GitLab

Those tests were not included in our systematic testing for
a long time. This should be corrected now.

Last change here was 84e52850.

As we now store Vidjil results in '.vdj' files, we can run the test suite without actually relaunching all tests.

Ambiguous terms should be surrounded by parentheses

We may want to launch the locus tests only without launching the whole test suite.

Never commit in a hurry!

The previous commit (see 8496cbb0) indicated a suffix for the backup files
(necessary with POSIX sed). Then we were overwriting the tap files with the
bak ones. This was totally stupid as we were overwriting the correct files
with the old ones…

This is now fixed we just delete the backup files as we don't really need them
(we just specify the option to ensure the compatibility between POSIX and GNU
sed).

GNU sed and POSIX sed have slightly different behaviors.  Those behaviors are
incompatible if we don't specify a “backup” suffix with -i. So we specify such
a suffix and then we move all the files so that they overwrite the original
files.
Moreover in this case, the -e option is necessary for FreeBSD's sed.

sed may return an error code if there is no tap file in one of the
subdirectory. I didn't find an option in sed to ignore such errors, so ignore
all errors.

Handle DDJ recombinations on TRD.
This corrects one bug but there are still some sequencesthat are not correctly segmented.
This should be further investigated.

-i option should be put before the filenames in FreeBSD sed.

When git is not available or the current directory is not under a git, an
error message is displayed each time the command is launched.

Using $(shell ...) construct the command will be launched at the affectation
and not each time the variable is used.

diffstat may not exist. We must prevent the build from failing because of that

grep may return a non-zero code when some files are unreadable, this
would fail the test which is not desired.

When the global test suite is passing, individual failed tests (counted in SHOULD_*_EXPECTED_FAILS)
can be marked as 'TODO' to make continuous integration happy

Follows a973c9dd

This will help further bug hunting.

This sequence, previously expertised in 84f2acae, fails since cd292aa0.
The tests are now passing.

We run should-vdj-to-tap.py unbuffered (python -u) to have the best tee experience.

The segmentations were previously xxx (because of a known bug on the evalue) and are now IGH
as expected. Since there are more affectations in the index, the xxx has more ambiguous
positions now, which increases its e-value.

Mikaël Salson authored Sep 21, 2016 and Mathieu Giraud committed Sep 23, 2016

Due to 9db2ea1c, there is one more test classified as BUG.
(release-2016.08 does not have the right number of expected fails)

Mikaël Salson authored Sep 21, 2016 and Mathieu Giraud committed Sep 23, 2016

Due to 79c2bebc the number of failed tests increase since we changed two TODOs
in BUG

Mikaël Salson authored Sep 01, 2016 and Mathieu Giraud committed Sep 23, 2016

All files inside refer to curated-vdj, not should-vdj

Commits ad0fe37e and 5815bf0d were very lazy, they are reverted.
It is safer to still check all these tests, allowing some failed tests corresponding
to recent or challenging .should-vdj sequences.

Te results of the tests were carefully compared to the ones before ad0fe37e/5815bf0d.
The difference (6 tests) come from some missing D segments due to 0c3cd0e.

as in algo/Makefile

Contains sequences from other files (redundant)

Those files contain fake sequences.