On 11/10/2010 07:16 PM, William Hubbs wrote:
> On Mon, Nov 08, 2010 at 10:05:17PM +0200, Petteri R??ty wrote:
>> On 11/08/2010 06:17 AM, Donnie Berkholz wrote:
>>> On 16:42 Sun 07 Nov     , Petteri R??ty wrote:
>>>> On 11/06/2010 11:22 AM, Krzysztof Pawlik wrote:
>>>>> Hello,
>>>>>
>>>>> I'm sending this patch for discussion, what it changes? The change is to where
>>>>> the final clone of repository will be placed, it used to be ${WORKDIR}/${module}
>>>>> (where module usually is the last component of source URI) to ${WORKDIR}/${P}
>>>>> (essentially ${S}). This has few effects:
>>>>>  - ebuilds using mercurial.eclass don't need to set S any longer
>>>>>  - mercurial.eclass behaves more like git.eclass
>>>>>  - it breaks all existing ebuilds using this eclass
>>>>
>>>> Which means that the doing the chance is not allowed as eclasses must
>>>> remain backwards compatible.
>>>
>>> Is that really still the case now that full environments have been saved 
>>> for a number of years? Clearly breaking things is unacceptable, but I 
>>> could envision someone simultaneously updating the eclass and all 
>>> consumers.
>>>
>>
>> There's no full environment saved before the package is installed and I
>> don't see why we should break overlays.
> 
> I didn't think we had to care about overlays since they aren't the main
> tree?  After all, how can we be sure what is happening in all overlays
> out there?
> 
> I could see this, like Donnie says, if he wasn't going to fix all of the
> ebuilds.  However I don't see a problem with it since he said he is
> going to fix all of the ebuilds.
> 

If there are options that don't require breaking with no big downsides
then we should rather go with them. There usually are.

Regards,
Petteri