Re: [PATCH v2 26/33] diff-merges: let new options enable diff without -p

[Sergey Organov <sorganov@gmail.com>]

Elijah Newren <newren@gmail.com> writes:

> On Fri, Dec 18, 2020 at 6:42 AM Sergey Organov <sorganov@gmail.com> wrote:
>>
>> Elijah Newren <newren@gmail.com> writes:
>>
>> > On Wed, Dec 16, 2020 at 10:50 AM Sergey Organov <sorganov@gmail.com> wrote:

[...]

>> >> diff --git a/log-tree.c b/log-tree.c
>> >> index f9385b1dae6f..67060492ca0a 100644
>> >> --- a/log-tree.c
>> >> +++ b/log-tree.c
>> >> @@ -899,15 +899,21 @@ static int log_tree_diff(struct rev_info *opt, struct commit *commit, struct log
>> >>         int showed_log;
>> >>         struct commit_list *parents;
>> >>         struct object_id *oid;
>> >> +       int is_merge;
>> >> +       int regulars_need_diff = opt->diff || opt->diffopt.flags.exit_with_status;
>> >
>> > So rev_info.diff has changed in meaning from
>> > commits-need-to-show-a-diff, to non-merge-commits-need-to-show-a-diff.
>> > That's a somewhat subtle semantic shift.  Perhaps it's worth adding a
>> > comment to the declaration of rev_info.diff to highlight this?  (And
>> > perhaps even rename the flag?)
>>
>> No, the meaning of rev_info.diff hopefully didn't change. rev_info.diff
>> still enables all the commits to pass further once set. It is still
>> exactly the same old condition, just assigned to a variable for reuse.
>> My aim was to avoid touching existing logic of this function and only
>> add a new functionality when opt->merges_need_diff is set.
>>
>> It looks like I rather choose confusing name for the variable, and it'd
>> be more clear if I'd call this, say:
>>
>>   int need_diff = opt->diff || opt->diffopt.flags.exit_with_status;
>>
>> ?
>>
>> What do you think?
>
> I think need_diff would actually be confusing.  It can be false when
> you need diffs (e.g. --diff-merges=cc with no -p, because then you'd
> need diffs for merge commits and not for non-merge commits).  I'd
> stick with your original local variable name.
>
> Perhaps opt->diff hasn't changed meaning and I just had a wrong mental
> model in my head for what it meant, but even then what seems like its
> obvious purpose given its name is mismatched with what it actually
> does.  Since you are already changing struct rev_info in this series,
> this was more a note that a name change or at least a comment for
> opt->diff might be useful.  I mean, you asked a couple times on the
> previous series for help trying to understand it, and I could only
> offer some flailing guesses and Junio responded with a couple bits of
> history.  Clearly, it isn't very clear and this patch reminded me of
> that and made me wonder if we're possibly making it a little harder
> for others further down the road to figure out.

I still don't see why opt->diff is needed in the first place, and
second, why opt->diffopt.flags.exit_with_status check is here? Why
whoever sets opt->diffopt.flags.exit_with_status doesn't just set
opt->diff as well (provided opt->diff is needed in the first place)?

From the aforementioned discussion it looks like opt->diff is an
optimization (maybe a remnant from the times when diff was a separate
script), and that apparently there is some code instance somewhere that
actually relies on the fact that clearing opt->diff is enough to disable
diff machinery (as followed from my experiment of removing the check
altogether and then getting only single seemingly unrelated test
failing.)

Overall, neither have I any idea how to clarify this, nor do I want to
bother in this patch series. It'd be nice though if somebody who really
does understand diff machinery in Git does the job.

Thanks,
-- Sergey