Careful. #104315 introduced a bug and I’m not sure @rsc has fixed it yet on master.

@benesch thanks, can you link to (or open) an issue for the introduced bug?

I'm not sure if there's an open issue or not. I was only notified by @rsc's comment here: https://go-review.googlesource.com/c/go/+/104315#message-17a7cff9ab15c2e73ca46760c16d0ba4ef5ae502

@benesch which CLs need to be cherry-picked to fix this issue?

I'm afraid I don't know. My solution in #104315 introduced a bug that @rsc said he'd fix:

FWIW, the code changes and justification here are wrong:
the split must start at pmain, because pmain can depend
on something (say, testing), which depends on p,
and that's not a cycle, but testing must still be
recompiled.

I will fix this as part of some other work I'm doing
in this code.

https://go-review.googlesource.com/c/go/+/104315#message-17a7cff9ab15c2e73ca46760c16d0ba4ef5ae502

As far as I can tell that fix has not yet been committed. I'd be happy to look into fixing it myself but I don't want to duplicate work with @rsc.

@FiloSottile Do you still feel that this should be backported? The backport justification didn't include the reason (ie. security issue, serious problem with no workaround).

If so, I'll follow up with @rsc to see if the described issue was resolved.

This issue doesn't have sufficient justification for being backported yet, so I moved it to Go 1.10.6 milestone.

Any update on the backport justification?

I flagged it for backport before we had the discussion about the criteria, and I assumed any regression causing an internal compiler error was worth discussing for backport. 1.11 came out in the meantime, and we agreed "use 1.11" is a valid workaround for cases like this, so closing.