[MERGE] Fix RemoteRepository.get_parent_map stacking

[Aaron Bentley]

John Arbash Meinel wrote:
> Aaron Bentley wrote:
>> Are you saying you want an effort test?  I think this is overkill for a
>> bugfix.  I don't mind adding a direct test that
>> _get_parent_map(['rev1']) returns nothing.
> 
> That is what I'm asking for.

Done.

> +    * RemoteRepository.get_parent_map now uses fallback repositories.
> +      (Aaron Bentley)
> +
> 
> ^- Isn't there a bug number associated with this?

Not firmly.  There are two bugs that *may* be fixed by it:
    * RemoteRepository.get_parent_map now uses fallback repositories.
      (Aaron Bentley, #297991?, #293679?)


> graph._StackedParentsProvider(providers)
> +        return self._parents_provider
> 
> ^- I sort of like that it is cached, but I'm concerned about someone
> mutating self._fallback_repositories that doesn't know to reset
> self._parents_provider.

I thought it was an unlikely case, so doing what you asked was an
improvement over doing nothing,

> I agree that this is necessary to get stacked branches to function
> properly, and we should consider getting this merged and getting a 1.9.1
> with it as well. (Did you happen to build your changes against 1.9 to
> start with?)

Unfortunately not.

> You can chose whether you want to track down when you need to get rid of
> the current "self._parents_provider" or just not cache the value. I'm
> leaning toward the latter as the quick fix.

As a quick fix, I've gone with not caching the value.

Aaron