Have you benchmarked the performance impact of this? It feels a bit like premature optimization to me.

• Binary size for release: 32535304 → 32530720
• PDF export time: 17665106 ns → 17410679 ns on 9n-proto with 1 run using the --timings flag

So the performance impact is pretty small, but I don’t see any reason why the code should perform worse after this change. And getting rid of these particular Lazy statics was easy compared to some of the ones elsewhere in the codebase.

So the performance impact is pretty small, but I don’t see any reason why the code should perform worse after this change.

I'm not concerned about the code performing worse, mostly about unnecessary complexity with build.rs and slower builds because some dependencies are built for the build script (and potentially even twice if host and target mismatch, like for wasm).

I'm not concerned about the code performing worse, mostly about unnecessary complexity with build.rs and slower builds because some dependencies are built for the build script (and potentially even twice if host and target mismatch, like for wasm).

Sure, that’s the one disadvantage of this change. Another option would be to perform the minifying and compression in typst-assets’s build script.

For me, personally, a build.rs needs to bring meaningful benefits to justify its existence. This does not feel like it's the case here, so I'll close this. Thanks anyway!