Sure, attached is a patch for the usock problem.
But hm, this was against the 0.1 release, should I do it against the trunk?
Regarding the patching of configure script, there must be something
wrong with the automake (or is it autoconf) on this machine.
The logic for setting NN_HAVE_EVENTFD works, and the logic for
NN_USE_EVENTFD seems ok. But even though it correctly finds that eventfd
doesn't exist on the system, it still sets NN_USE_EVENTFD.
AS_IF([test x"$ac_cv_func_eventfd"=xyes], [
AC_DEFINE([NN_USE_EVENTFD])], [
AS_IF([test x"$ac_cv_func_pipe"=xyes], [
AC_DEFINE([NN_USE_PIPE])], [
AS_IF([test x"$ac_cv_func_socketpair"=xyes], [
AC_DEFINE([NN_USE_SOCKETPAIR])], [
AC_MSG_ERROR([No signaling supported])
])
])
])
As expected, $ac_cv_func_eventfd has value "no".
/André
On 2013-09-17 09:40, Martin Sustrik wrote:
Hi Andre,
Would you mind providing a patch?
Martin
On 17/09/13 09:29, André Jonsson wrote:
On 2013-09-16 16:29, Martin Sustrik wrote:
On 16/09/13 16:13, André Jonsson wrote:
I can see that on my own (linux) machine that iov_base is a void *,
how
does that work? I mean, is the code in usock_posix.inc not used then?
It is used. No idea why it compiles.
In any case the line should look something like this:
((uint8_t*) hdr->msg_iov->iov_base) += nbytes;
Does that help?
Martin
Hm... I thought I replied to this already, but I can't find it... oh
well.
It does indeed help! The whole thing compiled like a charm (save for a
handful of warnings).
However, I had to do add an extra level of indirection for it to
compile:
*((uint8_t**)&(hdr->msg_iov->iov_base)) += nbytes;
Which is horrible... but without it, it said:
CC src/aio/usock.lo
"src/aio/usock_posix.inc", line 901: error #2137: expression must be a
modifiable lvalue
((uint8_t*)hdr->msg_iov->iov_base) += nbytes;
Anyway, I'll see if I can make some adjustments to configure.ac for it
to generate to correct defines for this platform (and hopefully not
break it for others).
/André
