usa / garbagecollect.jp wrote:
> Perhaps, we should write a test that actually causes EINTR during `connect`.
> 
> But, when I committed r7931, there was a failure of an existing test, I guess.
> At the moment, such failure has not been observed.
> So, I think that there is not a big portability problem in this patch.

Thanks for confirming.  Unfortunately, EINTR is not easy to trigger on a
connect test case to localhost.  Not even -Wl,--wrap in GNU ld is useful
for testing this, it is dependent on socket state in the kernel.