RDH: Don't attempt to sniff the mime type of requests that...

chrome/browser/renderer_host/buffered_resource_handler.cc

Index: chrome/browser/renderer_host/buffered_resource_handler.cc
===================================================================
--- chrome/browser/renderer_host/buffered_resource_handler.cc	(revision 24201)
+++ chrome/browser/renderer_host/buffered_resource_handler.cc	(working copy)
@@ -151,8 +151,12 @@
 
   const bool sniffing_blocked =
       LowerCaseEqualsASCII(content_type_options, "nosniff");
+  const bool no_data =
+      response_->response_head.headers &&  // Can be NULL if FTP.
+      (response_->response_head.headers->response_code() == 304 ||

[eroman, 2009/09/01 03:30:58]

I'm not sure that it is appropriate to put HTTP-code specific stuff outside of
the net/http module (I know we do this in other places, but that seems wrong).

In particular, this seems like logic duplicated from
http_network_transaction.cc.

Moreover, I don't think this list is exhaustive. For example, HEAD requests can
never have a body, regardless of the HTTP status code.

See HttpNetworkTransaction::DidReadResponseHeaders() which tries to handle all
of these cases in setting up |response_body_length_|.

I think it would be better to unify these interpretations of HTTP codes --
either by using shared functions living in src/net/http, or by exporting the
information from the network module through some field in the response object.

(Sorry to pitch in late, only just got back from vacation).

+       response_->response_head.headers->response_code() == 204);
   const bool we_would_like_to_sniff =
-      net::ShouldSniffMimeType(request_->url(), mime_type);
+      no_data ? false : net::ShouldSniffMimeType(request_->url(), mime_type);
 
   RecordSnifferMetrics(sniffing_blocked, we_would_like_to_sniff, mime_type);
 
@@ -165,7 +169,7 @@
     return true;
   }
 
-  if (sniffing_blocked && mime_type.empty()) {
+  if (sniffing_blocked && mime_type.empty() && !no_data) {
     // Ugg.  The server told us not to sniff the content but didn't give us a
     // mime type.  What's a browser to do?  Turns out, we're supposed to treat
     // the response as "text/plain".  This is the most secure option.
@@ -182,7 +186,7 @@
     return true;
   }
 
-  if (ShouldWaitForPlugins()) {
+  if (!no_data && ShouldWaitForPlugins()) {
     wait_for_plugins_ = true;
     return true;
   }