nightly 1.71-2023-04-17 overflows the recursion limit

[benthecarman]

I now get this error from one of my dependent crates (rust-miniscript) when compiling with the --release flag

error[E0275]: overflow evaluating the requirement `F: FnMut<(&Pk,)>`
  |
  = help: consider increasing the recursion limit by adding a `#![recursion_limit = "256"]` attribute to your crate (`miniscript`)
  = note: required for `&mut F` to implement `FnMut<(&Pk,)>`
  = note: 128 redundant requirements hidden
  = note: required for `&mut &mut &mut &mut &mut &mut &mut &mut &mut &mut &mut &mut &mut &mut &mut &mut &mut &mut &mut &mut &mut &mut &mut ...` to implement `FnMut<(&Pk,)>`
  = note: the full type name has been written to '/home/ben/projects/mutiny-node/target/wasm32-unknown-unknown/release/deps/miniscript-e33708ffc0e16b91.long-type-8328810264563709237.txt'

[jyn514]

Have you tried raising the recursion limit like it suggests? We don't have stability guarantees for the recursion limit.

If that doesn't help, please link to an exact version of your crate so it's possible to reproduce the error.

[compiler-errors]

The lack of a span on this error suggests that this is due to an overflow during codegen. Given the &mut &mut ..., I doubt raising the recursion limit will help here.

I'll bisect this I guess. Seems to reproduce on miniscript with cargo build --release --all-features.

[compiler-errors]

Well bisection came back with garbage:

searched nightlies: from nightly-2023-04-01 to nightly-2023-04-18 regressed nightly: nightly-2023-04-18 searched commit range: https://github.com/rust-lang/rust/compare/d0f204e4d750b62f9d6c2593405e828757126832...7908a1d65496b88626e4b7c193c81d777005d6f3 regressed commit: https://github.com/rust-lang/rust/commit/53ac4f8e2fc15e49ef3a04f98622a9b9db755fd4

bisected with cargo-bisect-rustc v0.6.5

Host triple: aarch64-unknown-linux-gnu Reproduce with: ```bash cargo bisect-rustc --start 2023-04-01 --end 2023-04-18 -- build --release --all-features ```

Though given the regression happened a few days ago, I'm suspicious of #109247. The overflow is actually happening in the MIR inliner. Here's a snippet of the call stack of the error:

