Failing tests merging bzr.dev to brisbane-core

[John Arbash Meinel]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Andrew Bennetts wrote:
> John Arbash Meinel wrote:
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA1
>>
>> I just merged bzr.dev 4157 into brisbane-core, and now we have some
>> "per_repository" tests that are failing. Specifically:
>>
>> ERROR:
>> per_repository.test_fetch.TestFetchSameRepository.test_fetch_to_knit3(RemoteRepositoryFormat-default)
>>     'RemoteRepository' object has no attribute
>> '_eliminate_revisions_not_present'
>>
>> My guess is this is caused by one of these revisions:
>>
>>  4153 Canonical.com Patch Queue Manager 2009-03-17 [merge]
>>       (andrew) Remove InterRemoteToOther and InterOtherToRemote.
>>
>>  4154 Canonical.com Patch Queue Manager 2009-03-17 [merge]
>>       (andrew) Remove InterPackToRemotePack too.
>>
>> And having brisbane-core having more/different scenarios. I can dig into
>> this, but I was hoping that Andrew might have a bit more context to do so.
> 
> Well, those revisions exposed the problem.  But the real cause is actually:
> 
>     john at arbash-meinel.com-20090311225935-p6b5tr1u0vwcg7gj
>     
>     Some work on rich-root support.
>     
>     Try to get InterDifferingSerializer to support upcasting to rich-root without having
>     to double handle the inventories.
> 
> Which makes InterDifferingSerializer always return True.
> 
> Uncommenting the code you commented out makes the test_fetch_to_knit3 tests all
> pass again.
> 
> Note that the fetch code itself now notices if it's upcasting to rich-root, and
> if so the StreamSource will generate root texts automatically (and generate the
> stream in topological order).  So we could perhaps look at removing
> InterDifferingSerializer altogether.  Currently removing it breaks
> test_fetch_to_knit3(RepositoryFormatPackDevelopment5Subtree) due to xml8's
> _unpack_entry having no entry_cache, but the other test_fetch_to_knit3 scenarios
> pass.
> 
> -Andrew.
> 
> 

So converting to GC format repos really needs to batch by some amount so
that autopack can resolve the overly-expanded texts. Not to mention it
is *really* nice to be able to stop a conversion and start it again (the
standard fetch code is always all-or-nothing).

So if you prefer, we can add a "isinstance(..., RemoteRepository)"
rather than the rich-root flags. But I really *do* want to keep
InterDifferingSerializer and have it support upcasting to rich-root
data. (All the gcchk formats are going to be rich-root as well as
'subtree'.)

John
=:->
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAknAKZAACgkQJdeBCYSNAANOHACdEsKydqf7bvTfD4xE3gDgDk0V
c0oAnAmAPtQbFSaNAuWfkQSfG3cSZaPI
=xF+9
-----END PGP SIGNATURE-----