Avoid building crates with official pre-built artifacts?

[NobodyXu]

I wonder if we should stop building crates with official pre-build artifacts.
We could use cargo binstall --dry-run --strategies crate-meta-data "$crate" --targets "$TARGET_ARCH" to find out if such artifacts are present.

[NobodyXu]

@alsuren What do you think about this?

[alsuren]

I remember making quickinstall tell binstall not to look for upstream builds, because it was too slow. This means that quickinstall will still need us to build those packages. Even if we managed to make it faster, we would need to wait for the existing clients to die before disabling builds. I guess we could find that out from the stats server, but binstall doesn't report which version of quickinstall is driving it, so it might be guesswork. In the grand scheme of things, we will always burn a bunch of compute time building crates that nobody installs. I think we still break even overall though (we could check by summing up the installs from the stats server and comparing it against the how many crates we build in the same period).

…

On Mon, 13 Mar 2023, 08:09 Jiahao XU, ***@***.***> wrote: @alsuren <https://github.com/alsuren> What do you think about this? — Reply to this email directly, view it on GitHub <#197 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAB6FN2L2HJUXP6E2NXLLH3W33I2JANCNFSM6AAAAAAVYWL2AI> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[NobodyXu]

I remember making quickinstall tell binstall not to look for upstream
builds, because it was too slow. This means that quickinstall will still
need us to build those packages.

This is improved with v0.21.x release, which uses GitHub Restful API to speedup resolution and avoid creating a lot of http requests to GitHub that slows it down and gets it rate limited.

Even if we managed to make it faster, we would need to wait for the
existing clients to die before disabling builds. I guess we could find that
out from the stats server, but binstall doesn't report which version of
quickinstall is driving it, so it might be guesswork.

Hmm yeah, that's indeed a problem.

In the grand scheme of things, we will always burn a bunch of compute time
building crates that nobody installs. I think we still break even overall
though (we could check by summing up the installs from the stats server and
comparing it against the how many crates we build in the same period).

True.

[NobodyXu]

Let's put this issue to hold for now.