Anyone know of a guide or tutorial on adding dependency inference? I'm thinking about adding a plugin to compile C code

You should look at the

cc

backend. We already have dependency inference for C code.

I did see that but didn't understand how it works

Incomplete is correct. A lot of the base stuff is there, but I was splitting it into 3-4 PRs for easier PRing before work became crazy. My original impetus for this was also embedded programming, which is why I had hooks for grabbing gcc-arm as well

Nice! Good to know it was also for embedded

I think trying to be too clever about generating objects, using the Pants cache and all that. That's nice and all, and maybe it's useful in some bootstrapping situations, but I feel like if I were to do it again (and maybe I will re-do it this way), I'd probably want Pants as a front-end to a ninja backend. Like, let the battle-tested, super fast tool do what it's supposed to, and have the Pants plugin act more like cmake to generate the appropriate ninja build files

cmake interop would be nice. right now we have complicated make system that ignores our deps' cmake files. my main goal is to unify our builds across boards

So, as an example, in the JS backend, Pants does some dep management, but npm or one of those package managers handles the nitty-gritty about the language. I have a Swift plugin somewhere, and that forks everything off to the swift compiler. There were talks about starting from the LLVM layer with Pants in CC, but I dunno, I see the practical side being higher-level configuration, management, toolchains. The actual act of compiling is honestly an "irrelevant" implementation detail

I've wanted to switch to clang but it doesn't have as small code size as gcc. So, I'd vote to keep them separate

Well, the toolchain part is just something you specify at the system level (or download if it's downloadable). Use gcc, gcc-arm, clang, zig, 🤷 who cares. I view it more as "here's the libraries I want out, here's the binaries I want out, Pants - go figure out via dep inference where everything is - and make that happen" Whether we use raw gcc commands and cache object files galore, vs just calling out to ninja to perform the compilation step - I think that's super irrelevant

what does ninja get you that pants doesn't?

Probably performance. In my limited tests, needing to create all those processes and caching results was slower than ninja. Not crazy amounts, but like... Why even. So, the idea would be that Pants is more equivalent to Cmake

either one will be faster than our current make system...

😆

it'll become a priority for us to work on when github starts charging open source projects for build time

Uhhh that's a thing?

not yet

🙏

you can always do the ninja backend for pants later

I agree. Its an experimental backend anyways - so, any of the low level changes wouldn't actually touch the API. And in a cruel twist of irony, the very project I'm backlogged working on right now, was the exact project I was building that plugin for. And now, I'm trying to cram a toolchain together to build this thing again 🤦‍♂️

🙂