Upgrade to rustc 1.3.0-dev (fddfd089b 2015-07-10)

[SimonSapin]

Depends on mozilla/webdriver-rust#12.

Fixes #6020.

[hoppipolla-critic-bot]

Critic review: https://critic.hoppipolla.co.uk/r/5516

This is an external review system which you may optionally use for the code review of your pull request.

In order to help critic track your changes, please do not make in-place history rewrites (e.g. via git rebase -i or git commit --amend) when updating this pull request.

[pcwalton]

Looks good to me, once upstream is merged feel free to land.

[SimonSapin]

@bors-servo r=pcwalton

[bors-servo]

📌 Commit b2a87db has been approved by pcwalton

[bors-servo]

⌛ Testing commit b2a87db with merge 35ea678...

[bors-servo]

💔 Test failed - android

[SimonSapin]

@bors-servo r=pcwalton

(Re-ran ./mach update-cargo -p webdriver and ./mach update-cargo -p hyper now that mozilla/webdriver-rust#12 is merged.)

[bors-servo]

📌 Commit bbed054 has been approved by pcwalton

[bors-servo]

⌛ Testing commit bbed054 with merge a02e7d8...

[bors-servo]

💔 Test failed - mac2

[SimonSapin]

ld: warning: directory not found for option '-L/Users/servo/buildbot/slave/mac2/build/ports/cef/lib/x86_64-apple-darwin'
ld: warning: directory not found for option '-L/Users/servo/.rust/lib/x86_64-apple-darwin'
duplicate symbol _TaskBasicInfoVirtualSize in:
    /var/folders/sd/sks3rs2s3pb9bdrg9ldf9d5w0000gp/T/rustc.pqY5S1Luk7Wa/libtask_info-6ccbc3024e2756cc.rlib(r-task_info-task_info.o)
    /var/folders/sd/sks3rs2s3pb9bdrg9ldf9d5w0000gp/T/rustc.pqY5S1Luk7Wa/libtask_info-6ccbc3024e2756cc.rlib(r-1-task_info-task_info.o)
duplicate symbol _TaskBasicInfoResidentSize in:
    /var/folders/sd/sks3rs2s3pb9bdrg9ldf9d5w0000gp/T/rustc.pqY5S1Luk7Wa/libtask_info-6ccbc3024e2756cc.rlib(r-task_info-task_info.o)
    /var/folders/sd/sks3rs2s3pb9bdrg9ldf9d5w0000gp/T/rustc.pqY5S1Luk7Wa/libtask_info-6ccbc3024e2756cc.rlib(r-1-task_info-task_info.o)
duplicate symbol __Z8SkDebugfPKcz in:
    /var/folders/sd/sks3rs2s3pb9bdrg9ldf9d5w0000gp/T/rustc.pqY5S1Luk7Wa/libazure-f792f70de9cc89e8.rlib(r-skia-SkDebug_stdio.o)
    /var/folders/sd/sks3rs2s3pb9bdrg9ldf9d5w0000gp/T/rustc.pqY5S1Luk7Wa/libazure-f792f70de9cc89e8.rlib(r-1-skia-SkDebug_stdio.o)
ld: 3 duplicate symbols for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)

[pcwalton]

<pcwalton> SimonSapin: sometimes the Skia one is the result of Skia and Azure not being compiled with the same -D preprocessor flags
<pcwalton> SimonSapin: I've seen that sort of thing happen with SK_DEBUG being set to 1 in skia-sys and 0 in rust-azure, or vice-versa

[pcwalton]

Maybe we have multiple versions of packages. Check the Cargo.lock?

[SimonSapin]

The only crates duplicated in Cargo.lock are android_glue and bitflags. I couldn’t reproduce locally on a mac. Let’s see if this is intermittent.

@bors-servo retry

[bors-servo]

⌛ Testing commit bbed054 with merge 21433eb...

