Improve protocol major macro

[roberth]

Why

• Prevent future surprises like Add Store::isTrustedClient() #7515 (comment)
• Fix an exception message in writeDerivedPaths

What

Most significant change: add PROTOCOL_MAJOR. All other changes follow from this. From the commit message:

PROTOCOL_MAJOR returns a simple major number, rather than the
number multiplied by 256. This is what one would reasonably
expect from GET_PROTOCOL_MAJOR, but we can't just change a
macro in a public header; hence the "rename".

[tfc]

Does anything speak against implementing that as a nicely typed function like this:

constexpr size_t protocol_minor(size_t x) { return x & 0xff; }

This way this happens at compile time if possible and no one will ever have to handle problems that occur from entering the wrong signed/unsigned type.

[tfc]

it seems like it would make sense to give the different major/minor macros (or functions if the other suggestion is followed) differing names as this also makes it easier to grep for their usage etc.

[thufschmitt]

Duh, that was indeed a nice footgun.

In addition to #7523 (comment), could we also deprecate the old macros? Deprecating a macro isn't really possible as far as I know, but we can hack something like:

constexpr size_t deprecated(size_t x) { return x }
#define GET_PROTOCOL_MAJOR(x) (deprecated((x) & 0xff00))
#define GET_PROTOCOL_MINOR(x) (deprecated((x) & 0x00ff))

[edolstra]

I would just remove the old macros. I think the only use is in Hydra which can be updated easily.

[roberth]

The old macro should stick around for compatibility, as Hydra isn't the only consumer. The new macros are also necessary for library-consuming code that needs to work around certain issues (shouldn't have to, but escape hatches are good).

TODO: clean up the implementation code to use a struct version.minor is nicer than a ton of macro calls. (thanks @Ericson2314)

[Ericson2314]

@roberth and I discussed this, and we think it would be good for the C++ code to fully parse the version once into a struct, and then pass that around instead.

@roberth also mentioned it would still be good to have a macro to --- in effect --- document the binary format itself separate from Nix's parser implementing it (which is the spirit of these foo-protocol.hh headers anyways), and that sounds fine to me.

[roberth]

fully parse the version once into a struct

Have tried this but failed. I accidentally broke something.

I would prefer to just merge this, to provide an alternative to the footgun, and then move on to some more relevant problem solving.

[Ericson2314]

@roberth Sounds good. Do you think you could push your attempt somewhere? I would be happy to take a stab at the follow-up if that helps.

[roberth]

It looks like I did a proper rage quit deleting my edits. It's mostly just a bunch of find and replace, so it's probably best to start over anyway. Sorry for dropping the actual struct bit though :/

EDIT: also this was like a week ago and my reflog is not helping either.