Build script that does nothing

[dtolnay]

Eventually we will want a build script that enables Serde impls for i128 and u128 -- #1136. As a first step here is a build script that does nothing to see whether we can roll this out without breaking anyone's workflow, without having a supported feature at stake in the event that it needs to be rolled back.

[dtolnay]

After releasing this I would like to give people a week to complain before we start using this to enable any observable behavior.

[oli-obk]

Would it be overkill to use the rustc_version crate?

Good idea on the slow rollout. We could probably publish an rc version first when we try out an actual feature.

[dtolnay]

The rustc_version crate adds 2.5 CPU-seconds of compile time on my computer above what this build script adds, which seemed excessive. We don't need a production quality semver parser or complete rustc metadata handling.

[hcpl]

FYI there is also the version_check crate which compared to rustc_version is much more lightweight feature-wise and doesn't have dependencies.

[dtolnay]

Nice, I hadn't seen version_check before. That looks a lot more reasonable for our use case.

[dtolnay]

Thanks for demonstrating the implementation with version_check! In my own applications I would definitely use it. But for Serde I am going to stick with some simple parsing in the build script. I believe one of the defining challenges for Serde over the next few years will be avoiding bloat (or the perception of bloat) and there is a vast difference in perception between having dependencies and having no dependencies. I took a look at cargo clean && time cargo build --no-default-features before this PR, with just a build script, and with build script + version_check and the build script by itself adds 9% to compile time which is already close to the limit of what it makes sense to accept for the sake of i128 impls; build script + version_check adds 18%. But let's revisit if this approach turns out unreliable or unmaintainable.