Obsolete Markers and Phases

[Patrick Mézard]

Le 02/08/12 17:30, v a écrit :
> 
> Patrick Mézard wrote
>>
>> Le 02/08/12 16:47, Scott Palmer a écrit :
>>>
>>> On 2012-08-02, at 10:29 AM, Pierre-Yves David <pierre-yves.david@>
>>> wrote:
>>>
>>>> On Thu, Aug 02, 2012 at 07:26:21AM -0700, v wrote:
>>>>>
>>>>> Pierre-Yves David wrote
>>>>>>
>>>>>> On Thu, Aug 02, 2012 at 07:20:53AM -0700, v wrote:
>>>>>>> http://mercurial.selenic.com/wiki/ChangesetsObsolescence and hg help
>>>>>>> say
>>>>>>> that
>>>>>>> obsolescence can't/shouldn't be used for public changesets.
>>>>>>>
>>>>>>> Isn't this the whole point?
>>>>>>
>>>>>> Golden rule:
>>>>>>
>>>>>>    The whole point of **public** changeset is to be immutable.
>>>>>>
>>>>>
>>>>> Does obsolescence mutate changesets?
>>>>
>>>> Obsolescence marker track changesets mutation.
>>>>
>>>> Public changeset will refuse to be Mutated.
>>>>
>>>> Note: Mutation of a changeset is actually the writting a whole new
>>>> changeset.
>>>>      As per DVCS design.
>>>
>>> Can you give us a typical use case?  I'm getting confused by this.  Since
>>> the introduction of "public" vs. "private" change sets, I thought
>>> rewriting history by actually mutating (or replacing) a changeset was now
>>> safe - because it was just a correction or alteration of local data that
>>> was not shared yet.
>>
>> Phase helps you not mutate changesets which have been published.
>>
>>
>> Obsolete helps you describe how non-public changesets have been rewritten.
>> You will not directly use this information but rewriting tools will. The
>> middle of this post has an example of using evolve extension to rewrite a
>> changeset in the middle of a changeset sequence:
>>
>>   http://selenic.com/pipermail/mercurial-devel/2012-July/042948.html
>>
>> Note, the "stabilize" call (similar to mq qpush for instance) use 3-way
>> merge instead of patching since the changeset being rewritten is not
>> destroyed but marked obsolete.
>>
>> Obsolescence markers can also be used in collaborative settings where
>> people work on the same sets of changes. See the end of the post
>> referenced above for more details about this.
>>
>> Also note, that all this stuff is still experimental :-)
>>
> 
> What would go wrong if obsolete was used with public changesets?

Nothing would go wrong as it can happen in a distributed setup: A publish revision REV while B marks it obsolete.

The marker has no effect on the public changeset. It will be taken in account to resolve conflicts with the non-public things. For instance B could have rewritten REV into REV2. In that case, he would have to rebase REV2 on top of REV.

--
Patrick Mézard