Re: Shepherd changes

On May 20, 2013, at 9:13 AM, Gérard Talbot wrote:

> 
> Le Lun 20 mai 2013 0:15, Peter Linss a écrit :
>> Today's upgrade to Shepherd introduced two significant changes.
>> 
>> First, Shepherd now tracks test files on a more individual basis, rather
>> than aggregating all files with the same file name.
> 
> 
> Tests, reftests and support files with same file name is still a problem...

That depends. If they are in the same test suite, then yes (although I may deal with it in the build system at some point), if the tests are not in the same test suite, then it's really not so much of an issue.

> 
> Also many tests from TTWF events do not follow, do not comply with the
> filename template.

True, but that's a separate problem. I'm going to be reworking how Shepherd tracks issues in the near future, once that's done I'll add automatic issues about file name format.

> 
> 
>> Multiple files with the same name _and the same content_ are still
>> aggregated as before, but when the file contents differ, they are
>> treated as separate testing assets (actually as separate branches of the
>> same asset, but still unique tests/references). The distinction is
>> subtle, and most users wont notice a difference, but in the cases where
>> name collisions have happened in the repository you'll now see the files
>> as distinct assets with their own page (and links to the other versions
>> of the files).
> 
> 
> When I visit
> http://test.csswg.org/shepherd/testcase/border-bottom-color-052/
> and then click on its associated reftest
> http://test.csswg.org/shepherd/reference/border-bottom-color-052-ref:-/#revision-4779-7267e029eea2
> I end up into a Shepherd file about a retracted version of a reftest
> which is not loadable.

Sorry, small bug in the code that computes reference links, fixed now. (In this case there was a deleted file with the same path as the current reference, it was pointing to the deleted version instead of the current version.)

As a side effect of the recent changes, Shepherd is much more aggressive about how it computes links between tests and references, the link will now be to the specific file referenced, not just the asset of that name, which is necessary when there are name collisions in the reference. It also reports more errors about broken reference links.

> 
> 
> I do not understand why there is a Shepherd file about a retracted
> reftest which can not be examined or viewed anyway...

Shepherd always maintains history of all files, even after they're deleted form the repository. You can still review the issue history and there are links to the previous versions in the repository. 


Peter

Received on Monday, 20 May 2013 17:16:36 UTC