We should retry shutdown on EINPROGRESS error. Fixes #3168

[akzhan]

Just like EAGAIN handling behavior but due to another reason.

[ysbaddaden]

Does it reliably fix the reduced test case in #3168 (granted you have the issue without the fix) ? If so, don't worry about writing a spec, we'll merge this long awaited fix!

[akzhan]

Yes, of course, it is real patch that fixes #3168.

[RX14]

I just tested this patch on my machine, it fixes #3168 for me.

[denysvitali]

Tested too, works for me!

dvitali@denvit-ws1:crystal/bin ‹retry-EINPROGRESS-on-unbuffered-close*›$ ./crystal ~/Documents/crystal/test/http_problem.cr 
Using compiled compiler at `.build/crystal'
dvitali@denvit-ws1:crystal/bin ‹retry-EINPROGRESS-on-unbuffered-close*›$ ./crystal -v                                      
Using compiled compiler at `.build/crystal'
Crystal 0.22.0+? [30c7410] (2017-05-20) LLVM 4.0.0

[ysbaddaden]

There is an issue on CI. Once fixed we'll be able to merge this.

[akzhan]

New commit to just reword comments to make it easy to understood after years.

[RX14]

Looks like circleci picked up this push and is building this PR. Travis OSX has been flaky on network for weeks, so build failures on travis but not circleci should probably be ignored.

[ysbaddaden]

Thank you!