Re: cvsimport weird

From: Pavel Roskin <proski@gnu.org>
Date: May 18, 2006, 05:44 UTC
Message-ID: <1147931094.32050.51.camel@dv>
In-Reply-To: <1147924771.32050.40.camel@dv>

On Wed, 2006-05-17 at 23:59 -0400, Pavel Roskin wrote:

> I'm quite sure that it's a bug in cvsps.  It displays such things on
> x86_64, but works properly on 32-bit PowerPC.

Address resolution is broken in cvsps on 64-bit machines. This patch to cvsps is needed:

--- cbtcommon/tcpsocket.c +++ cbtcommon/tcpsocket.c

@@ -198,7 +198,7 @@ convert_address(long *dest, const char *
     memcpy(dest, &ip.s_addr, sizeof(ip.s_addr));
   }
 #else
-  if ( (*dest = inet_addr(addr_str)) != -1)
+  if ( (*dest = inet_addr(addr_str)) != INADDR_NONE)
   {
     /* nothing */
   }


However, it's not sufficient to fix the original problem of empty CVS
server version string.  For some reason cvsps fails to authenticate with
pserver.  The ext protocol is working.

It's interesting that both git-cvsimport and cvsps have code to
authenticate over the pserver protocol.  I think maybe git-cvsimport
shouldn't do it, or maybe it should close some sockets before running
cvsps.

-- 
Regards,
Pavel Roskin