[bors-servo]

💔 Test failed - mac2

[SimonSapin]

Ah, I missed that this is in build-cef. I can reproduce locally.

[larsbergstrom]

Hrm. This is super confusing because when you do a build-cef, the only new things compiled are the servo crate (with a different set of features enabled) and the embedding crate.

[larsbergstrom]

Happens even on a clean build of CEF - https://pastebin.mozilla.org/8839475

[larsbergstrom]

@alexcrichton Have you ever seen something like this? We're getting duplicate symbols reported during link from a couple of crates (Skia and task-info) when we build a shared library (the CEF embedding stuff) but not when we build a binary (the final servo binary) - despite the fact that these crates are not rebuilt and we share target directories.

If I nm -g libtaskinfo, I do get what appears to be two copies of the symbol:

./deps/libtask_info-4f894c0f5635aa99.rlib
                 U _TaskBasicInfoVirtualSize
0000000000000000 T _TaskBasicInfoVirtualSize
0000000000000000 T _TaskBasicInfoVirtualSize

Working on it, but this is currently blocking our rustup, so if you have any ideas, they'd be appreciated :-)

The task_info sources are much easier to dig into than Skia: https://github.com/servo/servo/tree/rustup_2015-07-10/support/rust-task_info

[alexcrichton]

I suspect this is caused by rust-lang/rust#26869 (the -force_load flag for OSX archives), although it looks like the task_info library may be getting linked twice? For example these two object filenames show up:

r-task_info-task_info.o
r-1-task_info-task_info.o

I believe that this is linking the library once and this is linking it again, so if you remove the #[link] annotation you should be good to go.

Perhaps the skia bug is also somewhat related?

[larsbergstrom]

@alexcrichton That fixed it for the task info - thanks a ton!

I suspect it's a little more complicated for skia, as the static lib is pulled in via #[link] lines in multiple places in rust-azure plus one in the actual skia repo, but I should (hopefully) be able to track down what we should really be doing. There are hacks on top of hacks dating back to partial attempts to stand something up on Android, where linking/loading was... iffy... for a while.

[alexcrichton]

The pattern that I would suggest is the similar foo and foo-sys crates that crates.io has. For example if you have skia bindings I'd put them in the skia crate but the FFI definitions are in the skia-sys crate. Additionally, the #[link] (or linage via a build script) is only present for the skia-sys crate, so skia-sys is the one source of truth for linking the native library. That way the rust-azure crate could depend on skia-sys to be sure to link the library but doesn't necessarily need to pull in all the bindings of skia

[larsbergstrom]

@alexcrichton Yeah, I think most of our hacks are also to preserve link order, which is really finicky here. e.g., we need to link in fontconfig after skia or the skia bindings to FC functions will be unresolved.

[metajack]

@alexcrichton I think that's what we tried to achieve, but some of this is pre-Cargo legacy. We'll clean it up after this round of Rust upgrades and see if we run into further issues.

[bors-servo]

☔ The latest upstream changes (presumably #6596) made this pull request unmergeable. Please resolve the merge conflicts.

[SimonSapin]

I’ll rebase.

[SimonSapin]

I force-pushed my rebase of bbed054 without thinking of fetching first, so I overwrote any change made since. Sorry! @larsbergstrom I think you had some, could you push your local branch again? Either force-push and I can rebase again, or push to another branch and I can cherry-pick your changes.

Also, it looks like rust-layers now need to be updated for changes in Skia: https://travis-ci.org/servo/rust-layers/jobs/70484266

[mrobinson]

I've added the commit that gets rust-layers up to date, as well as reworking the code to work properly with it.

[pcwalton]

LGTM.

[larsbergstrom]

@bors-servo: r+

[bors-servo]

📌 Commit efdf2cd has been approved by larsbergstrom

[bors-servo]

⌛ Testing commit efdf2cd with merge 9af229b...

[bors-servo]

☀️ Test successful - android, gonk, linux1, linux2, linux3, mac1, mac2, mac3