC 32 vs 64-bit). * This removes LLVM as a build-time dependency of the standard library. This may end up becoming very useful if we move towards building the standard library with Cargo. In the immediate future, however, this commit should restore compatibility with LLVM 3.5 and 3.6.

bors · 2015-07-22T12:49:04Z

☀️ Test successful - auto-linux-32-nopt-t, auto-linux-32-opt, auto-linux-64-nopt-t, auto-linux-64-opt, auto-linux-64-x-android-t, auto-mac-32-opt, auto-mac-64-nopt-t, auto-mac-64-opt, auto-win-gnu-32-nopt-t, auto-win-gnu-32-opt, auto-win-gnu-64-nopt-t, auto-win-gnu-64-opt

vadimcn · 2015-07-22T16:30:00Z

:(
Couldn't we've ran .ll files through cpp (like is done with regular asm) for the same effect?

alexcrichton · 2015-07-22T16:38:01Z

In theory, yes, but that still requires the standard library to have llc available to compile itself, which we need to move away from if we want to build std with Cargo at some point. It also would have been interesting trying to detect the llc version, pass the right variables to the preprocessor, and then piping all that back into llc to create an object.

vadimcn · 2015-07-22T19:03:25Z

In theory, yes, but that still requires the standard library to have llc available to compile itself

I am not sure I understand this point. Didn't rust_trylive in rustrt_native library? Which is effectively a part of the compiler, which already requires LLVM to be around.
How it it different from all other runtime fns in rustrt_native and compiler-rt?

My concern with the new way is that:

Emitting IR via LLVM API takes a lot more work and is way less readable because of all the noise added by API usage.
One cannot override rust_try for a custom target, (like you we able to do for *-pc-windows-msvc) without changing the compiler.

vadimcn · 2015-07-22T19:04:52Z

(sorry for jumping in late - I was not aware of this change until bors told me that my PR collided with this one)

alexcrichton · 2015-07-22T21:55:28Z

I am not sure I understand this point. Didn't rust_trylive in rustrt_native library? Which is effectively a part of the compiler, which already requires LLVM to be around.
How it it different from all other runtime fns in rustrt_native and compiler-rt?

Ah by this I just mean that the tool llc is no longer needed to build the standard library.

The other shims and such in rustrt_native can be built with a normal C compiler and I believe that all the shims in compiler-rt can also be built with a normal C compiler. If you're in theory cross compiling the standard library to a new architecture it's much more likely you've got the right C compiler installed vs llc.

Emitting IR via LLVM API takes a lot more work and is way less readable because of all the noise added by API usage.

I don't disagree with this!

One cannot override rust_try for a custom target, (like you we able to do for *-pc-windows-msvc) without changing the compiler.

I'm not quite sure I follow this. Is this basically saying that it's easier to edit IR than it is to edit the compiler?

sorry for jumping in late - I was not aware of this change until bors told me that my PR collided with this one

No worries!

vadimcn · 2015-07-23T00:01:14Z

I'm not quite sure I follow this. Is this basically saying that it's easier to edit IR than it is to edit the compiler?

I meant that as a library function in rustrt_native, it could have been overridden. Not so much with it as compiler intrinsic.
In principle, rust_try could have been implemented in C++, assembly, or even in C, if the compiler supports suitable extensions. It's just hard to do so portably.

alexcrichton · 2015-07-23T01:01:13Z

How do you override something in a native library? You mean avoiding #[cfg(stage0)]?

The ABI of unwinding is very closely tied to the compiler itself, so even though it could be implemented in various locations it kinda makes sense to me to implement this as an intrinsic so the compiler has complete control over all ABI possibilities.

vadimcn · 2015-07-23T06:24:56Z

How do you override something in a native library?

I meant that when building rustrt_native for a new target, you could write a custom rust_try.

alexcrichton · 2015-07-23T15:08:23Z

That's true, but do you know if that's done often? From what I've seen the rust_try.ll file basically never changes.

eddyb · 2015-07-24T10:59:55Z

Was it known that this would break unwinding entirely on -Z no-landing-pads builds?
This means check-stage1-cfail fails miserably, it need at least the rust_try landing pad to run.

