Compilation taking unexpected long after 1.84

[mrain]

I'm building jf-pcs library from https://github.com/EspressoSystems/jellyfish

On rust version 1.83

rust version meta

[$] rustc --version --verbose                                                                                             
rustc 1.83.0 (90b35a623 2024-11-26)
binary: rustc
commit-hash: 90b35a6239c3d8bdabc530a6a0816f7ff89a0aaf
commit-date: 2024-11-26
host: aarch64-apple-darwin
release: 1.83.0
LLVM version: 19.1.1

It only took ~11s to build the library

[$] cargo build -p jf-pcs                                                                                                        
   Compiling proc-macro2 v1.0.86
   Compiling unicode-ident v1.0.12
   ... [excess outputs omitted]
   Compiling jf-utils v0.4.4 (https://github.com/EspressoSystems/jellyfish?tag=0.4.5#7d71dbef)
   Compiling jf-pcs v0.1.0 (/Users/chengyu/espressosys/jellyfish/pcs)
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 11.43s

Now on rust 1.84 the compilation runs almost forever:

rust version meta

rustc 1.84.0 (9fc6b4312 2025-01-07)
binary: rustc
commit-hash: 9fc6b43126469e3858e2fe86cafb4f0fd5068869
commit-date: 2025-01-07
host: aarch64-apple-darwin
release: 1.84.0
LLVM version: 19.1.5

[$] cargo build -p jf-pcs                                                                                                  
   Compiling proc-macro2 v1.0.86
   Compiling unicode-ident v1.0.12
   ... [excess outputs omitted]
   Compiling jf-utils v0.4.4 (https://github.com/EspressoSystems/jellyfish?tag=0.4.5#7d71dbef)
   Compiling jf-pcs v0.1.0 (/Users/chengyu/espressosys/jellyfish/pcs)
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 26m 43s

Suspect that this is due to the new trait solver.

[theemathas]

@rustbot label +regression-from-stable-to-stable

[lqd]

Indeed, this bisects to #130654 cc @rust-lang/initiative-trait-system-refactor. I'll look into minimizing this.

@rustbot label -E-needs-bisection

[lsunsi]

I have a pretty big project and I also see this regression: compile times are basically tripled on 1.84.
I'm not sure how useful it is, but self-profiling and diffing 1.83 and 1.84 on non-release gives me this

[lcnr]

I have a pretty big project and I also see this regression: compile times are basically tripled on 1.84. I'm not sure how useful it is, but self-profiling and diffing 1.83 and 1.84 on non-release gives me this

@lsunsi the above crate was bisected to the stabilization of the next-generation trait solver in coherence. Your crate seems to instead spend more time during monomorphization, i.e. while preparing for codegen. I expect this to be a separate problem and it would be good to open a new issue for it

[lsunsi]

@lcnr Awesome, I'll do that. How do I bisect in the same way? I'd love to pinpoint some change to the regression I'm seeing.