[Freeipa-devel] [PATCH] 0330 - Add comment about last change to VERSION

[Petr Viktorin]

On 12/09/2013 02:50 PM, Martin Kosek wrote:
> On 12/09/2013 02:35 PM, Simo Sorce wrote:
>> On Mon, 2013-12-09 at 12:39 +0100, Martin Kosek wrote:
>>> On 12/09/2013 12:08 PM, Tomas Babej wrote:
>>>> On 12/05/2013 01:37 PM, Petr Viktorin wrote:
>>>>> Consider this scenario:
>>>>>
>>>>> - Nathaniel submits RADIUS patches that update the API version (from 2.69 to
>>>>> 2.70)
>>>>> - I have ACI patches that also bump the version (from 2.69 to 2.70)
>>>>> - Nathaniel's patches gets accepted
>>>>> - I rebase my ACI patches onto master. Git thinks that the 2.69->2.70 change
>>>>> is already done, so it leaves VERSION unchanged.
>>>>>
>>>>> I can solve this locally by telling Git to not merge VERSION automatically,
>>>>> but I think it would be helpful to add a unique comment to each change so
>>>>> that everyone gets a conflict cases like this.
>>>>> Do you agree?
>>>>>
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> Freeipa-devel mailing list
>>>>> Freeipa-devel at redhat.com
>>>>> https://www.redhat.com/mailman/listinfo/freeipa-devel
>>>>
>>>> Makes sense to me.
>>>>
>>>> I'd just add a comment so that the purpose of the last change comment is also
>>>> obvious for the new developer perusing the VERSION file.
>>>>
>>>> Maybe something along the lines of:
>>>>
>>>>   ########################################################
>>>>   IPA_API_VERSION_MAJOR=2
>>>>   IPA_API_VERSION_MINOR=70
>>>> +# Update the last change entry to enforce conflict on merging two independent
>>>> branches into master.
>>>> +# Last change: npmccallum - RADIUS support

I don't think this is necessary, IMO "Last change:" is enough as 
instructions.

>>> I spoke with Petr offline, to me it would make bigger sense if we just forbid
>>> automatic merging of this line on the git server side (if possible) instead of
>>> adding other arbitrary work to our development process.
>>>
>>> IIRC, Petr3 said it should be possible to do.
>>
>> Except it may not fix the issue, if someone does a rebase on his machine
>> and resubmit a patch to the list w/o noticing the change was effectively
>> dropped.
>>
>> Simo.
>
> I thought that the point of the anti-merge protection is to prevent git merging
> tools to prevent automatic rebase of this particular line and force manual
> merging, i.e. force increasing the number.

When the file is equal on both sides of the merge, Git just uses the 
common file, and doesn't consider it for merging.
So unfortunately, git attributes won't work here; there needs to be 
another change in the file.

I did say otherwise, sorry for that.

-- 
Petr³