Fix flaky "t5516.81 deny fetch unreachable SHA1, allowtipsha1inwant=false"

[dscho]

This test seems to fail infrequently in our Azure Pipeline, apparently most often in the msvc job. See https://dev.azure.com/Git-for-Windows/git/_build/results?buildId=36820&view=ms.vss-test-web.build-test-results-tab for an example.

[dscho]

More information (from #257):

This seems to fail every once in a while, especially in Git for Windows' experimental MSVC job of our Azure Pipeline.

The issue seems to be that multiple processes report fatal errors at the same time, and our grep fails to verify that the expected error message was printed because they were interleaved. Example:

++ grep 'remote error:.*not our ref.*64ea4c133d59fa98e86a771eda009872d6ab2886$' err
++ echo 'error: '\''grep remote error:.*not our ref.*64ea4c133d59fa98e86a771eda009872d6ab2886$' 'err'\'' didn'\''t find a match in:'
error: 'grep remote error:.*not our ref.*64ea4c133d59fa98e86a771eda009872d6ab2886$ err' didn't find a match in:
++ test -s err
++ cat err
fatal: git uploadfatal: remote error: upload-pack: not our ref 6-pack: not our ref 64ea4c1334ea4c133d59fa98e86a771eda009872d6ab2886
d59fa98e86a771eda009872d6ab2886

[dscho]

This problem is related, but not identical, to the broken pipe issue on macOS for which a tentative fix was presented in https://public-inbox.org/git/[email protected]/.

[dscho]

Hmm. Seems that that thread has stalled for quite a while now, and VFS for Git still carries 246f269.

@derrickstolee do you think we should do something about it? Like e.g. change the signature of send_request() to return an int, and teach the callers to drain the reader and then call die_errno(_("unable to write to remote"));. From what I see, all three callers are in find_common(), therefore one could be labeled with write_failed and the other two could use a goto write_failed;.