Re: [PATCH 3/4 v6] cache-tree: subdirectory tests

From: Junio C Hamano <gitster@pobox.com>
Date: Jul 11, 2014, 15:40 UTC
Message-ID: <xmqq7g3j6gzr.fsf@gitster.dls.corp.google.com>
In-Reply-To: <xmqqbnsv6hmc.fsf@gitster.dls.corp.google.com>

Junio C Hamano <gitster@pobox.com> writes:

Show 6 quoted lines

>>> +       sed -n -e "s/[0-9]* subtrees//" -e '/#(ref)/d' -e '/^invalid /p' >actual &&
>
> Is the second one to remove "#(ref)", which appears for a good
> "reference" cache tree entry shown for comparison, necessary?  Do
> they ever begin with "invalid"?  If they ever begin with "invalid"
> that itself may even be a noteworthy breakage to catch, no?

Answering to myself...

Because test-dump-cache-tree uses DRY_RUN to create only an in-core copy of tree object, and we notice that the reference cache-tree created in the tests contains the object name of a tree that does not yet exist in the object database. We get "invalid #(ref)" for such node.

In the ideal world, I think whoever tries to compare two cache-trees (i.e. test-dump-cache-tree) should *not* care, because we are merely trying to show what the correct tree object name for the node would be, but this is only for testing, so the best way forward would be to:

 - Stop using DRY_RUN in test-dump-cache-tree.c;

 - Stop the code to support DRY_RUN from cache-tree.c (nobody but
   the test uses it); and

 - Drop the "-e '#(ref)/d'" from the above.

I would think.