Handle sync values of wrong type more graceful

Handle sync values of wrong type more graceful.

Clear prefs that the server reports TYPE_NULL values for and log a warning if the type is off otherwise.

BUG=70823
TEST=compiles and passes tests

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=72786

[Mattias Nissler (ping if slow), 2011-01-26 17:49:39]

Please review. This avoids a DCHECK on debug builds, but doesn't change
semantics for release builds.

[albertb, 2011-01-26 18:09:05]

Do you mean that some of your preferences are stored as null on the sync server?
If that's the case, it seems like it should be possible to go from null to
proper type without DCHECKing. Otherwise, those preferences would become
out-of-sync with no way of ever syncing back...

Shouldn't PrefService::Set allow a pref to go from TYPE_NULL to TYPE_LIST or
TYPE_DICTIONARY?

[Mattias Nissler (ping if slow), 2011-01-26 18:27:50]

On 2011/01/26 18:09:05, albertb wrote:
> Do you mean that some of your preferences are stored as null on the sync
server?

Yes. To be more precise, they are stored as TYPE_NULL values.

> If that's the case, it seems like it should be possible to go from null to
> proper type without DCHECKing. Otherwise, those preferences would become
> out-of-sync with no way of ever syncing back...

When the client downloads these values and tries to apply them it now DCHECKs
(see below). But I don't see why we wouldn't be able to sync our actual value
back to the server?

> 
> Shouldn't PrefService::Set allow a pref to go from TYPE_NULL to TYPE_LIST or
> TYPE_DICTIONARY?

PrefService::Set() accepts only values of the type the pref was registered with.
Previously, you could also set TYPE_NULL values for lists and dictionaries to
reset them. Also, Get() would return TYPE_NULL values if these weren't set. So
if we ignore the TYPE_NULLs from the server now and upload new values later, we
should be fine, no?

[albertb, 2011-01-26 19:09:25]

As discussed over chat, it seems like whenever the server has a pref stored as
TYPE_NULL, the client should clear the local preference.

In the case where the server's and client's type really don't match (eg.
TYPE_LIST vs. TYPE_DICT) a DCHECK is probably the right behaviour.

[Mattias Nissler (ping if slow), 2011-01-26 19:17:02]

New version with the ClearPref()/warning approach.

[albertb, 2011-01-26 19:17:16]

LGTM!