[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: gawk 3.1.6 bug in /inet/tcp/lport/0/0
|
From: |
Corinna Vinschen |
|
Subject: |
Re: gawk 3.1.6 bug in /inet/tcp/lport/0/0 |
|
Date: |
Mon, 3 Dec 2007 11:38:01 +0100 |
|
User-agent: |
Mutt/1.5.16 (2007-06-09) |
Hi Aharon,
On Nov 30 11:16, Aharon Robbins wrote:
> Richard and Corinna:
> > On Nov 15 23:07, Aharon Robbins wrote:
> > > Greetings. Re the below. I don't get any such message under Linux.
> > > Please talk to the cygwin people directly. I have bcc'ed the Cygwin
> > > contact for gawk, so I hope she will be in touch w/you directly.
> > > [...]
> > > > From: Richard Narum
> > > >[...]
> > > > #!/bin/sh
> > > >
> > > > gawk '
> > > > BEGIN {
> > > > RS = ORS = "\r\n"
> > > > HttpService = "/inet/tcp/8080/0/0"
> > > > Hello = "<HTML><HEAD></HEAD><BODY><PRE>Hello
> > > > World!</PRE></BODY></HTML>"
> > > > Len = length(Hello) + length(ORS)
> > > > while ("awk" != "complex") {
> > > > print "HTTP/1.0 200 OK" |& HttpService
> > > > print "Content-Length: " Len ORS |& HttpService
> > > > print Hello |& HttpService
> > > > while ((HttpService |& getline) > 0)
> > > > continue;
> > > > close(HttpService)
> > > > }
> > > > }
> > > > '
> > > >
> > > > I get the following error.
> > > >
> > > >
> > > > gawk: cmd. line:8: fatal: remote host and port information (0, 0)
> > > > invalid
> >
> > That's a regression in io.c, function socketopen. The 3.1.5 code
> > called gethostbyname on a hostname "0", which leads to gethostbyname
> > returning NULL. However, this wasn't a problem because the code
> > explicitely checked for an input remote address of "0" and, if so,
> > allowed gethostbyname to fail without socketopen to fail:
> > [...]
> > In 3.1.6, the function getaddrinfo is used. If getaddrinfo returns -1,
> > gawk fails to open the socket, [...]
>
> I looked into this. The use of `any_remote_host' changed from 3.1.5 to 3.1.6.
> In 3.1.6 it is a correctly-used boolean and the logic that's there is fine.
>
> Under Linux, even when using the replacement getaddrinfo that calls
> gethostbyname,
> a hostname argument of "0" still seems to "work", in that gethostbyname does
> NOT
> return NULL. This is a bit weird.
I'm not quite sure if that's ok. Giving a "0" to gethostbyname is
beyond the scope of the gethostbyname function. SUSv3 has only this to
say on the subject(*):
"The name argument of gethostbyname() shall be a node name; the
behavior of gethostbyname() when passed a numeric address string is
unspecified. For IPv4, a numeric address string shall be in the
dotted-decimal notation described in inet_addr()."
It looks like returning a "0" hostent structure is covered by allowing
unspecified behaviour. Personally I'd rather think that gethostbyname
should always return NULL with h_errno set to HOST_NOT_FOUND, as Cygwin
(resp. WinSock) does.
> Please try it out and let me know.
It fixes this problem on Cygwin.
Thanks,
Corinna
> * io.c (socketopen): Use NULL as first argument to `getaddrinfo'
> if any_remote_host is true. Should help on Non-GLIBC systems.
>
>
> ===================================================================
> --- io.c 11 Aug 2007 19:49:23 -0000 1.13
> +++ io.c 30 Nov 2007 09:11:22 -0000
> @@ -1208,7 +1208,7 @@
> rhints.ai_family = lhints.ai_family;
> rhints.ai_protocol = lhints.ai_protocol;
>
> - rerror = getaddrinfo (remotehostname, remotepname, &rhints,
> &rres);
> + rerror = getaddrinfo (any_remote_host ? NULL : remotehostname,
> remotepname, &rhints, &rres);
> if (rerror) {
> if (lres0 != NULL)
> freeaddrinfo(lres0);