Re: [PATCH] diffcore: add a filter to find a specific blob

[Jonathan Nieder <jrnieder@gmail.com>]

Hi,

Stefan Beller wrote:
> On Thu, Dec 14, 2017 at 1:22 PM, Jonathan Nieder <jrnieder@gmail.com> wrote:

>> - what about mode changes?  If the file became executable but the
>>   blob content didn't change, does that commit match?
>
> ./git log --find-object=$(git rev-parse ba67504f:t/perf/p3400-rebase.sh)
>
> claims it does find the mode change (commit ba67504f is just a mode
> change)

Thanks.  Reminder to self to add a test + docs about that (as a followup
change; this isn't a complaint about the patch).

>> - are copies and renames shown (if I am passing -M -C)?
>
> It restricts the commits shown, not the renamed files. But I assume
> you mean it the same way as with mode changes.
> I did not find a good commit in gits history to demonstrate, but as
> it is orthogonal to the object id restrictions, I would think it works

Ok, will add test + doc.

>> Nit, not related to this change: it would be nice to have a long
>> option to go along with the short name '-t' --- e.g. --include-trees.
>
> follow up patches welcome. :)

Will think more and try to send a patch if it still seems like a good
idea in a day or so. ;)

>> Another nit: s/gitlink entry/submodule commit/, perhaps.  The commit
>> object is not a gitlink entry: it is pointed to by a gitlink entry.
>
> Well, what if the user doesn't have a submodule, but uses gitlinks
> for other purposes? We do inspect the gitlink, so it is correct IMHO.

It's a language nit.  The argument to --find-object is a commit object
name, not a gitlink entry.  A gitlink entry looks like

 160000 <path> <object>

>> Another documentation idea: it may be nice to point out that this
>> is only about the preimage and postimage submodule commit and that
>> it doesn't look at the history in between.
>
> That is sensible. One might be tempted to ask: "Which superproject
> commit contains a submodule pointer, that has commit $X in the
> submodule history?", but this new option is totally not answering this.

Ok, will try to come up with wording.

>>>                                                          The
>>> reason why these commits both occur prior to v2.0.0 are evil
>>> merges that are not found using this new mechanism.
>>
>> Would it be worth the doc mentioning that this doesn't look at merges
>> unless you use one of the -m/-c/--cc options, or does that go without
>> saying?
>
> I assumed it goes without saying, just like the lacking -t could mean
> to ignore trees. ;)

I suspect it's worth a mention, based on the discussion in this thread
(i.e. without such docs it was non-obvious and took some time to
diagnose).

[...]
>>> +--find-object=<object-id>::
>>> +     Restrict the output such that one side of the diff
>>> +     matches the given object id. The object can be a blob,
>>> +     gitlink entry or tree (when `-t` is given).
>>
>> I like this name --find-object more than --blobfind!  I am not sure it
>> quite matches what the user is looking for, though.  We are not
>> looking for all occurences of the object; we only care about when the
>> object appears (was added or removed) in the diff.
>
> Thanks! Yes, but the 'edges' are so few commits that a further walk
> will reveal all you need to know?

Sorry for the lack of clarity: I actually like this behavior *more*
than a "find trees pointing to object" behavior.  I'm just saying the
name sets an unclear expectation.

[...]
> Regarding finding a better name, I would want to hear from others,
> I am happy with --find-object, though I can see --pickaxe-object
> or --object--filter to be a good narrative as well.

Drat, I was hoping for an opinion.

Based on the answers above about mode changes and renames, at the
moment --object-filter seems clearest to me.

Thanks,
Jonathan