From: Derrick Stolee <stolee@gmail.com>
Date: Wed, 30 May 2018 19:18:24 GMT
Subject: Re: [RFC PATCH 00/35] object-store: lookup_commit
Message-ID: <6e18e6bb-cc11-a1b4-becb-4115df124284@gmail.com>
In-Reply-To: <CAGZ79kaEHhE=suT2-Rzrbu2P7z8KPzRdCOFNCTqsOiV0+HjJ9A@mail.gmail.com>

On 5/29/2018 11:18 PM, Stefan Beller wrote:
> On Tue, May 29, 2018 at 6:05 PM, Derrick Stolee <stolee@gmail.com> wrote:
>> On 5/29/2018 8:47 PM, Stefan Beller wrote:
>>> This applies on the merge of nd/commit-util-to-slab and
>>> sb/object-store-grafts,
>>> and is available at http://github.com/stefanbeller/ as branch
>>> object-store-lookup-commit
>>> as the merge has some merge conflicts as well as syntactical conflicts
>>> (upload-pack.c
>>> and fetch-pack.c introduce new calls of functions that would want to take
>>> a repository struct
>>> in the object-store-grafts series)
>>>
>>> As layed out in
>>> https://public-inbox.org/git/20180517225154.9200-1-sbeller@google.com/
>>> this is getting close to finishing the set of object store series though
>>> the last
>>> unfinished part of this RFC hints at new work on the plate:
>>> * To give this series a nice polish, we'd want to convert parse_commit,
>>> too.
>>>     But that requires the conversion of the new commit graph. Maybe we need
>>>     to split this series into 2.
>>
>> I'll take a look at this series tomorrow. I've been working in
>> ds/commit-graph-fsck to make many of the methods take a 'struct commit_graph
>> *' parameter, which could easily be 'r->commit_graph' for a 'struct
>> the_repository *r' instead of the global 'commit_graph'. Those graph-local
>> methods will make the transformation to be repo-local a lot easier. (There
>> still may be some need to insert a repository, though.)
>>
>> Since you are working on the commit-slab stuff in this patch, I'll (continue
>> to) delay my patch series on using the commit-slab for generation numbers to
>> avoid collisions with your work.
> This series touches the commit slabs in the last few commits only.
> I don't think there will be huge conflicts, as all I had to do was
> move the definition from a global to inside the parsed object store.

I went through the whole series and didn't find any issue with the code. 
The commit messages reference coccinelle scripts that I never saw, so 
those should be struck from the messages.

If anyone else plans to review, these two commits are worth special 
attention compared to the others:

[PATCH 27/35] commit-slabs: remove realloc counter outside of slab struct
[PATCH 28/35] commit.c: migrate the commit buffer to the parsed object store

I found no issue with them, but they do actually change behavior slightly as opposed to the other commits which have no change other than to relax the dependence on the_repository as a global. I also like the pattern of storing the slabs inside the parsed object structure in the_repository. I'll use a similar pattern when looking into using slabs for generation number.

For Junio: I know this will conflict with ds/commit-graph-fsck (specifically, I add new references to lookup_commit() and manipulate parse_commit_gently()). If this series is ready to be queued quickly, I can rebase my v4 onto these commits as they are placed into 'pu' (and please eject ds/commit-graph-fsck at that time).

Thanks,
-Stolee