"thread 'rustc' has overflowed its stack" introduced in 1.45.0

[epage]

Code

git clone https://github.com/crate-ci/typos.git
cd typos/crates/typos-dict
cargo +1.44.0 check  # works
cargo +1.45.0 check  # overflows
RUST_MIN_STACK=20000000 cargo check  # works

(also reproduced in beta and nightly)

Meta

rustc --version --verbose:

rustc 1.45.0 (5c1f21c3b 2020-07-13)
binary: rustc
commit-hash: 5c1f21c3b82297671ad3ae1e8c942d2ca92e84f2
commit-date: 2020-07-13
host: x86_64-apple-darwin
release: 1.45.0
LLVM version: 10.0

Error output

Checking typos-dict v0.2.1 (/Users/edpage/src/personal/typos/crates/typos-dict)

thread 'rustc' has overflowed its stack
fatal runtime error: stack overflow
error: could not compile `typos-dict`.

Caused by:
  process didn't exit successfully: `rustc --crate-name typos_dict --edition=2018 crates/typos-dict/src/lib.rs --error-format=json --json=diagnostic-rendered-ansi --crate-type lib --emit=dep-info,metadata -Cembed-bitcode=no -C debuginfo=2 -C metadata=439c5583e8d5e9ec -C extra-filename=-439c5583e8d5e9ec --out-dir /Users/edpage/src/personal/typos/target/debug/deps -C incremental=/Users/edpage/src/personal/typos/target/debug/incremental -L dependency=/Users/edpage/src/personal/typos/target/debug/deps --extern log=/Users/edpage/src/personal/typos/target/debug/deps/liblog-6af237d64da748f9.rmeta --extern phf=/Users/edpage/src/personal/typos/target/debug/deps/libphf-3c962e541346fbff.rmeta --extern unicase=/Users/edpage/src/personal/typos/target/debug/deps/libunicase-561e4e81056a68fd.rmeta` (signal: 6, SIGABRT: process abort signal)

Backtrace

No backtrace generated

[oli-obk]

This is already fixed on nightly, I just tried it locally. Can you confirm?

[epage]

Installed nightly last night but to double check:

$ rustup update
...
nightly-x86_64-apple-darwin unchanged - rustc 1.47.0-nightly (db0492ace 2020-07-29)

$ cargo +nightly check
   Compiling version_check v0.9.2
    Checking siphasher v0.3.3
   Compiling log v0.4.8
    Checking cfg-if v0.1.10
   Compiling unicase v2.6.0
    Checking phf_shared v0.8.0
    Checking phf v0.8.0
    Checking typos-dict v0.2.1 (/Users/edpage/src/personal/typos/crates/typos-dict)

thread 'rustc' has overflowed its stack
fatal runtime error: stack overflow
error: could not compile `typos-dict`.

Caused by:
  process didn't exit successfully: `rustc --crate-name typos_dict --edition=2018 crates/typos-dict/src/lib.rs --error-format=json --json=diagnostic-rendered-ansi --crate-type lib --emit=dep-info,metadata -Cembed-bitcode=no -C debuginfo=2 -C metadata=d5a6aa6c6eb18cbb -C extra-filename=-d5a6aa6c6eb18cbb --out-dir /Users/edpage/src/personal/typos/target/debug/deps -C incremental=/Users/edpage/src/personal/typos/target/debug/incremental -L dependency=/Users/edpage/src/personal/typos/target/debug/deps --extern log=/Users/edpage/src/personal/typos/target/debug/deps/liblog-194f264183e56651.rmeta --extern phf=/Users/edpage/src/personal/typos/target/debug/deps/libphf-49669432db2fabaf.rmeta --extern unicase=/Users/edpage/src/personal/typos/target/debug/deps/libunicase-0d7595d4b238a095.rmeta` (signal: 6, SIGABRT: process abort signal)

[oli-obk]

Hmm... must be apple specific. Do you have a gdb handy that you could run on that last rustc invocation to obtain a stack trace?

