On Tue, Feb 11, 2014 at 5:09 PM, Peter Cock <p.j.a.cock@googlemail.com> wrote:
On Tue, Feb 11, 2014 at 4:42 PM, Peter Cock <p.j.a.cock@googlemail.com> wrote:
On Tue, Feb 11, 2014 at 3:41 PM, Dave Bouvier <dave@bx.psu.edu> wrote:
Peter,
Thanks for pointing this out, I've applied your changes in 12478:15fc8675064e on the default branch.
--Dave B.
Thanks Dave,
RE: http://lists.bx.psu.edu/pipermail/galaxy-dev/2014-February/018154.html https://bitbucket.org/galaxy/galaxy-central/commits/15fc8675064ea46b7e081d96...
I'm still digging into the underlying problem, by focussing on just one unit test ...
As a workaround, I could probably provide expected data for all four output files...
Confirmed, providing a reference file for each output works: <test> <param name="job_type" value="genome" /> <param name="job_quality" value="accurate" /> <param name="type" value="none" /> <param name="filenames" value="ecoli.fastq" ftype="fastqsanger" /> <output name="out_fasta" file="ecoli.mira4_de_novo.fasta" ftype="fasta" /> <output name="out_bam" file="ecoli.mira4_de_novo.bam" ftype="bam" /> <output name="out_maf" file="ecoli.mira4_de_novo.mira" ftype="mira" /> <output name="out_log" file="ecoli.mira4_de_novo.log" ftype="txt" lines_diff="2000" /> </test> However, at only 5kb the FASTA file is small and fine to bundle - but the BAM (49kb), log (123kb) and MAF (764kb) quickly add up so I would prefer not to have to include these (under source code control, and for the Tool Shed upload). Also, as you might guess from the large number of allowed differences, the log file is quite variable (lots of time stamps plus Galaxy's temp filenames appear).
I've been talking to John (off list), he will reply here later, but in the meantime pointed me at a relevant commit, https://bitbucket.org/galaxy/galaxy-central/commits/d05be33ad6b772c2a688d875...
Basically the current Twill tests are extremely fragile and expect the test output files in the same order used in the <outputs> definitions and also there must be sample data for ALL the output files.
(I may have hit this bug before, I forget now).
I think it ought be possible to refactor the Twill test code to use the same dictionary approach used in the API based testing - and so support listing the test output in any order and only testing some of the files. However, I'll wait and see what you guys have to say. Peter.