On Wed, 2014-07-02 at 20:25 +0000, Hoda, Sahir wrote:
> Hi all,
>
> I've been working on getting the proton-c library to build on Solaris, and I
> have had to implement a few workarounds. I'd like to get the group's opinion
> on these changes, and hopefully get some sort of official Solaris support
> going.
>
> First and foremost, I hit the following error upon compiling posix/io.c:
>
> #error "Don't know how to turn off SIGPIPE on this platform"
>
> As the error indicates, Solaris does not have a way to turn off SIGPIPE. My
> workaround here was to simply not attempt to ignore SIGPIPE:
I think that Solaris does - it's just that no one has had access to
Solaris to implement this.
> ..
> This is based on the assumption that given that SIGPIPE cannot be disabled on
> Solaris, most applications built for Solaris must have mechanisms in place to
> deal with SIGPIPE. What do others think?
Nope, this is unfortunately wrong, applications cannot be expected to
cope with signals being caused by a library which is basically outside
their control, and the signal will be raised sometimes.
At worst you'll have to ignore the signal and hope that none of your
applications use it.
>
> The second issue I hit was that in pn_listen and pn_connect, the call to
> getaddrinfo always failed, unless I added a hint of ai_socktype=SOCK_STREAM:
>
> --- a/proton-c/src/posix/io.c
> +++ b/proton-c/src/posix/io.c
> @@ -123,8 +123,11 @@ static inline int pn_create_socket(int af);
> pn_socket_t pn_listen(pn_io_t *io, const char *host, const char *port)
> {
> - struct addrinfo *addr;
> - int code = getaddrinfo(host, port, NULL, &addr);
> + struct addrinfo hints = {0}, *addr;
> + int code;
> +
> + hints.ai_socktype = SOCK_STREAM;
> + code = getaddrinfo(host, port, &hints, &addr);
> if (code) {
> pn_error_format(io->error, PN_ERR, "getaddrinfo(%s, %s): %s\n", host,
> port, gai_strerror(code));
> return PN_INVALID_SOCKET;
> @@ -163,8 +166,11 @@ pn_socket_t pn_listen(pn_io_t *io, const char *host,
> const char *port)
> pn_socket_t pn_connect(pn_io_t *io, const char *host, const char *port)
> {
> - struct addrinfo *addr;
> - int code = getaddrinfo(host, port, NULL, &addr);
> + struct addrinfo hints = {0}, *addr;
> + int code;
> +
> + hints.ai_socktype = SOCK_STREAM;
> + code = getaddrinfo(host, port, &hints, &addr);
> if (code) {
> pn_error_format(io->error, PN_ERR, "getaddrinfo(%s, %s): %s", host,
> port, gai_strerror(code));
> return PN_INVALID_SOCKET;
>
> Would there be any reason not to provide this hint for all platforms?
That sounds like a bug in the Solaris getaddrinfo. But there's nothing
wring with specifying hints - you will need to specify all the correct
defaults once you have it not just SOCK_STREAM - look at the posix
manpage for this.
>
> The third major issue I encountered was that 'errno' was not being read in my
> multi-threaded application. This would manifest as constant failure returns
> from pn_connect - because the 'connect()' socket call was returning
> EINPROGRESS - but errno was always set to zero. To fix this I needed to pass
> the '-pthread' option to gcc, by adding '-DCMAKE_C_FLAGS=-pthread' to the
> cmake command line. Perhaps this option could be set inside the
> CMakeLists.txt itself, but I am not familiar with cmake. If anybody can
> suggest how the the -pthread argument can be integrated into cmake, I would
> be happy to try it out.
Hmm, I don't understand this explanation, perhaps you can elaborate?
Andrew