[epage]

Having problems getting it working, not normally a debugger person (been years since I've run gdb) and not much of a Mac person (work computer).

gdb --args /Users/edpage/.rustup/toolchains/nightly-x86_64-apple-darwin/bin/rustc --crate-name typos_dict --edition=2018 crates/typos-d
ict/src/lib.rs --error-format=json --json=diagnostic-rendered-ansi --crate-type lib --emit=dep-info,metadata -Cembed-bitcode=no -C debuginfo=2 -C metadata=d5a6aa6c6eb1
8cbb -C extra-filename=-d5a6aa6c6eb18cbb --out-dir /Users/edpage/src/personal/typos/target/debug/deps -C incremental=/Users/edpage/src/personal/typos/target/debug/incr
emental -L dependency=/Users/edpage/src/personal/typos/target/debug/deps --extern log=/Users/edpage/src/personal/typos/target/debug/deps/liblog-194f264183e56651.rmeta
--extern phf=/Users/edpage/src/personal/typos/target/debug/deps/libphf-49669432db2fabaf.rmeta --extern unicase=/Users/edpage/src/personal/typos/target/debug/deps/libun
icase-0d7595d4b238a095.rmeta

I installed rustc-dev for nightly and I get

Reading symbols from /Users/edpage/.rustup/toolchains/nightly-x86_64-apple-darwin/bin/rustc...

warning: Could not open OSO archive file "/Users/runner/work/1/s/build/x86_64-apple-darwin/stage1-rustc/x86_64-apple-darwin/release/deps/libjemalloc_sys-404271e36c84d6
5d.rlib"
(No debugging symbols found in /Users/edpage/.rustup/toolchains/nightly-x86_64-apple-darwin/bin/rustc)

Then when I run, I get

(gdb) run
Starting program: /Users/edpage/.rustup/toolchains/nightly-x86_64-apple-darwin/bin/rustc --crate-name typos_dict --edition=2018 crates/typos-dict/src/lib.rs --error-fo
rmat=json --json=diagnostic-rendered-ansi --crate-type lib --emit=dep-info,metadata -Cembed-bitcode=no -C debuginfo=2 -C metadata=d5a6aa6c6eb18cbb -C extra-filename=-d
5a6aa6c6eb18cbb --out-dir /Users/edpage/src/personal/typos/target/debug/deps -C incremental=/Users/edpage/src/personal/typos/target/debug/incremental -L dependency=/Us
ers/edpage/src/personal/typos/target/debug/deps --extern log=/Users/edpage/src/personal/typos/target/debug/deps/liblog-194f264183e56651.rmeta --extern phf=/Users/edpag
e/src/personal/typos/target/debug/deps/libphf-49669432db2fabaf.rmeta --extern unicase=/Users/edpage/src/personal/typos/target/debug/deps/libunicase-0d7595d4b238a095.rm
eta
Unable to find Mach task port for process-id 76733: (os/kern) failure (0x5).
 (please check gdb is codesigned - see taskgated(8)

[oli-obk]

hmm... not sure what our gdb capabilities on mac are. I'll try to find someone who's building rustc on macs. Thanks for the attempt!

[Mark-Simulacrum]

I'm looking into producing a build locally.

[Mark-Simulacrum]

@oli-obk backtrace: https://gist.github.com/Mark-Simulacrum/bb263799f9057d7000714caa21367fcc

[epage]

Looks like my CI saw a stack overflow on Windows as well

https://dev.azure.com/crate-ci/crate-ci/_build/results?buildId=1015&view=logs&jobId=f2aa1cb7-ddd8-5e13-2793-3e55ac100c6d&j=f2aa1cb7-ddd8-5e13-2793-3e55ac100c6d&t=5a4762e8-c879-51cf-cd2f-a8a50b37302f

[oli-obk]

lol I recently (two days ago) looked at collapse_goto_chain and went "oh this could stack overflow, why is this not a loop".

I'll (try to) fix this without inserting more stacker invocations