Images hosted at a different TCP port on the same host are not included in PDF - patch included

[michaelrsweet]

Version: 1.8-current
Original reporter: Iain W. Bird

Where a web server hosts sites on multiple ports, if a site on e.g. port 81 includes images or other content from e.g. port 82, then HTTP keepalive prevents disconnect and reconnect to the new port, hence ERR404 is emitted.

The fix is to always reconnect if the port number differs.

Patch included

[michaelrsweet]

"htmldocimg.txt":

Index: htmldoc/file.c
===================================================================
--- htmldoc/file.c	(revision 1736)
+++ htmldoc/file.c	(working copy)
@@ -96,6 +96,7 @@
 char	cookies[1024] = "";		/* HTTP cookies, if any */
 char	referer_url[HTTP_MAX_VALUE] = "";
 					/* HTTP referer, if any */
+static int last_connport = 0;		/* Port of last connection */ 
 
 
 /*
@@ -364,6 +365,7 @@
  * 'file_find_check()' - Check to see if the specified file or URL exists...
  */
 
+
 static const char *			/* O - Pathname or NULL */
 file_find_check(const char *filename)	/* I - File or URL */
 {
@@ -461,15 +463,22 @@
         strlcpy(connpath, resource, sizeof(connpath));
       }
 
-      if (strcasecmp(httpGetHostname(http, tempname, sizeof(tempname)),
-                     hostname))
+      if (last_connport != connport || (strcasecmp(httpGetHostname(http, tempname, sizeof(tempname)),
+                     hostname)))
       {
+	/*
+	 * Even if the host is the same, we must disconnect if the port is
+	 * different to avoid persisted HTML keepalives.
+	 */
+
+	last_connport = 0;
         httpClose(http);
         http = NULL;
       }
 
       if (http == NULL)
       {
+	last_connport = 0;
         progress_show("Connecting to %s...", connhost);
 
 #ifdef HAVE_SSL
@@ -488,6 +497,10 @@
 	                 "Unable to connect to %s!", connhost);
           return (NULL);
         }
+	else
+	{
+	  last_connport = connport;
+	}
       }
 
       progress_show("Getting %s...", connpath);

[michaelrsweet]

[master 87a317d] Fix a URL referencing bug when the same hostname but a different port was used (Issue #290)