On Fri, Feb 5, 2016 at 5:42 AM, Joachim Achtzehnter <joac...@kraut.ca> wrote: > >> Are you sure the earlier sononblock() call succeeded (ie. my patch on >> socket_bucket_read() does not help)? > > It did not get called. The reason was that it assumed the socket was already > in non-blocking mode. This happened because the build was configured with > 'ac_cv_o_nonblock_inherited=yes', assuming that this property is inherited > from the listen socket, but it is not.
Ooook, now it makes sense. > > I have to apologize for having wasted your time. I feel really bad about > this. The only small excuse I have is that we were given the build > configuration for this cross-compile setup by another company, so in some > sense we are also a victim ourselves. No problem, at least we could validate the 2.4.18 changes :) > > It doesn't help that the configure script for the apr library silently makes > the above assumption: > >> if test "$cross_compiling" = yes; then : >> >> ac_cv_o_nonblock_inherited="yes" > > > This makes it non-obvious that one is creating an incorrect configuration. > There are several other places in the configure script that make assumptions > like this. In other places the configure script produces an error instead, > forcing the user to specify a value, which is much better: > >> if test "$cross_compiling" = yes; then : >> as_fn_error $? "cannot check setpgrp when cross compiling" "$LINENO" 5 Cross compiling APR probably needs fixes, patches welcome ;) > > Again, sorry about the wasted time. Thanks for the follow up. Regards, Yann.
