Partial type hint breakage

[sfackler]

This is a minimal reproduction of something I saw when updating some of my projects to use Vec.

use std::vec_ng::Vec;

fn main() {
    let s = [1i, 2, 3];
    let v: Vec<_> = s.iter().collect();
}

test.rs:5:21: 5:39 error: cannot determine a type for this bounded type parameter: unconstrained type
test.rs:5     let v: Vec<_> = s.iter().collect();
                              ^~~~~~~~~~~~~~~~~~

This manifested in other ways in larger programs, ICEing in one case:

 ❯ env RUST_LOG=std::rt::backtrace make                                                                                          [rust-postgres/master *$=]
rustc -O --cfg ndebug -L submodules/rust-openssl/build/ -L submodules/rust-openssl/build/ -L submodules/rust-phf/build/ -L submodules/rust-phf/build/ -L submodules/rust-phf/build/ --dep-info build/postgres.d \
    --out-dir build src/lib.rs
src/lib.rs:612:17: 612:42 error: the type of this value must be known in this context
src/lib.rs:612                 PgUnknownType { oid, .. } =>
                               ^~~~~~~~~~~~~~~~~~~~~~~~~
src/lib.rs:622:19: 622:26 error: the type of this value must be known in this context
src/lib.rs:622             match desc.ty {
                                 ^~~~~~~
error: internal compiler error: no type for node 74135: local oid (id=74135) in fcx 0x7f639ffba890
note: the compiler hit an unexpected failure path. this is a bug.
note: we would appreciate a bug report: http://static.rust-lang.org/doc/master/complement-bugreport.html
note: run with `RUST_LOG=std::rt::backtrace` for a backtrace
task 'rustc' failed at '~Any', /build/rust-git/src/rust/src/libsyntax/diagnostic.rs:123
stack backtrace:
   1:     0x7f63a9db6740 - rt::backtrace::imp::write::h584392c0aaf88f8agYb::v0.10.pre
   2:     0x7f63a9d1b370 - rt::unwind::begin_unwind_inner::h39118a123ff832cbUzb::v0.10.pre
   3:     0x7f63a80eac40 - <unknown>
   4:     0x7f63a80eb820 - diagnostic::Handler::bug::h3def07bec623c5d9q3b::v0.10.pre
   5:     0x7f63a5aa8580 - driver::session::Session_::bug::hc25e766e0c47b22fxyc::v0.10.pre
   6:     0x7f63a5cc2f70 - middle::typeck::check::FnCtxt::node_ty::h3bb3bf7881b4ddbc4W6::v0.10.pre
   7:     0x7f63a5ce35e0 - <unknown>
   8:     0x7f63a5cea980 - <unknown>
   9:     0x7f63a5cea980 - <unknown>
  10:     0x7f63a5cea3e0 - <unknown>
  11:     0x7f63a5ce8af0 - <unknown>
  12:     0x7f63a5cea220 - <unknown>
  13:     0x7f63a5ce8af0 - <unknown>
  14:     0x7f63a5ce8af0 - <unknown>
  15:     0x7f63a5cea220 - <unknown>
  16:     0x7f63a5ce8af0 - <unknown>
  17:     0x7f63a5ce8af0 - <unknown>
  18:     0x7f63a5ce8800 - <unknown>
  19:     0x7f63a5cea700 - <unknown>
  20:     0x7f63a5ced8b0 - middle::typeck::check::writeback::resolve_type_vars_in_fn::h60db597186ed5046Vs1::v0.10.pre
  21:     0x7f63a5d38ac0 - <unknown>
  22:     0x7f63a5d424e0 - <unknown>
  23:     0x7f63a5d33c20 - middle::typeck::check::check_item::h2dd1675c2d31364e1T5::v0.10.pre
  24:     0x7f63a5d38910 - middle::typeck::check::check_item_types::ha39a6b2247f5d06aaA5::v0.10.pre
  25:     0x7f63a5e32b70 - <unknown>
  26:     0x7f63a5e31870 - middle::typeck::check_crate::h1776a18a6977a78cZvr::v0.10.pre
  27:     0x7f63a6187f70 - driver::driver::phase_3_run_analysis_passes::hef56a7cb46a523e2WOa::v0.10.pre
  28:     0x7f63a618eb80 - driver::driver::compile_input::h692b87098ca31f542bb::v0.10.pre
  29:     0x7f63a61b2630 - run_compiler::h5fa6736bc743af23tNi::v0.10.pre
  30:     0x7f63a61c5380 - <unknown>
  31:     0x7f63a61c3cb0 - <unknown>
  32:     0x7f63a61bf6e0 - <unknown>
  33:     0x7f63a551d8c0 - <unknown>
  34:     0x7f63a9db0080 - <unknown>
  35:     0x7f63a9dbd1a0 - rust_try
  36:     0x7f63a9dafec0 - rt::task::Task::run::h413709669c303bc5s58::v0.10.pre
  37:     0x7f63a551d660 - <unknown>
  38:     0x7f63a9db37b0 - <unknown>
  39:     0x7f63a4a2afe0 - start_thread
  40:     0x7f63a522ace9 - clone
  41:                0x0 - <unknown>

This can be reproduced by taking sfackler/rust-postgres@c5abe8c and changing lines 592 and 598 of src/lib.rs to use partial type hints.

cc @Kimundi

[huonw]

Maybe partial type hints need feature gating.

[Kimundi]

I knew that feature was too easy to implement... Lets see if I can find the cause.

[Kimundi]

Well, I'm not sure how to fix this. This inference error implies to me that partial type hints actually don't work at all for the Vec<_> = .collect() case right now, and the existing tests didn't catch it because they inferred right from both sides.

According to @eddyb this is because the T in Vec<T> and FromIterator<T> don't have any direct constrain to be identical. (Which would be possible with HKT though)

While there is a impl<T> FromIterator<T> for Vec<T>, this link is apparently not currently considered, leading to the inference failure here.

cc @nikomatsakis

[sfackler]

@alexcrichton Any idea why there are so many <unknown> stacks in the backtrace? Does a non--g build not contain enough debug info for libbacktrace to work off of?

[thestinger]

I expect that there would be other ways of hitting these same inference bugs.

[japaric]

This snippet is working on 0.12 and master:

use std::collections::HashMap;

fn main() {
    let s = [(1i, 1u), (2, 2), (3, 3)];
    let v: Vec<_> = s.iter().collect();
    let m: HashMap<_, _> = s.iter().map(|&x| x).collect();
}

I'm sure one of @nikomatsakis' PR that got merged in the last two months fixed inference from partial hints. Not sure which one though.

I think this should be closed. (Unless someone got another test case that fails to infer from partial hints)

[nikomatsakis]

💃 I've been meaning to look into why this doesn't work. Happy to not have to.