On Fri, Mar 14, 2014 at 9:04 PM, John Chilton <jmchilton@gmail.com> wrote:
On Fri, Mar 14, 2014 at 10:24 AM, Peter Cock <p.j.a.cock@googlemail.com> wrote:
Thanks John,
I suggest making this test framework perform this check by default (the twill and API based frameworks) and seeing what - if anything - breaks as a result on the Test Tool Shed.
Hey Peter,
I hope it is okay, but I do not want to make this change to the Twill driven variant of tool tests, I consider that code at end of life - new development would be a waste I think.
Running all tools against a modified environment that switched all tests to target the APIs would be nice, but it sounds like there is not really the infrastructure in place for doing that right now.
Upon further consideration I am not sure there are really any backward compatibility concerns anyway - or at least no more so than anything else when switching over to the API driven tests. I'll let the pull request sit open for a few days and then merge it as is.
Note that one area of fuzziness is subclasses, e.g. if the tool output was labelled "fastqsanger", but the <test> just said "fastq", I would say the test was broken. On the other hand, if the <test> used a specific datatype like "fastqsanger" but the tool produced a dataset tagged with a more generic datatype like "fastq" I think that is a again a real failure.
100% agreed on both points. I believe the implementation proposed in pull request #347 reflects this resolution of the "fuzziness".
Thanks and have a great weekend, -John
That sounds like a plan :) Hmm. I wonder if it would be trivial to tweak our TravisCI setup to run the functional tests twice, once with the old twill framework and once with the new API based framework? Seems doable (but would up the run time quite a bit). Thanks, Peter