Add missing nested socket pools to socket pool info page.

Add missing nested socket pools to socket pool info page.  Each socket pool should be displayed once and only once.  Also, if more than one proxy is in use, socket pools with the same proxy will appear adjacent to each other.

BUG=39756
TEST=manual


Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=58826

[eroman, 2010-09-03 21:39:46]

LGTM,

Could you have vandebo could take a look as well? There have been some subtle
issues surrounding destruction ordering of pools/sockets, and I am not sure if
holding additional references here will cause problems.

In fact, rather than adding the extra scoped_refptrs<> to the pools, I wander if
we could reach directly into the connect job factory and access the pointers
from there?

http://codereview.chromium.org/3328009/diff/13002/3003
File chrome/browser/resources/net_internals/socketsview.js (right):

http://codereview.chromium.org/3328009/diff/13002/3003#newcode39
chrome/browser/resources/net_internals/socketsview.js:39: // Puts socket pools
with the same name in lists contained within a new list.
nit: typically we use java-doc style comment blocks for javascript function
comments.

as in:

/*
  * XXXX
  */

http://codereview.chromium.org/3328009/diff/13002/3003#newcode44
chrome/browser/resources/net_internals/socketsview.js:44: if
(!socketPoolNameLists[socketPoolInfo[i].name]) {
nit: I suggest aliasing socketPoolInfo[i].name to a variable within the "for"
loop, as it used 4 times.

[mmenke, 2010-09-07 14:24:26]

On 2010/09/03 21:39:46, eroman wrote:
> LGTM,
> 
> Could you have vandebo could take a look as well? There have been some subtle
> issues surrounding destruction ordering of pools/sockets, and I am not sure if
> holding additional references here will cause problems.
> 
> In fact, rather than adding the extra scoped_refptrs<> to the pools, I wander
if
> we could reach directly into the connect job factory and access the pointers
> from there?

Certainly doable.  I just went with the extra socket pool pointers because it
required a bit less code.  I'll hold off on changing anything until hearing from
vandebo.

[vandebo (ex-Chrome), 2010-09-07 19:57:41]

LGTM with respect to pool references.

On 2010/09/03 21:39:46, eroman wrote:
> Could you have vandebo could take a look as well? There have been some subtle
> issues surrounding destruction ordering of pools/sockets, and I am not sure if
> holding additional references here will cause problems.

The additional references shouldn't change anything.  The ConnectJobFactories
hold references to subordinate pools, and the factories are referenced from
ClientSocketPoolBase.  So by adding subordinate pool references to the pool
classes, you're just holding two references (at two different abstraction
levels) instead of one in the same class.

http://codereview.chromium.org/3328009/diff/13002/3006
File net/http/http_proxy_client_socket_pool.h (right):

http://codereview.chromium.org/3328009/diff/13002/3006#newcode223
net/http/http_proxy_client_socket_pool.h:223: const
scoped_refptr<TCPClientSocketPool> tcp_pool_;
These need to go before base_, or the constructor needs to be changed. (same in
socks and ssl)

[mmenke, 2010-09-07 20:51:38]

On 2010/09/07 19:57:41, vandebo wrote:
> LGTM with respect to pool references.
> 
> On 2010/09/03 21:39:46, eroman wrote:
> > Could you have vandebo could take a look as well? There have been some
subtle
> > issues surrounding destruction ordering of pools/sockets, and I am not sure
if
> > holding additional references here will cause problems.
> 
> The additional references shouldn't change anything.  The ConnectJobFactories
> hold references to subordinate pools, and the factories are referenced from
> ClientSocketPoolBase.  So by adding subordinate pool references to the pool
> classes, you're just holding two references (at two different abstraction
> levels) instead of one in the same class.
> 
> http://codereview.chromium.org/3328009/diff/13002/3006
> File net/http/http_proxy_client_socket_pool.h (right):
> 
> http://codereview.chromium.org/3328009/diff/13002/3006#newcode223
> net/http/http_proxy_client_socket_pool.h:223: const
> scoped_refptr<TCPClientSocketPool> tcp_pool_;
> These need to go before base_, or the constructor needs to be changed. (same
in
> socks and ssl)

Thanks.  I'll keep it as-is, then (With the nits taken care of, of course). 
Getting the factory back out from the ConnectJobFactoryAdaptor is just a bit too
messy for my tastes.