[Users] McLachlan-rewrite has landed

[Ian Hinder]

On 24 Jul 2015, at 19:24, Erik Schnetter <schnetter at cct.lsu.edu> wrote:

> On Fri, Jul 24, 2015 at 1:21 PM, Erik Schnetter <schnetter at cct.lsu.edu> wrote:
> On Fri, Jul 24, 2015 at 12:04 PM, Ian Hinder <ian.hinder at aei.mpg.de> wrote:
> 
> On 24 Jul 2015, at 17:44, Erik Schnetter <schnetter at cct.lsu.edu> wrote:
> 
>> On Fri, Jul 24, 2015 at 11:23 AM, Ian Hinder <ian.hinder at aei.mpg.de> wrote:
>> 
>> On 24 Jul 2015, at 17:04, Ian Hinder <ian.hinder at aei.mpg.de> wrote:
>> 
>>> 
>>> On 24 Jul 2015, at 16:59, Ian Hinder <ian.hinder at aei.mpg.de> wrote:
>>> 
>>>> 
>>>> On 24 Jul 2015, at 14:13, Erik Schnetter <schnetter at cct.lsu.edu> wrote:
>>>> 
>>>>> I just pushed all the changes related to the McLachlan-rewrite merge. Apart from the McLachlan Kranc scripts and the respective generated code, there are a few thorns that needed to have test results updated.
>>>>> 
>>>>> I see zero test failures with these changes on Bethe, Blue Waters, Carver, Datura, Edison, Hopper, Mike, Nvidia (a Perimeter workstation), Philip, Queen Bee, Shelob, and Stampede.
>>>> 
>>>> Hi Erik,
>>>> 
>>>> It looks like there are some issues with backward compatibility.  I get this error:
>>>> 
>>>> WARNING[L2,P0] (Cactus): ParameterSetKeyword: Unable to set keyword 'ML_BSSN::my_rhs_boundary_condition', new value 'static' is not in any active range
>>>> WARNING[L1,P0] (Cactus): Major error in parameter file '/lustre/datura/ianhin/simulations/einsteintoolkit/rewrite_ML_BSSN_bench_o4/output-0000/ML_BSSN_bench.par' line 30: Range error setting parameter 'ML_BSSN::my_rhs_boundary_condition' to 'static'
>>> 
>>> Also, any output parameters referring to the old variables are now invalid; e.g. phi has been renamed phiW.  Given this, it looks like old parameter files will very likely all have to be modified for the new version.
>> 
>> Hi Erik,
>> 
>> Something is wrong with the merge: the merge commit should have two parents, but it only has one.  
>> 
>> 	https://bitbucket.org/einsteintoolkit/mclachlan/commits/all
>> 
>> It looks like the "merge" is just a set of changes on the master branch.
>> 
>> I merged the master branch into the rewrite branch, and the result of this then back into the master. The former was non-trivial because of other changes that happened to the master in the mean time, including regenerating code, so I introduced a temporary branch "integrate-rewrite" for this. I didn't look at the parent structures of the commits.
>> 
>> I now see that integrate-rewrite doesn't have rewrite as parent. Very strange. I only used SourceTree, and only used standard mechanisms (merging, branching, committing).
> 
> (summarising IM conversation with Erik)
> 
> It looks like some functionality from master was lost in the "merge".  e.g. the ability to run the tests on any number of processes.  This suggests to me that other changes may also be missing.
> 
> Leaving out this ability was on purpose to simplify the merge. In particular, this means that the commit that renamed "phi" to "phiW" could simultaneously rename the respective output files. Completely changing the output format requires re-generating all test results, making it impossible to track whether the code has an error. It is easy to do so after the fact -- I'll do that right now.
> 
> If you find other functionality that is missing, let me know. I'm not aware of others, except as listed on the wiki.
> 
> Was this functionality ever enabled? I see a very prominent "NPROCS 2" in ML_BSSN_Test.

That is for the multipatch test only, which actually requires Llama, which is not in the ET.  Since I couldn't test it with the configuration I was using at the time, I didn't change it.  But all other tests should have no NPROCS limit.

test.ccl has:

> 
> TEST ML_BSSN_MP_O8_bh
> {
>   NPROCS 2
>   ABSTOL 1e-11
> }
> 
> TEST ML_BSSN_NewRad
> {
>   ABSTOL 1e-11
> }
> 
> TEST ML_BSSN_MP_O8_bh
> {
>   ABSTOL 1e-10
> }


and the last commit to this file was

> commit 89adf9faf2d6d10f9cddb57bc91f7e8ad068b190
> Author: Ian Hinder <ian.hinder at aei.mpg.de>
> Date:   Thu Nov 27 18:33:19 2014 +0100
> 
>     Allow most tests to run on any number of processes
>     
>     The multipatch test is excluded, as I don't have multipatch in this
>     Cactus tree.



-- 
Ian Hinder
http://members.aei.mpg.de/ianhin

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.einsteintoolkit.org/pipermail/users/attachments/20150724/ec140d5e/attachment.html