Stable regression in nightly

[frankmcsherry]

Type inference broke, or something. Hard to see why, as the file (broadcast.rs) is probably one of the simpler cases.

Easiest repro to is clone the repo (it is also up on crates, though as an earlier version)

% git clone https://github.com/frankmcsherry/timely-dataflow
Cloning into 'timely-dataflow'...
remote: Counting objects: 1959, done.
remote: Compressing objects: 100% (70/70), done.
remote: Total 1959 (delta 32), reused 0 (delta 0), pack-reused 1889
Receiving objects: 100% (1959/1959), 596.67 KiB | 741.00 KiB/s, done.
Resolving deltas: 100% (1393/1393), done.
Checking connectivity... done.
% cd timely-dataflow 
% multirust override stable
multirust: using existing install for 'stable'
multirust: override toolchain for '/Users/mcsherry/timely-dataflow' set to 'stable'
% cargo build
    Updating git repository `https://github.com/frankmcsherry/abomonation`
    Updating registry `https://github.com/rust-lang/crates.io-index`
   Compiling byteorder v0.3.11
   Compiling libc v0.1.8
   Compiling abomonation v0.1.0 (https://github.com/frankmcsherry/abomonation#6bdf392c)
   Compiling log v0.3.1
   Compiling getopts v0.2.11
   Compiling timely v0.0.7 (file:///Users/mcsherry/timely-dataflow)
% multirust override nightly
multirust: using existing install for 'nightly'
multirust: override toolchain for '/Users/mcsherry/timely-dataflow' set to 'nightly'
% cargo build               
   Compiling abomonation v0.1.0 (https://github.com/frankmcsherry/abomonation#6bdf392c)
   Compiling libc v0.1.8
   Compiling byteorder v0.3.11
   Compiling log v0.3.1
   Compiling getopts v0.2.11
   Compiling timely v0.0.7 (file:///Users/mcsherry/timely-dataflow)
src/progress/broadcast.rs:31:51: 31:70 error: the type of this value must be known in this context
src/progress/broadcast.rs:31                 while let Some((update, delta)) = recv_messages.pop() {
                                                                               ^~~~~~~~~~~~~~~~~~~
note: in expansion of while let expansion
src/progress/broadcast.rs:31:17: 33:55 note: expansion site
note: in expansion of while let expansion
src/progress/broadcast.rs:30:13: 37:14 note: expansion site
src/progress/broadcast.rs:34:51: 34:70 error: the type of this value must be known in this context
src/progress/broadcast.rs:34                 while let Some((update, delta)) = recv_internal.pop() {
                                                                               ^~~~~~~~~~~~~~~~~~~
note: in expansion of while let expansion
src/progress/broadcast.rs:34:17: 36:55 note: expansion site
note: in expansion of while let expansion
src/progress/broadcast.rs:30:13: 37:14 note: expansion site
error: aborting due to 2 previous errors
Could not compile `timely`.

To learn more, run the command again with --verbose.
% cargo --version
cargo 0.4.0-nightly (15b497b 2015-07-08) (built 2015-07-08)
% rustc --version
rustc 1.3.0-nightly (16f64c388 2015-07-09)

[alexcrichton]

This is likely due to the new implementation of Extend on collections for &T. Do you have some source code that we could poke around with? This may just entail an annotation on a call to collect.

[frankmcsherry]

Well, you can poke around with a copy of the repo? The file that is causing the issues is https://github.com/frankmcsherry/timely-dataflow/blob/master/src/progress/broadcast.rs, and though it doesn't seem to use Extend or extend directly, it used to (so, the issue might be similar).

If I misunderstand what you mean by "code to poke around with", let me know. I don't have a smaller reproduction, in that keeping things building on stable means not breaking them by ripping out lines, but I'm happy to try and help.

[bluss]

I didn't understand the reason it didn't infer, looks like it should be quite simple. Smells a bit like a regression in the language, not caused by libstd changes(?) So are you sure the exact same code compiles on stable, but not on nightly?

[alexcrichton]

Oh oops sorry I missed the git link in the initial report! Nominating as this does look like a regression

[alexcrichton]

cc @rust-lang/lang and @rust-lang/comipler

[alexcrichton]

er, @rust-lang/compiler

[bkoropoff]

Note that changes to type inference that require the addition of annotations to existing code are considered "fair game". I'd still like to understand what changed, though.

[jroesch]

I don't think any intentional inference changes have landed, but I agree @bkoropoff we should at least figure out what changed and advise people accordingly.

[frankmcsherry]

More data on when this broke:

A coworker has a stale nightly, rustc 1.3.0-nightly (faa04a8 2015-06-30), which builds the git pull just fine. I'll see about getting a build against the crates.io version (his project isn't compatible with 0.0.6, but a new project that just loads the timely crate explodes).

My version, 1.3.0-nightly (16f64c3 2015-07-09) explodes on the git pull and crates.io's timely 0.0.6 (which is older than the git pull).

Crater with 1.3.0-nightly (2015-07-10) explodes on crates.io's timely 0.0.6. I suspect it would break on the git drop too.

This may help narrow down when the regression dropped in. I think multirust can pull down specific nightlies, so I may just do a few more of those.

[frankmcsherry]

courtesy multirust, against crates.io timely 0.0.6, using nightlies from http://static.rust-lang.org/dist/

nightly-2015-06-30: pass
nightly-2015-07-01: pass
nightly-2015-07-07: fail
nightly-2015-07-08: fail
nightly-2015-07-09: fail

I don't know why there is a week gap, but it seems to be somewhere in there.

[nikomatsakis]

triage: P-high -- it's a regression