Looks like there are 291 formula that still use GOPATH.

While we're here - would it be worth standardizing some of the flags that we use for building Go formulae? Specifically, I would advocate for always including

• -ldflags "-s -w", unless it breaks build
• -trimpath: "remove all file system paths from the resulting executable. Instead of absolute file system paths, the recorded file names will begin with either "go" (for the standard library), or a module path@version (when using modules), or a plain import path (when using GOPATH)."

It looks like Debian sets -trimpath but doesn't strip binaries with -s -w. Would it be useful to add something like std_go_args (like we have for CMake builds)?

Would it be useful to add something like std_go_args (like we have for CMake builds)?

I think it would be useful. I've already seen some formulae floating around with stripped binaries, which is why I've brought this up. From my limited testing with the PRs above, there's a reduction of 15-20% in total formula size. I personally don't think it would be a bad idea to strip binaries, seeing as Go binaries are all statically linked.

Maybe you want to wait @chenrui333 and do gopath at the same time as the go_std_args?

Would it be useful to add something like std_go_args (like we have for CMake builds)?

+1 for std_go_args idea.

We can definitely pipeline the work between std_go_args and gopath removal though.
After that we can refine our template for easy bootstrap golang formula.

All right! we are all done ( i think). We have PRs for all the tools listed in @carlocab's comment.

While we have filed issues with a LOT of the tools there I doubt that many of them will actually upgrade to go modules by the time go1.17 is out. Would the idea be to deprecate them or disable them or add a go1.16 formula when that time comes?

Possibly both - we could deprecate as "unmaintained" if there is no response, but we'll let them depend on go@1.16 for that deprecation period.

I'm very likely going to start tests with 1.17 RC1 when that lands rather than the final version so that we're better prepared.

For 1.16: we just now need to double check a couple of the other failures but they should be easy to prove they are pre-existing or not and then we can rebase on master (with the bootstrap change we agreed on in #71370 (comment)).

Thanks very much for all the help, @roopakv. There are still a handful of formulae that don't use GOPATH but nevertheless did failed with go1.16: #71289 (comment)