  24:     0xffffa865f308 - <rustc_trait_selection[59d5fff7e50533b4]::traits::select::SelectionContext>::candidate_from_obligation
  25:     0xffffa8653904 - <rustc_trait_selection[59d5fff7e50533b4]::traits::select::SelectionContext>::select_from_obligation
  26:     0xffffa865e974 - <rustc_trait_selection[59d5fff7e50533b4]::traits::select::SelectionContext>::select
  27:     0xffffa8521564 - <rustc_trait_selection[59d5fff7e50533b4]::traits::fulfill::FulfillProcessor>::process_trait_obligation
  28:     0xffffa8520820 - <rustc_trait_selection[59d5fff7e50533b4]::traits::fulfill::FulfillProcessor as rustc_data_structures[7f1a4c6933e5419]::obligation_forest::ObligationProcessor>::process_obligation
  29:     0xffffa84db470 - <rustc_data_structures[7f1a4c6933e5419]::obligation_forest::ObligationForest<rustc_trait_selection[59d5fff7e50533b4]::traits::fulfill::PendingPredicateObligation>>::process_obligations::<rustc_trait_selection[59d5fff7e50533b4]::traits::fulfill::FulfillProcessor>
  30:     0xffffa851e104 - <rustc_trait_selection[59d5fff7e50533b4]::traits::fulfill::FulfillmentContext as rustc_infer[8facaf8255ab7cac]::traits::engine::TraitEngine>::select_where_possible
  31:     0xffffa77519f0 - <dyn rustc_infer[8facaf8255ab7cac]::traits::engine::TraitEngine as rustc_infer[8facaf8255ab7cac]::traits::engine::TraitEngineExt>::select_all_or_error
  32:     0xffffa77eeb58 - rustc_traits[88a431509884da59]::codegen::codegen_select_candidate
  33:     0xffffa7cbbb64 - rustc_query_system[7175af1eaca55fc6]::query::plumbing::try_execute_query::<rustc_query_impl[9bf706fa0936a2ef]::queries::codegen_select_candidate, rustc_query_impl[9bf706fa0936a2ef]::plumbing::QueryCtxt>
  34:     0xffffa7b789b8 - <rustc_query_impl[9bf706fa0936a2ef]::Queries as rustc_middle[c197f7dd5ada7ea7]::ty::query::QueryEngine>::codegen_select_candidate
  35:     0xffffa69d81b8 - rustc_ty_utils[e19ef8fcd720bdb9]::instance::inner_resolve_instance
  36:     0xffffa69d7758 - rustc_ty_utils[e19ef8fcd720bdb9]::instance::resolve_instance
  37:     0xffffa7c86000 - rustc_query_system[7175af1eaca55fc6]::query::plumbing::try_execute_query::<rustc_query_impl[9bf706fa0936a2ef]::queries::resolve_instance, rustc_query_impl[9bf706fa0936a2ef]::plumbing::QueryCtxt>
  38:     0xffffa7b859fc - <rustc_query_impl[9bf706fa0936a2ef]::Queries as rustc_middle[c197f7dd5ada7ea7]::ty::query::QueryEngine>::resolve_instance
  39:     0xffffa8885cd4 - <rustc_middle[c197f7dd5ada7ea7]::ty::instance::Instance>::resolve_opt_const_arg
  40:     0xffffa8885808 - <rustc_middle[c197f7dd5ada7ea7]::ty::instance::Instance>::resolve
  41:     0xffffa6f5e6b0 - rustc_mir_transform[d2e7dbab7faafc68]::inline::cycle::mir_callgraph_reachable::process
[ like 400 identical calls ]
 425:     0xffffa6f5e934 - rustc_mir_transform[d2e7dbab7faafc68]::inline::cycle::mir_callgraph_reachable::process
 426:     0xffffa6f5e2c4 - rustc_mir_transform[d2e7dbab7faafc68]::inline::cycle::mir_callgraph_reachable
 427:     0xffffa7cb6054 - rustc_query_system[7175af1eaca55fc6]::query::plumbing::try_execute_query::<rustc_query_impl[9bf706fa0936a2ef]::queries::mir_callgraph_reachable, rustc_query_impl[9bf706fa0936a2ef]::plumbing::QueryCtxt>
 428:     0xffffa7b73ee8 - <rustc_query_impl[9bf706fa0936a2ef]::Queries as rustc_middle[c197f7dd5ada7ea7]::ty::query::QueryEngine>::mir_callgraph_reachable
 429:     0xffffa6efde84 - <rustc_mir_transform[d2e7dbab7faafc68]::inline::Inliner>::try_inlining
 430:     0xffffa6efcf60 - <rustc_mir_transform[d2e7dbab7faafc68]::inline::Inliner>::process_blocks
 431:     0xffffa6efca28 - <rustc_mir_transform[d2e7dbab7faafc68]::inline::Inline as rustc_middle[c197f7dd5ada7ea7]::mir::MirPass>::run_pass
 432:     0xffffa6f396ec - rustc_mir_transform[d2e7dbab7faafc68]::pass_manager::run_passes_inner
 433:     0xffffa6e51ee8 - rustc_mir_transform[d2e7dbab7faafc68]::optimized_mir

cc @cjgillot @saethlin

[cjgillot]

From preliminary investigation, for_each_key is is polymorphically recursive: for_each_key<'_, F> calls for_each_key<'_, &mut F>. Downstream crates calling it should emit an error when building an executable. Do they?

[apoelstra]

@cjgillot they do not, until some recent nightly compilers, in which they now start emitting errors.

[apoelstra]

@cjgillot sorry, I am incorrect. Callers do emit errors. Previously this would happen later in the compile stage, and provide line number information, and only occur if they actually try to use the function. Now it occurs earlier, with no line number information, and regardless of whether the function is actually used.

[lnicola]

I think we're hitting this on https://github.com/georust/geo/issues/1010, but I can't actually reproduce it. Could it only fire intermittently or for some specific platforms?

The problematic code is https://github.com/georust/geo/blob/0.24.1/geo/src/algorithm/map_coords.rs#L850.

[apoelstra]

Looks like this bug has hit stable now :(.

[NotGovernor]

I'm new to rust, having this error, but also am worried my amatuerity? will probably mean anything I could add here would be more noise than signal. :D So forgive me if that is the case here.

