v1.49.x -- Ruby: Output a single binary file per platform-specific gem

[johnnyshields]

In #30933, I discovered the Ruby UCRT gem attempting to load the MSVCRT binary, and failing because that binary file is a zero-byte placeholder. Specifically, the gem root dir looks like this:

grpc_c.32-msvcrt.ruby - 0 KB
grpc_c.64-msvcrt.ruby - 0 KB
grpc_c.64-ucrt.ruby - 9,650 KB

It would be possible to fix the reference so the UCRT binary is loaded (probably a C build flag issue), however...

Rather than having 3 binaries files (1 real + 2 placeholders), the most elegant solution is to only create 1 binary file for each platform-specific, named grpc_c.ruby, which is:

• On Windows: the correct binary file for the given Windows architecture
• On Linux/Mac: a 0-byte placeholder

For reference, the UCRT gem root dir will now look like this:

grpc_c.ruby - 9,650 KB

This solution unambiguously avoids loading the wrong binary, and also gets rid of a lot of cruft in the build code at the same time. Obviously we'll never need more than one grpc_c binary for any given platform gem.

[johnnyshields]

@apolcyn ready for your review.

[johnnyshields]

@apolcyn any update on this?

[apolcyn]

Thanks for the investigation, @johnnyshields.

This PR looks reasonable to me. But given the size of this PR, and the fragility of the windows build, I'm a little apprehensive to merge this on 1.49. We'd need to release this change straight into a 1.49.1 patch release, without a pre-release, so we'd have more risk of breaking the whole windows build.

So I'd prefer to fix the build issue, which seems to not be setting the macro correctly, if possible. That said if it's very hard, I'm fine with this as a fall back.

Do you have any idea what's wrong there? The current extconf.rb code looks right w.r.t. detection of the ucrt platform, and I checked that rake-compiler-dock does fill in x64-mingw-ucrt as the RUBY_PLATFORM when running that code.

[johnnyshields]

@apolcyn unfortunately I've exhausted my bandwidth I can spend investigating this issue. Re: "fragility of Windows build", this PR makes it significantly less fragile IMHO by reducing complexity.

I'm fine to get this either in 1.49.1 or 1.50.0. Perhaps you could cut a pre-release of a tiny version v1.49.1-pre1. I'm glad to test on all Windows versions when you have a gem cut either way.

[apolcyn]

I think I've found the cflags issue - opened #31051. I'm backporting that and aiming to get that into 1.49.1

[johnnyshields]

@apolcyn thank you. I'm closing this PR (1.49.x backport) however will leave the PR for master open.