Defer the proxy service initialization to first time resolve is called....

Defer the HttpNetworkSession initialization to first time CreateTransaction is called.

http://code.google.com/p/chromium/issues/detail?id=3639


Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=3767

[darin (slow to review), 2008-10-22 07:34:31]

LGTM, nice find!

[wtc, 2008-10-22 18:27:20]

LGTM.  Thanks for tracking this down.

http://codereview.chromium.org/7854/diff/1/2
File net/proxy/proxy_service.cc (right):

http://codereview.chromium.org/7854/diff/1/2#newcode262
Line 262: DCHECK(config_last_update_time_.ToInternalValue() == 0);
Initializing config_last_update_time_ to 0 may not ensure that the first call to
ResolveProxy calls UpdateConfig.  See below.

http://codereview.chromium.org/7854/diff/1/2#newcode272
Line 272: if ((TimeTicks::Now() - config_last_update_time_) >
kProxyConfigMaxAge)
If TimeTicks::Now() returns the application's uptime, this boolean expression
may be false if the applications calls ResolveProxy immediately after startup.

On Windows, TimeTicks::Now() returns the system's uptime, so this problem won't
happen.  Also the problem will correct itself after 5 seconds.  But you get my
idea that this could happen.

I hope we can come up with a solution that doesn't have this issue at all. 
Perhaps creating ProxyService lazily?  How does our WinHTTP code avoid this
issue?

[eroman, 2008-10-22 18:53:31]

http://codereview.chromium.org/7854/diff/1/2
File net/proxy/proxy_service.cc (right):

http://codereview.chromium.org/7854/diff/1/2#newcode272
Line 272: if ((TimeTicks::Now() - config_last_update_time_) >
kProxyConfigMaxAge)
> I hope we can come up with a solution that doesn't have this issue at all. 
> Perhaps creating ProxyService lazily?  How does our WinHTTP code avoid this
> issue?

Done -- the session is now lazily created, instead of lazily-initializing the
proxy service.

Please review the latest snapshot.

[wtc, 2008-10-22 19:15:30]

LGTM.

http://codereview.chromium.org/7854/diff/203/204
File net/http/http_network_layer.cc (right):

http://codereview.chromium.org/7854/diff/203/204#newcode64
Line 64: HttpNetworkSession* HttpNetworkLayer::session() {
I studied our WinHTTP code.  We only need to do this lazy creation in
CreateTransaction.  So we don't really need a 'session()' method.

http://codereview.chromium.org/7854/diff/203/204#newcode84
Line 84: return session()->auth_cache();
The WinHTTP code doesn't have any check for session_ in this method.

http://codereview.chromium.org/7854/diff/203/204#newcode90
Line 90: if (suspend)
In the WinHTTP code, we just check for (suspend && session_).

[eroman, 2008-10-22 20:08:54]

All done. Thanks for the help!

[wtc, 2008-10-22 21:11:48]

LGTM.

Remember to update the CL's description, something like:
  Defer the HttpNetworkSession initialization to first time CreateTransaction is
called