nalgebra and differential-dataflow regressed

[Zoxc]

nalgebra v0.10.1 and differential-dataflow v0.0.3 works on 5aaa606, but are broken on 9f15631.

UPDATE: nalgebra fixed, different-dataflow not yet.

[est31]

broken in which way? do you have an error message or ICE you can share with us?

[Zoxc]

nalgebra:

   Compiling nalgebra v0.10.1
error[E0391]: unsupported cyclic reference between types/traits detected
   --> C:\Users\John\.cargo\registry\src\github.1485827954.workers.dev-1ecc6299db9ec823\nalgebra-0.10.1\src\structs\dmatrix.rs:148:5
    |
148 |     mij:   [N; 2 * 2],
    |     ^^^^^^^^^^^^^^^^^ cyclic reference
    |
note: the cycle begins when processing `structs::dmatrix::DMatrix2::mij`...
   --> C:\Users\John\.cargo\registry\src\github.1485827954.workers.dev-1ecc6299db9ec823\nalgebra-0.10.1\src\structs\dmatrix.rs:148:5
    |
148 |     mij:   [N; 2 * 2],
    |     ^^^^^^^^^^^^^^^^^
note: ...which then requires const-evaluating `structs::dmatrix::DMatrix2::{{initializer}}`...
   --> C:\Users\John\.cargo\registry\src\github.1485827954.workers.dev-1ecc6299db9ec823\nalgebra-0.10.1\src\structs\dmatrix.rs:148:16
    |
148 |     mij:   [N; 2 * 2],
    |                ^^^^^
note: ...which then requires processing `structs::dmatrix::DMatrix2::{{initializer}}`...
   --> C:\Users\John\.cargo\registry\src\github.1485827954.workers.dev-1ecc6299db9ec823\nalgebra-0.10.1\src\structs\dmatrix.rs:148:16
    |
148 |     mij:   [N; 2 * 2],
    |                ^^^^^
note: ...which then requires coherence checking all impls of trait `std::ops::Mul`...
note: ...which then requires processing `structs::dmatrix::DMatrix`...
   --> C:\Users\John\.cargo\registry\src\github.1485827954.workers.dev-1ecc6299db9ec823\nalgebra-0.10.1\src\structs\dmatrix.rs:20:1
    |
20  | / pub struct DMatrix<N> {
21  | |     nrows: usize,
22  | |     ncols: usize,
23  | |     mij:   Vec<N>
24  | | }
    | |_^
note: ...which then requires computing the variances for items in this crate...
    = note: ...which then again requires processing `structs::dmatrix::DMatrix2::mij`, completing the cycle.

differential-dataflow:

   Compiling differential-dataflow v0.0.3
error: no associated item named `new` found for type `timely_sort::Sorter<_>` in the current scope
   --> C:\Users\John\.cargo\registry\src\github.1485827954.workers.dev-1ecc6299db9ec823\differential-dataflow-0.0.3\src\operators\arrange.rs:122:26
    |
122 |         let mut sorter = LSBRadixSorter::new();
    |                          ^^^^^^^^^^^^^^^^^^^
    |
    = help: items from traits can only be used if the trait is in scope; the following trait is implemented but not in scope, perhaps add a `use` for it:
    = help: candidate #1: `use timely_sort::RadixSorterBase;`

error: no associated item named `new` found for type `timely_sort::Sorter<_>` in the current scope
   --> C:\Users\John\.cargo\registry\src\github.1485827954.workers.dev-1ecc6299db9ec823\differential-dataflow-0.0.3\src\operators\arrange.rs:222:26
    |
222 |         let mut sorter = LSBRadixSorter::new();
    |                          ^^^^^^^^^^^^^^^^^^^
    |
    = help: items from traits can only be used if the trait is in scope; the following trait is implemented but not in scope, perhaps add a `use` for it:
    = help: candidate #1: `use timely_sort::RadixSorterBase;`

error: no associated item named `new` found for type `timely_sort::Sorter<_>` in the current scope
   --> C:\Users\John\.cargo\registry\src\github.1485827954.workers.dev-1ecc6299db9ec823\differential-dataflow-0.0.3\src\operators\consolidate.rs:110:70
    |
110 |                 let entry = inputs.entry_or_insert(index.time(), || (LSBRadixSorter::new(), 0, default_threshold));
    |                                                                      ^^^^^^^^^^^^^^^^^^^
    |
    = help: items from traits can only be used if the trait is in scope; the following trait is implemented but not in scope, perhaps add a `use` for it:
    = help: candidate #1: `use timely_sort::RadixSorterBase;`

[TimNN]

Looks like a duplicate of #41936, hopefully fixed by #41937 once it lands.

[frewsxcv]

Is this fixed now?

[Zoxc]

I think nalgebra is fixed now. It may not compile with RUST_LOG=rustc_typeck::check::regionck due to that failing with rustc_serialize.

[nikomatsakis]

What about differential-dataflow ?

Note that compiling with RUST_LOG is not really a supported workflow, though the current setup is generating a lot of cycle errors in those cases and needs to be fixed.

[Zoxc]

nalgebra compiles on rustc 1.19.0-nightly (d015610db 2017-06-05), differential-dataflow gives the same error.

[nikomatsakis]

OK. Regression. Nominating.

[nikomatsakis]

cc @rust-lang/compiler -- regression, anybody have any ideas?

[eddyb]

differential-dataflow is a completely different error, and a much more serious one. Either traits-in-scope is broken or there's an inherent method that's not found.

[est31]

I've tried to bisect the regression, but all I could find was rust compiler versions where the compilation failed.

• rust 1.11 stable fails compilation
• rust 1.16 stable fails compilation
• rust 1.17 stable fails compilation
• commit 4465f22 fails compilation, including all commits bisect-rust tested down to 927c55d where uploading on each merge started.

Therefore, @Zoxc can you please point to a rustc version where compilation works?

[Zoxc]

It did work on a crater run on top of 5aaa606. That looks pretty close to 1.16 stable though.

[Zoxc]

Version 0.0.3 was published on crates.io Jun 7, 2016, so it should probably work on rust 1.9.

[est31]

Still fails compilation on 1.9: http://paste.ubuntu.com/24804993/
And on 1.8 as well.

[Zoxc]

I built 5aaa606 locally and it doesn't work there either :/

[nikomatsakis]

A bit surprising that we've not seen this in cargobomb runs or other things. I'm inclined to believe that was a bug in that release of differential-dataflow -- but hard to say without closer investigation
.
cc @frankmcsherry -- any thoughts? we're looking at a (potential, unverified) regression in v0.0.3 of differential-dataflow (I see you're up to 0.1.0 now)

[nikomatsakis]

triage: P-high

I'm marking this P-high, but mostly because I want to try and confirm if there has been a regression here or not (hence no assignee)

[frankmcsherry]

When PartialOrd was ret-conned to mean "the same as Ord, if Ord exists" all correctness for 0.0.3 went out the window. The behavior of sorting changed for types where partial_cmp was not identical to cmp, and I'm sure that there was fall-out.

[nikomatsakis]

@frankmcsherry hmm, ok. This is a compilation error though.

[frankmcsherry]

Ah, sorry. The reason I haven't been tracking it is for that reason. The reason there is a compile error above is because the timely_sort crate had a breaking change. It got fixed in master, but can't fix 0.0.3.

[nikomatsakis]

@frankmcsherry thanks! :) I'm going to close this bug then.