On Fri, 20 Sep 2002, Michael Rueger wrote:

> Bert Freudenberg wrote:
> 
> > 
> > I have a version that might work through the api on hopefully most Linux 
> > browsers. If only I could test it!
> 
> OK, I disabled the "magic" on squeakalpha.org (our test server), so 
> every browser should try to use the api.

Okay. Still, the default in the image for API supported is "false". So on 
any page that does not explicitely include the param, the api is not used. 
For example, the no-api hack fails horribly when loading a page from the 
local disk. Going through the API works fine here. Another case is when 
the SRC is given as relative name. Works fine through browser, fails 
otherwise.

I hacked HTTPClient>>browserSupportsAPI to always return true if
Smalltalk platformName = 'unix'. *Even* then, an HTTP connection is opened 
when I load, e.g.,  the Essay from squeakalpha. Can this be because 
CachingHTTPRequest is derived from HTTPRequest? IMO it should be a wrapper 
that either uses HTTPRequest or PluginHTTPRequest. 

Hmm, I tried subclassing CachingHTTPRequest from PluginHTTPRequest.  It
still opens an HTTP connection. Where does this come from? It must be the
loading of project ressources, because these are not requested from the
API. Could you fix that?

And could you included this changeset in at least the plugin image? It 
wouldn't hurt in the bas image, too:
	http://swiki.gsug.org:8080/sqfixes/1930.html
This removes that endless loop when the image tries to ensure the
existence of the user directory, but fails to do that because of the file 
sandbox. Another possibility would be fixing FileDir>>ensureExistence.

> If you get unsupported browser, please send me the id strings.

Opera/6.02 (Linux 2.4.9-31smp i686; U) [en]
Mozilla/5.0 (Linux 2.4.9-31smp i686; U) Opera 6.02 [en]
Mozilla/4.78 (Linux 2.4.9-31smp i686; U) Opera 6.02 [en]
Mozilla/3.0 (Linux 2.4.9-31smp i686; U) Opera 6.02 [en]
Mozilla/4.0 (compatible; MSIE 5.0; Linux 2.4.9-31smp i686) Opera 6.02 [en]

Aren't these in your log files? ;-)

-- Bert