Individual stage1rustc invocations will abort with fatal runtime error: Could not unwind stack, error = 5 if any errors are found.
And then the compiletest driver will panic! during handling of the unexpected SIGILL in rustc, triggering another fatal runtime error: Could not unwind stack, error = 5 abort, hiding the original rustc one.

eddyb · 2015-07-24T11:14:32Z

Actually, scratch that, it's because the snapshot compiler doesn't have the try intrinsic so this PR uses a noop, breaking unwinding for the stage1 compiler and tests until the next snapshot.

alexcrichton · 2015-07-24T16:13:28Z

@eddyb yes this was known.

On the functionality side, morestack is gone (rust-lang/rust#27338), and rustrt_native is gone (rust-lang/rust#27176). On the implementation side, connect has been renamed to join (rust-lang/rust#26957).

rust-lang/rust#27176 "removes LLVM as a build-time dependency of the standard library. This may end up becoming very useful if we move towards building the standard library with Cargo."

rust-highfive assigned pcwalton Jul 21, 2015

rust-highfive assigned brson and unassigned pcwalton Jul 21, 2015

Aatch reviewed Jul 21, 2015
View reviewed changes

Comment threadsrc/librustc_trans/trans/intrinsic.rs Outdated
AatchJul 21, 2015
Copy link
Copy Markdown
Contributor
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
"ony" -> "only"

alexcrichton force-pushed the fix-stock-llvm branch 2 times, most recently from 51016bd to 6046fb7Compare July 21, 2015 07:47

alexcrichton force-pushed the fix-stock-llvm branch 4 times, most recently from 825860c to 970a3caCompare July 21, 2015 22:31

nagisa reviewed Jul 21, 2015
View reviewed changes

alexcrichton force-pushed the fix-stock-llvm branch from 970a3ca to c35b2bdCompare July 21, 2015 23:09

bors merged commit c35b2bd into rust-lang:masterJul 22, 2015

This was referenced Jul 22, 2015
refactor svh #26629
Closed
SIMD groundwork part 1 #27169
Merged
Implement Win64 eh_personality natively. #27210
Merged
Refactor how closures are represented to expose the types of the upvars #27087
Merged

alexcrichton deleted the fix-stock-llvm branch July 22, 2015 15:48

arielb1 mentioned this pull request Sep 15, 2015
1.3 relnotes #28408
Merged

Uh oh!

Uh oh!

Conversation

alexcrichton commented Jul 21, 2015

Uh oh!

rust-highfive commented Jul 21, 2015

Uh oh!

alexcrichton commented Jul 21, 2015

Uh oh!

AatchJul 21, 2015

Choose a reason for hiding this comment

Uh oh!

brson commented Jul 21, 2015

Uh oh!

bors commented Jul 21, 2015

Uh oh!

brson commented Jul 21, 2015

Uh oh!

alexcrichton commented Jul 21, 2015

Uh oh!

bors commented Jul 21, 2015

Uh oh!

nagisaJul 21, 2015

Choose a reason for hiding this comment

Uh oh!

alexcrichton commented Jul 21, 2015

Uh oh!

bors commented Jul 22, 2015

Uh oh!

bors commented Jul 22, 2015

Uh oh!

alexcrichton commented Jul 22, 2015

Uh oh!

bors commented Jul 22, 2015

Uh oh!

bors commented Jul 22, 2015

Uh oh!

vadimcn commented Jul 22, 2015

Uh oh!

alexcrichton commented Jul 22, 2015

Uh oh!

vadimcn commented Jul 22, 2015

Uh oh!

vadimcn commented Jul 22, 2015

Uh oh!

alexcrichton commented Jul 22, 2015

Uh oh!

vadimcn commented Jul 23, 2015

Uh oh!

alexcrichton commented Jul 23, 2015

Uh oh!

vadimcn commented Jul 23, 2015

Uh oh!

alexcrichton commented Jul 23, 2015

Uh oh!

eddyb commented Jul 24, 2015

Uh oh!

eddyb commented Jul 24, 2015

Uh oh!

alexcrichton commented Jul 24, 2015

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

9 participants