Re: Bug: git log: boundary commits do not respect order (e.g. date-order, topo-order) 2

[Josh Tepper <josh@clarifai.com>]

I just wanted to follow up -- what do you think?

On Thu, Feb 22, 2018 at 9:21 PM, Josh Tepper <josh@clarifai.com> wrote:
> My use case is that when combined with --graph, the ordering is
> necessary to render a reasonable looking commit graph; by placing the
> commits at the end, each boundary commit essentially produces another
> vertical line all the way to the bottom where the commits reside.
> More specifically, the case where I noticed this was:
>
> git log --boundary --graph ^master feature
>
> which will have one vertical line going all the way to the bottom for
> each merge from master into feature.
>
> Another issue that I've noticed is that if one is using
> --show-linear-break (instead of --graph), the linear breaks are
> missing between the boundary commits and the other commits.
>
> Regarding the documentation, while it doesn't explicitly say where the
> boundary commits will be ordered with the other commits (nor does it
> say that they'll have accurate linear breaks), the documentation for
> the order flags (and for the linear break flag) makes no mention that
> certain commits are excluded.  The implicit understanding of the
> documentation for --date-order is that it will apply to all of the
> commits.  Certainly, if you add the flag --full-history, one expects
> the ordering to apply to the additionally traversed commits.  I
> suppose boundary commits are a little different since they're not
> explicitly part of the range -- nonetheless I expected them to be
> treated similarly.
>
> Maybe this could at least be documented?
>
> Best,
> ~Josh
>
> On Thu, Feb 22, 2018 at 2:29 PM, Derrick Stolee <stolee@gmail.com> wrote:
>> On 2/21/2018 6:57 PM, Josh Tepper wrote:
>>>
>>> When using git log, boundary commits (ie, those commits added by
>>> specifying --boundary) do not respect the order (e.g., --date-order,
>>> --topo-order).  Consider the following commit history, where number
>>> indicates the order of the commit timestamps:
>>>
>>> <view with a fixed with font! 3's ancestor is 1, 6's ancestors are 4,5>
>>> 0----1----2----5  <--A
>>>         \         \
>>>           3----4----6  <--B
>>>
>>>
>>> Executing the following command:
>>>
>>> $ git log --boundary --date-order ^A B
>>>
>>> Should produce the following order (boundary commits shown with dashes):
>>> 6 -5 4 3 -1
>>>
>>> However, it in fact produces:
>>> 6 4 3 -5 -1
>>>
>>> Please advise.
>>>
>>
>> Hi Josh,
>>
>> Looking at the docs [1], I don't see any specifics on how the boundary
>> commits should be ordered.
>>
>> Clearly, the implementation specifies that the boundary is written after all
>> other commits. For a full discussion of this, see the commit message for
>> 86ab4906a7c "revision walker: Fix --boundary when limited". Here is an
>> excerpt:
>>
>>      - After get_revision() finishes giving out all the positive
>>        commits, if we are doing the boundary processing, we look at
>>        the parents that we marked as potential boundaries earlier,
>>        see if they are really boundaries, and give them out.
>>
>> The boundary commits are correctly sorted by topo-order among themselves as
>> of commit 4603ec0f960 "get_revision(): honor the topo_order flag for
>> boundary commits".
>>
>> So, I'm not sure that this is a bug (it is working "as designed") but it
>> certainly is non-obvious behavior.
>>
>> In what use case do you need these boundary commits to appear earlier?
>>
>> Thanks,
>> -Stolee
>>
>> [1] https://git-scm.com/docs/git-log
>>
>>