[Users] Spurious test failures due to now bboxset code in Carpet

[Erik Schnetter]

On Jan 20, 2015, at 13:24 , Ian Hinder <ian.hinder at aei.mpg.de> wrote:
> 
> On 20 Jan 2015, at 19:18, Erik Schnetter <schnetter at gmail.com> wrote:
> 
>> We enabled a new, much scalable implementation of the bboxset class in Carpet. This class describes sets of "bounding boxes", i.e. the locations and shapes of the components of Carpet's grids. (This new implementation requires C++11 features, and is thus only available by default if the C++ compiler supports the C++11 standard.)
>> 
>> As it turns out, the new class numbers these components differently than the old code. This numbering scheme is ultimately arbitrary; if there are two refined regions, then it does not matter for physics which one is labelled #0 and which one is labelled #1. However, since the old and the new code differ, some ASCII output will now differ. In the Einstein Toolkit test cases, this is triggered by the test ml-gw1d-small-amr of PeriodicCarpet. Inspection shows that, apart from this re-labelling, the test output is identical.
>> 
>> The new bboxset code has a defined ordering of the bboxes; they are ordered lexicographically by their upper left corner. (This is implemented in the "less" functions in bbox.hh and vect.hh, following STL conventions.) The old code used no particular ordering, although the ordering was guaranteed to remain the same between runs.
>> 
>> I suggest to ignore the test case failures caused by this transition, and to update this (and potentially other) test cases.
>> 
>> I plan to update the old bboxset code to explicitly sort bboxes in the same order.
> 
> Hi Erik,
> 
> We recently put some work into making the ASCII output from the tests the same independent of the number of processes that they are run on.  This used the fact that the output ordering of components from Carpet was the same under certain conditions.  This was using the old bboxset class.  Do you know if the new bboxset class has the same property?

Ian

Yes, this property does not change. What changes is the order in which the bboxes are output. They are now "more sorted" than before.

The test case in question does not use the reduced output format, so that it is still dependent on the number of processes.

-erik

--
Erik Schnetter <schnetter at gmail.com>
http://www.perimeterinstitute.ca/personal/eschnetter/

My email is as private as my paper mail. I therefore support encrypting
and signing email messages. Get my PGP key from https://sks-keyservers.net.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 203 bytes
Desc: Message signed with OpenPGP using GPGMail
Url : http://lists.einsteintoolkit.org/pipermail/users/attachments/20150120/b8f912e0/attachment.bin