GitLab

Makefile: Allow to launch a single functional test with valgrind by prefixing its name with valgrind_

This has no consequence since this rule is forced anyway.

Relative path is less robust when changing directory.

These informations are not so easy to parse... but at leastthey are here.

As k-mers with extended nucleotides are now ignored, the count of k-mers
in '-c germline' has slightly changed (< 0.05% variation in 'stanford-germlines.should_get').
There are also slight changes in the number of windows found in 'stanford-w100.should_get'.

The challenging sequence isolated in 1f12079d gives now the same window
when it is reversed-complemented. There was probably a 'N' in some IGHV germline.

There are some 'N' and other extended nucleotides in the germline sequences.
As we store in the indexes both the k-mers and their reverse complement, and as
we handle extended nucleotides almost randomly (see tools:nuc_to_int()),
we may have slight differences when analyzing some reads and their reverse complement.
Ignoring such k-mers allow thus to be more deterministic, getting the same
results on a (pure ACGT) read and its reverse complement.

Another option (harder to implement) could be to add several k-mers in the index,
but this would decrease the effective weight of the seed.

Note that we should also improve the analysis of reads that includes extended nucleotides.

This should change nothing, we should never go to the '?'.

As noted by @mikael-s, e963e165 hided a failed test in vidjil_revcomp_s22.should_get.
This test is one challenging sequence from Stanford-S22:
we currently do not find the same window on the sequence and on its revcomp.

valgrind_should: valgrind on functional tests
valgrind_tests:  valgrind on unit tests

pip install unittest2
pip install unittest-xml-reporting
python web2py.py -S vidjil -M -R testRunner.py
output folder > server/web2py/test-reports

This test was too long. We now just check the first 1000 reads and the main clone.

This invocation should also be more portable
See http://stackoverflow.com/questions/2792675/how-portable-is-mktemp1

By default the launcher is put at the start of the command line. In some cases
it must not be at the start and in other cases we don't want to use valgrind
(or another launcher) so we deactivate it with the !NO_LAUNCHER: directive

If we find it, we don't add it at the beginning (see doc.)

Makefile: For tests ensure that vidjil is rebuilt with -O0 -g so that we get useful information with Valgrind (slower...)

Use " rather than ' so that LAUNCHER can be expanded
which is useful when LAUNCHER contains some bash commands.

This allows to launch the script from another directory while
still benefiting from the default otions -p and -q.

If we overwrite the_kseg (in case it exists), it is definitely lost and we have a memory leak.