remove chrome dependencies from win sandboxing headers.

remove chrome dependencies from win sandboxing headers. Wrap sandbox code to
make the main routine a little cleaner. Unify the parameters of each of the "main" entry points so we can more easily abstract platform differences in the future. BUG=5323
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=7825

[cpu_(ooo_6.6-7.5), 2009-01-08 22:51:03]

http://codereview.chromium.org/17426/diff/22/23
File chrome/app/chrome_dll_main.cc (right):

http://codereview.chromium.org/17426/diff/22/23#newcode9
Line 9: 
I think we have reached the point were the above comment becomes critical. The
number and disposition of #ifdefs on this file indicates that we need urgently
to come with a better (even if not final) abstraction. For no platform except
windows chrome_dll_main.cc really makes sense. It is time to pay the design debt
and move the common initalization to a separate cc/h and the windowisms will
stay here and each plaform needs to get its file.

[brettw, 2009-01-08 22:54:15]

Darin's idea was that we could pass a struct to BrowserMain instead. This would
include the necessary parameters necessary for that platform. This struct would
be created once and then we pass and use it in all the *Main() functions.

So for example: chrome/common/main_function_params.h:

struct MainFunctionParams() {
  CommandLine command_line;
#if defined(OS_WIN)
...
#elif defined(...

#endif
}

Then I would put the Ole stuff in a wrapper class ifdefed to nothing on Mac.
Then you don't need CallBrowserMain anymore (just use that class in the if
statement directly).

How about putting the big SandboxWrapper (maybe renamed to SandboxInitWrapper)
in a separate file? I think it would also be much easier to follow if there were
just different declarations, there is little in common with the ifdefs, and it's
very hard to follow.

[pink (ping after 24hrs), 2009-01-08 22:56:11]

http://codereview.chromium.org/17426/diff/22/23
File chrome/app/chrome_dll_main.cc (right):

http://codereview.chromium.org/17426/diff/22/23#newcode9
Line 9: 
Until we understand more about how the renderer and browser need to launch, I
don't think we can effectively do that, and that requires some experimentation,
which is ongoing.

[pink (ping after 24hrs), 2009-01-08 22:59:10]

> Then I would put the Ole stuff in a wrapper class ifdefed to nothing on Mac.
> Then you don't need CallBrowserMain anymore (just use that class in the if
> statement directly).

I don't understand this part. How would the wrapper be invoked?

> How about putting the big SandboxWrapper (maybe renamed to SandboxInitWrapper)

Sure thing, I just figured it was small enough that it didn't need to go into a
separate file, but I can do that.

[pink (ping after 24hrs), 2009-01-09 16:51:17]

I broke out the headers as suggested which removes a bunch of ifdefs from
chrome_dll_main. Take another look.

[brettw, 2009-01-09 18:10:47]

LGTM with style nits.

http://codereview.chromium.org/17426/diff/213/35
File chrome/app/scoped_ole_initializer.h (right):

http://codereview.chromium.org/17426/diff/213/35#newcode31
Line 31: ScopedOleInitializer() { };
Nit: no semicolon.

http://codereview.chromium.org/17426/diff/213/39
File chrome/common/main_function_params.h (right):

http://codereview.chromium.org/17426/diff/213/39#newcode19
Line 19: : command_line_(cl), sandbox_info_(sb) { }
Indent this 2 more spaces.

http://codereview.chromium.org/17426/diff/213/37
File chrome/common/sandbox_init_wrapper.h (right):

http://codereview.chromium.org/17426/diff/213/37#newcode39
Line 39: };
Maybe DISALLOW_COPY_AND_ASSIGN here? (And below.)

[Amanda Walker, 2009-01-09 18:34:57]

On 2009/01/08 22:54:15, brettw wrote:
> Darin's idea was that we could pass a struct to BrowserMain instead. This
would
> include the necessary parameters necessary for that platform. This struct
would
> be created once and then we pass and use it in all the *Main() functions.

This ties in pretty well with some refactoring I've been doing for test_shell to
define a "platform delegate" object to hold platform-specific initialization,
teardown, and state.  It's about ready for review, so maybe we can use that as a
jumping-off point to do the same thing with the browser/renderer/etc.

[pink (ping after 24hrs), 2009-01-09 20:01:10]

all done, i'll commit once the tree re-opens.