Re: [jgit-dev] fixing bug 321502 - lost in polymorphia

[Shawn Pearce <spearce@xxxxxxxxxxx>]

On Fri, Aug 13, 2010 at 9:05 AM, Matthias Sohn
<matthias.sohn@xxxxxxxxxxxxxx> wrote:
> http://egit.eclipse.org/r/#change,1299 doesn't fix bug 321502
> (Shawn happened to accept the change when I concurrently tried to vote -1).

Well, to be fair the change was still reasonable cleanup.  It just
didn't fix the bug like we expected it would.  :-(

> We found this doesn't work since looking up the list of Refs to be shown in
> the
> history view implemented in PlotWalk.getTags(AnyObjectId) fails. We found
> that
> the failing hashmap lookup ends up calling RevObject.equals() which does a
> pointer comparison obviously failing to match the SWTCommit passed into
> PlotWalk.getTags(AnyObjectId) with the keys of the hashmap returned by
> Repository.getAllRefsByPeeledObjectId().
>
> Any hints how to fix this ? We are lost in polymorphia ...

Ooooh.  This is a common problem for us.  The objects coming back from
Repository are straight-up ObjectId, but the HashMap is using the key
that is passed to get() to implement equals(), which means the
overridden equals() from RevObject wins here.

Three solutions:

1)  When looking up an object in the map, use .copy() to convert from
RevObject to ObjectId.  This actually works by allocating a new
ObjectId with the same identity.  So its somewhat inefficient due to
the lookup cost.

2)  Use an ObjectIdSubclassMap rather than straight Map for the
returned map from the Repository.  The ObjectIdSubclassMap knows how
to use our own equality function for ObjectId without falling into the
trap of the RevObject override of .equals().

3)  We admit that RevObject overridding .equals() is wrong and back
out that implementation.

-- 
Shawn.