Add rapidfuzz

[michaelweinold]

No description provided.

[michaelweinold]

@DerThorsten, thank you for adding your fixes to the build recipe - I can see the PR build is now stuck on a build_with_rattler function. Is this a known problem?

[DerThorsten]

@michaelweinold the problem is, it drags in a outdated build of the emscripten compiler

emscripten_emscripten-wasm32 ┆ 3.1.45 ┆ h7b187fe_15 ┆ emscripten-forge ┆ 7.19 KiB │
the problem is the build number, it is at 15, we need build 29.
Normally this is picked up without any problems, but some package constraints drag in this old build.
I probably need to remove the old builds of the compiler by hand :/

[DerThorsten]

@michaelweinold the old unusable builds of the compiler will be remove at some point in the nearby future (cannot give you a specific time as I cannot do this by myself)

[DerThorsten]

@michaelweinold So now it uses the proper emscripten compiler, but fails with some different stuff

[michaelweinold]

"At least" the error message is similar to those of other failing (auto-)builds, like:

• Update pydantic-core from 2.20.1 to 2.21.0 #1260

[maxbachmann]

@michaelweinold I am the author of the rapidfuzz package. First a couple of notes:

• I did improve the build of rapidfuzz to include -fexceptions so it doesn't crash when rapidfuzz uses C++ exceptions in emscripten.
• The package does now use scikit-build-core. This mentions supporting emscripten while scikit-build doesn't.
• In my CI I can now successfully build + test the package using cibuildwheel
• I am not completely sure whether numpy should be a runtime dependency. In the python package it's an optional dependency since it's only required by a small subset of the library. Given the size I assume it could make sense not to include it here and let the users install it themselves if needed for their application.

Where is the standard place to add emscripten/pyodide packages? I noticed there is both emscripten-forge and pyodide. Would a package usually get added to both of them?

[michaelweinold]

Hello @maxbachmann! Thank you for getting in touch.
Pyodide is a monolithic distribution - any packages added there will be shipped with Pyodide itself. Emscripten Forge (if I understand correctly) works like the Conda Forge package repository, which allows you to install packages as needed.
I will try to find time to re-do the build here based on the latest version of rapidfuzz early next year.

[psychemedia]

I'm trying to make use of rapidfuzz in a JupyterLite xeus-python environment, which installs wasm32 build packages from emscripten-forge, so I'd +1 on this being added as a feedstock. :-)