When more than one write buffer (iomatrix) was getting used to compose a response, only the first buffer was getting sent. Instead, try to call writev() again immediately if another iomatrix is pending (until writev() returns EAGAIN or some other error). This defect has likely been present for a long time, since at least 0cc3309 (2010 Jun 22) when Feersum changed from write() to writev().
This seems to fail frequently via CPAN Testers. Replaced looking for plack in the PATH with a search through the site/vend bin/script dirs. If an executable script isn't found, skip the tests.
* also fix t/60-plack.t since changing the version number
stash ‣ miyagawa_: qq: should psgi.version say [1,0] or [1,1]? i noticed Starman (Starlet?) had that. should Feersum? miyagawa_ ‣ stash: if you use psgix.io then 1.1 Spec: https://github.com/miyagawa/psgi-specs/blob/master/PSGI.pod
…plicit in using ExtUtils::Makefile 6.51 in Makefile.PL.
The likely/unlikely assertions were made based on (in order): * expected common use cases (e.g. GET > POST) * gcov branch stats * guessing
Prevents a *potential* blocking situation where the socket couldn't be made Non-blocking, etc.