Pre-generate tables.rs instead of doing it in build.rs?

[NobodyXu]

Generating tables.rs requires unicode-linebreak to pull in regex and hashbrown as build-dependencies, plus running the generation at compile-time, which makes compilation a lot slower.

Since LineBreak.txt doesn't change that often, generating it manually won't make it much harder to maintain,

I can submit a PR to convert the build.rs to a separate crate that is never published and can be run locally to consume the LineBreak.txt and generate src/tables.rs, does it sound good to you?

If you insist on keeping the build.rs, it might be a good idea to switch to regex-lite rust-lang/regex#961

[axelf4]

I can submit a PR to convert the build.rs to a separate crate that is never published and can be run locally to consume the LineBreak.txt and generate src/tables.rs, does it sound good to you?

Sure!, as long as it does not entail checking in the generated src/tables.rs into source control. (I think that should be avoidable by adding it to both .gitignore and the include field in Cargo.toml?)

[NobodyXu]

Sure!, as long as it does not entail checking in the generated src/tables.rs into source control. (I think that should be avoidable by adding it to both .gitignore and the include field in Cargo.toml?)

Yes, that's doable.
I will submit a PR for this.

[BurntSushi]

Can you say why you don't want to check the generated tables into source code? I do it for all my crates, like regex-syntax and bstr. It's the best of both worlds: you don't need to pass on your build-dependencies and you don't need any bespoke build instructions and you don't need to do things like cargo publish --allow-dirty. It's just cargo build and everything works.