[PATCH] Improve NNTPconnect error reporting

[Julien ÉLIE]

Hi Ian,

> In inn2 2.4.5, NNTPconnect does not always report errors very well.

I see that INN 2.5.2 has a similar behaviour.


> In the patch below I address this by always arranging for the buffer
> to contain a useful message.

The problem is that I cannot apply the patch.  The whole code has changed
in INN 2.5.2.


> @@ -53,8 +64,11 @@ int NNTPconnect(char *host, int port, FILE **FromServerp, FILE **ToServerp, char
>     hints.ai_family = PF_UNSPEC;
>     hints.ai_socktype = SOCK_STREAM;
>     sprintf(portbuf, "%d", port);
> -    if (getaddrinfo(host, portbuf, &hints, &addr) != 0)
> +    if ((gai_r = getaddrinfo(host, portbuf, &hints, &addr)) != 0) {
> + store_error(buff, "getaddrinfo failed: ", gai_strerror(gai_r));
> + errno = 0;
>  return -1;
> +    }
>
>     for (ressave = addr; addr; addr = addr->ai_next) {
>  if ((i = socket(addr->ai_family, addr->ai_socktype,

It is now driven in network_connect_host() in lib/network.c

/*
**  Like network_connect, but takes a host and a port instead of an addrinfo
**  struct list.  Returns the file descriptor of the open socket on success,
**  or -1 on failure.  If getaddrinfo fails, errno may not be set to anything
**  useful.
*/

and not directly in lib/remopen.c where we have:

    fd = network_connect_host(host, port, NULL);
    if (fd < 0)
        return -1;

with no information from the getaddrinfo return code.
The network API (headers in include/inn/network.h) would then need to be
changed if we have to pass it a buffer to write the error.
Is it advisable to do that?

-- 
Julien ÉLIE

« Il était une fois un homme fidèle, c'est une belle histoire.
  Il était une fois une femme fidèle, c'est un conte de fées. » (Maurice Jeanneret)