I wanted to contribute my path to duplicating this. My project uses SurrealDB which depends on the geo library mentioned above. When I compile locally on my windows machine there is no issue. However, when I use the official rust docker image https://hub.docker.com/_/rust which builds using the RUN cargo install --path . command I get the recursion overflow error.

When I modify the docker to do a normal RUN cargo build I no longer get the error. I'm new enough to not know why, I just wanted to drop this here in case it helps. <3

[lnicola]

@NotGovernor I'm not familiar with SurrealDB, but the problematic code was removed in geo 0.25.0, and SurrealDB bumped that dependency from 0.25.0 to 0.25.1 about three weeks ago, so you shouldn't be seeing this.

I suggest checking which version you're using. cargo tree might help.

[sergei-pq]

I'm hitting this bug too with rust-miniscript. It's really strange error because it only happens if you have too many dependencies in your cargo crate. I can only reproduce this in a large repo with many dependencies / a lot of code, in a smaller isolated project with just miniscript and a few other dependencies, the issue doesn't happen, but it does in a multi-module crate which is really frustrating.

[apoelstra]

@sergei-pq which version of rust-miniscript are you using? We have a patch but we haven't backported it to every version.

[sergei-pq]

@apoelstra I observed this behavior on 9.0.1 being pulled in from bdk crate -- the problem is I can't change bdk directly. Instead I tried to use this patch operation to fix it.

bdk = {version = "0.28.0", features = ["default", "keys-bip39", "bip39"]}
miniscript = "10.0.0"

[patch.crates-io]
miniscript = { version = "10.0.0" }

This actually works for hotfixing BDK (or it at least fixes the issue since bdk is miniscript ^9 and seems to not break the code yet. This fix worked on my most recent issue with miniscript -- although I'm not certain it's entirely resolved. If I can't get it working I'm likely going to have to fork bdk and upgrade that directly -- or use the backported patch strategy.

[benthecarman]

@apoelstra would be great if it was backported for v9.x

[apoelstra]

I have an open PR https://github.com/rust-bitcoin/rust-miniscript/pull/566

[apoelstra]

Done.

[ActuallyHappening]

Still getting this issue in rustlang/rust:nightly docker image Cargo.toml contains geo="0.26.0", building with cargo +nightly chef cook --release ... and cargo leptos build --release

[apiraino]

@ActuallyHappening I'm trying to reproduce your error. Using the latest rustc 1.73.0-nightly (076887268 2023-08-17) I can compile the project leptos with cargo build --release. I think in your screenshot I'm reading geo version 0.24.1 instead of 0.26.0 (as per your comment)?

[ActuallyHappening]

Your right, after some investigation my git dependancy on surrealdb was using the outdated version of geo:

[dependencies.surrealdb]
git = "https://github.com/surrealdb/surrealdb.git"
tag = "v1.0.0-beta.9"
optional = true

which lead to this dependancy being included (from surreal/lib/Cargo.toml in commit c3773b2e):

geo = { version = "0.24.1", features = ["use-serde"] }

My full project failed to compile (using cargo leptos build --release) on the latest nightly nightly-x86_64-unknown-linux-gnu unchanged - rustc 1.74.0-nightly (ef85656a1 2023-08-21), but DOES compile on stable stable-x86_64-unknown-linux-gnu unchanged - rustc 1.71.1 (eb26296b5 2023-08-03). Using cargo tree shows that geo 0.24.1 is a dep of surrealdb.

I haven't completely solved this issue for my specific use case, but changing my Cargo.toml section to this seems to help:

# upgrading surrealdb so that it uses newer geo version
[dependencies.surrealdb]
git = "https://github.com/surrealdb/surrealdb.git"
# tag = "v1.0.0-beta.9"
branch = "main"
optional = true

# [patch.crates-io.geo]
# git = "https://github.com/georust/geo.git"
# branch = "main"

[apiraino]

Can this issue can be closed since the geo crate was updated?

thanks for a feedback cc @benthecarman @ActuallyHappening

@rustbot label -I-prioritize

[apoelstra]

Probably everybody hitting this bug (the miniscript crate also did) has updated to fix or remove the affected code (which could not have been used, since this bug only affects unused code). Since multiple actual rustc releases went out with this bug, even though it was reported with multiple months' lead time and at least two instances of deployed affeced code.

It theoretically represents a regression though, in that you can write Rust code which compiles before 1.71 and does not compile after.

For my part (as a maintainer of miniscript) I wouldn't mind closing it.