Remove sys-info dependency

[rgwood]

Hi there,

This PR removes the sys-info dependency, which is relatively large with a few dependencies of its own. Because we only use a tiny part of sys-info (the Linux part of os_release()), it's easy to "inline" what we need and avoid taking a dependency.

I think this makes sense to keep compile times and binary size down, especially on non-Linux platforms where we don't need anything from sys-info. Let me know what you think!

Licensing

sys-info is MIT-licensed, so as I understand it this is OK because I've copied the copyright+permission notices verbatim. Doing so in source code comments seemed easiest.

[TheLarkInn]

Looks great and makes sense! 🎉🎉 Thanks for the contribution! 🙇

[TheLarkInn]

Desirability as compile time and downloaded binary size is a new realization for me. However, in regards to binary size, doesn't the compiler on Release remove unused parts of a library?

[rgwood]

in regards to binary size, doesn't the compiler on Release remove unused parts of a library?

That’s a good point. I believe you’re right, so this is mostly a compile time improvement (IIRC the entire dependent crate gets compiled even if only part of it is used in the final executable).

From a binary size perspective, one (small, hypothetical) issue with the previous approach would be non-Linux platforms. The compiler didn’t have enough info to know that os_release() would never be called on non-Linux platforms and it looks like the non-Linux path for os_release() has quite a bit going on. But this is admittedly a bit speculative, I shoulda measured binary size :)

[TheLarkInn]

I love this idea to be able to measure binary sizes for a compiled library. Perhaps this would be a nice tool to create for PR workflows! I'm going to write this in my notebook to work on for the future!

[rgwood]

You might find cargo bloat useful for inspiration: https://github.com/RazrFalcon/cargo-bloat