Hacker News Reader

Compiling Python extensions is a nightmare because we allow Autoconf, CMake, Visual Studio, Bazel etc. to make it complicated and nonportable; when someone sets out to wrap some library for Python the quality of the result is limited by the quality of the tools and by low expectations.

A serious engineering effort along the lines of the Zig compiler would allow Python to build almost everything from source out of the box; exotic compilers and binary dependencies, not "Frankenstein code bases" per se, are the actual obstacles.

What you’re proposing here is essentially “if we fix the C/C++ build systems environment, this would be easy!”. You’re absolutely right, but fixing that mess has been a multi-decade goal that’s gone nowhere.

One of the great victories of new systems languages like Rust and Zig is that they standardized build systems. But untangling each individual dependency’s pile of terrible CMake (or autoconf, or vcxproj) hacks is a project in itself, and it’s often a deeply political one tied up with the unique history of each project.

The typical build scripts and tools and terrible hacks that are adequate for a standalone C/C++/Fortran project are lacking for building the same project as a Python extension, which requires portability and supported, not too custom, build steps.

The ambition and usefulness of aiming for the latter higher standard is quite new: it has gone (relatively) nowhere because it has been a multi-decade non-goal, which only a small minority of users cares about.

Hard-to-build python extensions are basically .so files with some special symbols exposed.

As others said, a general task to "build libfoo.so" is very complex, and currently requires a wide variety of build systems. I don't see why this will get any easier if we require this .so file to export some Python-specific symbols.

> What you’re proposing here is essentially “if we fix the C/C++ build systems environment, this would be easy!”. You’re absolutely right, but fixing that mess has been a multi-decade goal that’s gone nowhere.

Not sure I would call it easy, as it would still take a lot of effort to update how PyPI works to account for these new capabilities, but that's exactly what the Zig compiler & build system solved.

Rust is completely hands off when it comes to C/C++ dependencies, Zig can package and build them. That's why I created https://github.com/allyourcodebase/.

As I've mentioned on lobsters, look a this example `build.zig.zon` file: https://github.com/allyourcodebase/srt/blob/main/build.zig.z...

It mentions 3 dependencies:

    Haivision/srt, the upstream C++ project
    mbedtls
    googletest

When you run zig build all these 3 dependencies are downloaded and their build.zig is run if present (the first one doesn’t have a build.zig since it’s just the vanilla C/C++ upstream project that we are providing a build script for).

The work to package everything must still happen, but once it’s done correctly you get the ability to build from any host for any target, which you can literally see happening in the CI runs of that repo: https://github.com/allyourcodebase/srt/actions/runs/10982569...

This kind of skepticism is exactly why I wrote this post. The details of actually coming up with a realistic upgrade path for PyPI are certainly much more nuanced than what I wrote in the post, but the core insight is that the C/C++ build intractability problem has been solved... and that you shouldn't depend on free big tech money if you can.

The "upgrade path" is for individual packages (providing portable build scripts) and for client-side Python package management (actually providing and running next-generation tools), not for PyPI which already supports package metadata stating what platforms a package is compatible for.

In my experience, `zig cc` is great at cross-compiling (which is a hard problem that it solves!) but doesn't actually help with the rest of the problem.

You still have to run cmake/autoconf/meson/whatever else, which is the part that's project-specific and often quite fiddly.

This has been my experience as well. I previously used cargo-zigbuild for packaging at work (and contributed both to cargo-zigbuild and Zig as a byproduct), and I still had several troubles that I had to analyze and tackle in spite of them.

Well if you're limiting yourself to zig cc, then all you get is a C compiler.

If you take the time to kick out of your project that soup of build systems and replace them with a build.zig, then you got yourself a complete solution. Takes some effort but it's perfectly doable, see https://github.com/allyourcodebase

They do, Zig ships clang. The easiest way to install clang is via pip.

    pip install ziglang

    python -m ziglang clang --version
    clang version 18.1.6 (https://github.com/ziglang/zig-bootstrap 98bc6bf4fc4009888d33941daf6b600d20a42a56)
    Target: aarch64-unknown-darwin23.6.0
    Thread model: posix
    InstalledDir: /usr/bin

Very soon, Zig will all your (code)base.

