Move experiments settings (about:flags) from profile to local state.

This patch moves the experiments settings from the user profile to the browser's local state, since these settings correspond to command line switches.
This is a release blocker for NaCl, so we decided to push this through.

TEST=manual
BUG= http://code.google.com/p/nativeclient/issues/detail?id=1322

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=72225

[Nico, 2011-01-21 19:21:22]

The first line of the CL description is misleading, as this CL only moves
about:flags to local state.

This CL means that users will lose their current flag settings, right? But since
about:flags is now registered before there's a profile, writing migration code
is kinda hard, and we have a that scary warning message, so that's fine. It
would be nice if there was code that removed the now no-longer-used pref from
the user state.

Other than that, this looks good.

[Miranda Callahan, 2011-01-21 19:24:03]

On 2011/01/21 19:21:22, Nico wrote:
> The first line of the CL description is misleading, as this CL only moves
> about:flags to local state.
> 
> This CL means that users will lose their current flag settings, right? But
since
> about:flags is now registered before there's a profile, writing migration code
> is kinda hard, and we have a that scary warning message, so that's fine. It
> would be nice if there was code that removed the now no-longer-used pref from
> the user state.
> 
> Other than that, this looks good.

LGTM -- go ahead, and I'll add code to remove the no longer used prefs as part
of my next multi profile migration code.  Thanks!

[bbudge, 2011-01-21 22:55:05]

Changed the issue title.

On 2011/01/21 19:24:03, Miranda Callahan wrote:
> On 2011/01/21 19:21:22, Nico wrote:
> > The first line of the CL description is misleading, as this CL only moves
> > about:flags to local state.
> > 
> > This CL means that users will lose their current flag settings, right? But
> since
> > about:flags is now registered before there's a profile, writing migration
code
> > is kinda hard, and we have a that scary warning message, so that's fine. It
> > would be nice if there was code that removed the now no-longer-used pref
from
> > the user state.
> > 
> > Other than that, this looks good.
> 
> LGTM -- go ahead, and I'll add code to remove the no longer used prefs as part
> of my next multi profile migration code.  Thanks!

[stevenjb, 2011-01-22 06:02:02]

Unfortunately in ChromeOS, this change creates the following problem which
breaks all of the ChromeOS browser_tests and ui_tests. (See
http://build.chromium.org/p/chromium/builders/Linux%20Builder%20%28ChromiumOS...).

about_flags::ConvertFlagsToSwitches() calls
about_flags::MigrateChromeOSLabsPrefs() which expects a bunch of preferences to
be set.

These preferences are not actually set until ProfileImpl::GetPrefs() gets called
from ProfileManager::GetDefaultProfile () (the first occurrence for ChromeOS
being from g_browser_process->profile_manager()->GetDefaultProfile() called in
browser_main.cc).

I'm not sure whether we need to call GetDefaultProfile() earlier, call
ConvertFlagsToSwitches() later, or something more complicated.

[Nico, 2011-01-22 06:27:58]

stevenjb: I think we can just delete the ChromeOS migration code. It's
been there for a while, and stuff should be migrated for everyone who
cared about these settings. Do you think that will help?

On Fri, Jan 21, 2011 at 10:02 PM,  <stevenjb@chromium.org> wrote:
> Unfortunately in ChromeOS, this change creates the following problem which
> breaks all of the ChromeOS browser_tests and ui_tests. (See
>
http://build.chromium.org/p/chromium/builders/Linux%20Builder%20%28ChromiumOS...).
>
> about_flags::ConvertFlagsToSwitches() calls
> about_flags::MigrateChromeOSLabsPrefs() which expects a bunch of preferences
> to
> be set.
>
> These preferences are not actually set until ProfileImpl::GetPrefs() gets
> called
> from ProfileManager::GetDefaultProfile () (the first occurrence for ChromeOS
> being from g_browser_process->profile_manager()->GetDefaultProfile() called
> in
> browser_main.cc).
>
> I'm not sure whether we need to call GetDefaultProfile() earlier, call
> ConvertFlagsToSwitches() later, or something more complicated.
>
>
> http://codereview.chromium.org/6267009/
>

[Miranda Callahan, 2011-01-22 14:11:13]

I think that getting rid of the migration code is good in any case; as Nico
says, it's been there for a while -- and "about:flags" comes with so many
caveats as to things breaking and changing that slipping this particular rug out
from underneath those who haven't updated in a while seems completely
unproblematic.

Most importantly, it's good to ensure that migration code like this is
temporary, and it looks like it's time for it to go.

On 2011/01/22 06:27:58, Nico wrote:
> stevenjb: I think we can just delete the ChromeOS migration code. It's
> been there for a while, and stuff should be migrated for everyone who
> cared about these settings. Do you think that will help?
> 
> On Fri, Jan 21, 2011 at 10:02 PM,  <mailto:stevenjb@chromium.org> wrote:
> > Unfortunately in ChromeOS, this change creates the following problem which
> > breaks all of the ChromeOS browser_tests and ui_tests. (See
> >
>
http://build.chromium.org/p/chromium/builders/Linux%2520Builder%2520%2528Chro...).
> >
> > about_flags::ConvertFlagsToSwitches() calls
> > about_flags::MigrateChromeOSLabsPrefs() which expects a bunch of preferences
> > to
> > be set.
> >
> > These preferences are not actually set until ProfileImpl::GetPrefs() gets
> > called
> > from ProfileManager::GetDefaultProfile () (the first occurrence for ChromeOS
> > being from g_browser_process->profile_manager()->GetDefaultProfile() called
> > in
> > browser_main.cc).
> >
> > I'm not sure whether we need to call GetDefaultProfile() earlier, call
> > ConvertFlagsToSwitches() later, or something more complicated.
> >
> >
> > http://codereview.chromium.org/6267009/
> >