Some tests are failed, can you look at them?

Ah, I was pretty sure pnmp test was working, maybe I missed something, I'll check again later today :)

ok... it was working, but I did a cleanup and ended up removing a flush that was actually necessary 😅, that was tricky to debug

I also forgot to update the dts files. They should be updated now

Could you show how you run the benchmark (so we can check this is really faster)?

Could you show how you run the benchmark (so we can check this is really faster)?

Sure, take this as input:

@for $i from 1 through 20000 {
  .foo-#{$i} {
    color: red;
  }
}

And run the following commands (on my phone currently, I can double check if needed on Monday):

sass input.scss output_large.css

\time -v npx postcss output_large.css > output_min.css

Uses default cssnano config.

I use \ for time so that I can pass -v, otherwise it tries to run the flag as a command 🤣

alright, I'm at the PC now. This gist should help https://gist.github.com/Goodwine/3759bf6dd78a180d6153b752a05d00be

1. Download this gist to a directory next to your cssnano repo
2. Pull the changes from this PR to your repo.
3. Run:

npm install
\time -v npm run test

@ludofischer Can you take a look? Thanks

It does seem to speed up the test case considerably. I have benchmarked the cli using hyperfine and these are the results:

without the PR:

  Time (mean ± σ):     46.373 s ±  1.104 s    [User: 45.849 s, System: 2.046 s]
  Range (min … max):   45.168 s … 48.938 s    10 runs

WIth the PR applied:

  Time (mean ± σ):      8.717 s ±  0.237 s    [User: 8.028 s, System: 1.806 s]
  Range (min … max):    8.370 s …  9.019 s    10 runs

Thanks! I understand that this makes the implementation more complex. I've tried to add documentation to explain it without overwhelming the code.

Ultimately, the problem is due to the expensive getters/setters in PostCSS. If PostCSS were to parse lazily, this wouldn't have been an issue (or at least, not as significant 😅).