Zig is divorcing itself from Clang: https://github.com/ziglang/zig/issues/16270

Which doesn't mean what you're implying it means: https://github.com/ziglang/zig/issues/16270#issuecomment-161...

I think it does. If you want Clang, install Clang.

More info https://github.com/ziglang/zig/issues/16270#issuecomment-161...

You're the second person to post a link to a comment that's prominently highlighted on the issue I already linked. Yes, I've read that; nowhere do I see a statement like "Zig will always ship Clang with it" and instead I see a number of statements that imply it won't. I'm not even saying that getting Clang out of Zig is a bad thing. It's not like CMake or Rust come bundled with a C compiler.

I suggest you read it again then, because it heavily implies that.

> These use cases can still be satisfied by, again, an independently maintained project that combines Clang main() and Zig main() together. For users of these CLI tools, I don't expect there to be any difference in user experience.

Means when someone installs ziglang from their package manager, it will be able to build c.

Yeah, I've read it multiple times. Every time, it says to me "somebody somewhere else can package these things together, it just won't be Zig". There must be something about Andrew Kelley's communication style that clicks with other people but not me. But I just can't read into it what you say is there. Since this all boils down to interpreting one somewhat arrogant man's words, it ceases to be a technical discussion and just becomes an argument in semantic parsing. I'm just not going to comment much on Zig anymore as clearly other people know what's going on and I keep getting it wrong.

Zig is and will be shipping clang for a great long while. Not only will it be shipping clang so you can build other code with it, they aren't removing LLVM support.

https://github.com/ziglang/zig/issues/16270#issuecomment-205...

I raise you https://github.com/ziglang/zig/issues/20875

It is not entirely clear what any of this means for the future, and in any case, it keeps changing all the time.

The promise to maintain LLVM support just applies to Zig code. That doesn't require Clang. LLVM alone is not a C compiler, and neither is Zig without Clang.

What very well may happen is that the "standard" distribution of Zig will continue to bundle Clang, while a "minimal" distribution also gets created which omits Clang, and thus cannot compile C code. However, none of this has been spelled out clearly yet, and so I think it's reasonable to say don't depend on Zig to give you Clang.

Zig is migrating away from LLVM as the mandatory backend, it will still ship clang and llvm so that it can build everything it needs which is critical for its mission for all your (code) base.

Autoconf is easy, it's the highly bespoke build systems that someone thought would be a good idea that require the right phase of the moon that are the challenge.

Problem: saving bandwidth for pypi.

Solution: unify all software build stack.

Tomorrow, chat, we will tackle world hunger in an attempt to save people from anorexia.

Like and subscribe.

It's not that farfetched, Nix is living proof that it's possible. Zig's perspective is just from the inside of your codebase.

It wasn't even intentional for Zig at the beginning, though. And any such solution should be usable across all platforms, while Nix on Windows is still not remotely usable (the latest attempt seems to be [1]).

[1] https://lastlog.de/blog/libnix_roadmap.html

You and your respondents see this as a Python problem. I see it as a Zig problem.

As in, Zig will seamlessly add Python to its build system long before Python's build story is so robust.

Wheels do predate conda (though manylinux did base itself on the experience of Anaconda and Enthought's base set of libraries), and there were distributions like Enthought (or even more field specific distributions like Ureka and individuals like Christoph Gohlke) that provided binaries for the common packages.

What the conda ecosystem did was provide a not-horrible package manager that included the full stack (in a pinch to help fix up a student's git repository on a locked-down Windows system I used conda to get git, and you can get R and do R<->python connections easily), and by providing a standard repository interface (as opposed to a locked down and limited version that the other providers appeared to do), conda pushed out anyone doing something bespoke and centralised efforts so spins like conda-forge, bioconda, astroconda could focus on their niche and do it well.

If I didn’t have access to binary wheels, I wouldn’t have been able to learn Python in high school as a heavy Windows user.

Even now, running into package compile issues is a sure fire way to lose half an hour of my time.

> was here before Anaconda popularized the idea of binary packages for Python

This has incredible "I was there, Gandalf, three thousand years ago" vibes =P

the article does not suggest removing wheeels