error introduced in geo v0.22.1 by "rustdoc: Add item_template macro"

[coffinmatician]

Please see #112202 for the PR introducing the regression.

Code

geo v0.22.1 with the following Cargo.lock file. I can try to minify it, but I don't really want to.

Cargo.lock.txt

Meta

See the bisect below (I originally bisected over a larger range, but I wanted to recreate the message quickly):

~/Downloads/geo-0.22.1$ cargo bisect-rustc --preserve --start 2c718d12599ca00dc3957ec604cb3e87fe437124 --end ffb9b61294b96c389d343a4c55b15400249d74e6 -- build --release
bisecting ci builds
starting at 2c718d12599ca00dc3957ec604cb3e87fe437124, ending at ffb9b61294b96c389d343a4c55b15400249d74e6
opening existing repository at "rust.git"
Found origin remote under name `origin`
refreshing repository at "rust.git"
fetching (via local git) commits from 2c718d12599ca00dc3957ec604cb3e87fe437124 to ffb9b61294b96c389d343a4c55b15400249d74e6
opening existing repository at "rust.git"
Found origin remote under name `origin`
refreshing repository at "rust.git"
looking up first commit
looking up second commit
checking that commits are by bors and thus have ci artifacts...
finding bors merge commits
found 2 bors merge commits in the specified range
  commit[0] 2023-07-16: Auto merge of #113738 - jyn514:rollup-mjcya4c, r=jyn514
  commit[1] 2023-07-16: Auto merge of #113430 - Zalathar:hash, r=b-naber
validated commits found, specifying toolchains

checking the start range to verify it passes
installing 2c718d12599ca00dc3957ec604cb3e87fe437124
testing...
RESULT: 2c718d12599ca00dc3957ec604cb3e87fe437124, ===> No

checking the end range to verify it does not pass
installing ffb9b61294b96c389d343a4c55b15400249d74e6
testing...
RESULT: ffb9b61294b96c389d343a4c55b15400249d74e6, ===> Yes

searched toolchains 2c718d12599ca00dc3957ec604cb3e87fe437124 through ffb9b61294b96c389d343a4c55b15400249d74e6
checking last toolchain to determine final result
installing ffb9b61294b96c389d343a4c55b15400249d74e6
testing...


********************************************************************************
Regression in ffb9b61294b96c389d343a4c55b15400249d74e6
********************************************************************************

Attempting to search unrolled perf builds
ERROR: couldn't find perf build comment

Error output

The same error appears without the RUSTFLAGS=-Awarnings it just has a lot of spam.

~/Downloads/geo-0.22.1$ RUSTFLAGS=-Awarnings cargo +1.71.0 build --release
   Compiling geo v0.22.1 (/home/alex/Downloads/geo-0.22.1)
error[E0275]: overflow evaluating the requirement `[closure@src/algorithm/map_coords.rs:888:69: 888:72]: Fn<(geo_types::Coord<T>,)>`
  |
  = help: consider increasing the recursion limit by adding a `#![recursion_limit = "256"]` attribute to your crate (`geo`)
  = note: required for `&[closure@src/algorithm/map_coords.rs:888:69: 888:72]` to implement `Fn<(geo_types::Coord<T>,)>`
  = note: 128 redundant requirements hidden
  = note: required for `&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&...` to implement `Fn<(geo_types::Coord<T>,)>`
  = note: the full type name has been written to '/home/alex/Downloads/geo-0.22.1/target/release/deps/geo-31a6d43c26e8a8cc.long-type-10550508034351353399.txt'

For more information about this error, try `rustc --explain E0275`.
error: could not compile `geo` (lib) due to previous error

[clubby789]

Assuming https://github.com/georust/geo/tree/geo-0.22.1/geo is the crate and version with the issue, I'm not able to compile it as-is, because of a #[deprecated(since = 0.7.5, ..)] attribute, which is invalid syntax for the attribute. If I remove that line, I'm able to bisect it to

searched commit range: 5bd28f5...3307274
regressed commit: a20a04e -> #113108 cc @compiler-errors

bisected with cargo-bisect-rustc v0.6.6

Host triple: x86_64-unknown-linux-gnu
Reproduce with:

cargo bisect-rustc --preserve --start 2023-06-01 --end 2023-07-17 --prompt -- build --release 

which seems more likely as it changes the type system. Removing the ICE label as this just seems to be a query recursion overflow, please let me know if this is incorrect

[clubby789]

Actually, this doesn't seem to error any more on the current nightly. Can you verify if the issue still reproduces after updating?

[coffinmatician]

Here is a further minified test case. I am very confused on why changing the module names of things (too much, but minor changes seem to still trigger the issue) seems to stop triggering the issue. It seems very odd.
geo-0.22.1-minified.tar.gz

[clubby789]

Thank you for providing code to reproduce the issue - this one does indeed produce an error on current nightly. Having bisected a couple of times and gotten a different result every time, it seems a little inconsistent to track down.

However, both the error you posted and the one I'm getting are a regular query overflow error (although still likely a bug). An ICE is a specific error where the compiler panics.

[saethlin]

I strongly suspect that this regressed due to my MIR inliner change #109247, but I do not know how/why it was subsequently fixed (if it even was). Probably the only way to understand this is to brute-force instead of bisecting this because of the number of times this has probably changed between being an error and not.

[Jules-Bertholet]

@rustbot label regression-untriaged

[coffinmatician]

An ICE is a specific error where the compiler panics.

I was under the impression that ICE means any compiler error that keeps (reasonable) code that is expect to compile from compiling. Thank you for explaining that to me.

@clubby789 Also sorry that I was grumpy yesterday.