chore(docs): notes on unstable rustc in HACKING.md

[thoughtpolice]

Some notes on rustc unstable usage, after talking in #320. Figured this would be nice to add.

I have a small upcoming change (I'll file it separately after trying to make a few more) that will use the tagref syntax; it's nothing to worry about for anyone else except me, I think. (e.g. int_roundings can be replaced with a small bit of code, but it would be nice to undo that when it does become stable, so keeping a note on it is nice.)

Also rework some changes from Neil's recent commit; everything in HACKING.md uses - for bullet points instead of *, so it's just a tiny consistency change.

[ndmitchell]

Happy enough with your reformat.

For the unstable features, I think we are only interested in removing unstable features if it does not harm the quality of the code too much. In most cases, the use of the unstable feature was to make things better, so normally I am against removing it. In the fullness of time most unstable features become stable, so I think it will happen eventually. E.g. for int_roundings, thoughtpolice@62d892d looks like it makes things worse, while int_roundings is likely to be stabilised. In contrast, thoughtpolice@c69f195 is just clearly cleaner code with less extension.

[ndmitchell]

I dropped some thoughts in #265 so suggest we figure it out there

[thoughtpolice]

Yeah, and I didn't put up thoughtpolice@62d892d yet actually because I was wondering if an analogue to div_ceil should just be put in Gazebo, or something

[ndmitchell]

Seems weird to put something that is soon going to be in the standard library in Gazebo. Putting it in buck2_core might be feasible, but as far as I remember, host stuff doesn't depend on buck2_core (for good internal reasons).

[thoughtpolice]

Yeah, I expect it will be in the standard library in a few months (there seems to be one or two final nits but the Rust core people are busy). Like I said, a bit of it is just getting a feel for what will be big changes and what will be small changes. The div_ceil thing is kind of funny though; it's a good example of where like, if someone had added this function somewhere else, changing it wouldn't be a big deal, but because the addition and the usage are happening all at once, it's easy to think whether it's worth it at all or just redundant. But it's not an important patch, either way.

[thoughtpolice]

In light of #292 happening, I think my need for this has been lowered a bit, and I don't think it's as big a deal as long as we have statically linked binaries for users to use; the need for distributions etc is a bit different than what this is catering to.

If the Meta team decides to move to rustc stable, it'll be loudly announced anyway, so I doubt it's a big deal.

I might still submit the occasional cleanup for things like this.