Cutadapt

-----------------

* Added ``--action=crop`` for removing everything *except* the adapter sequence.
(The sequence before and after the adapter is removed.) This can be useful if the
"adapter" contains wildcards (such as ``N``) that you are interested in.
* :issue:`788`: Added option ``-L`` as a counterpart to ``-l``/``--length``,
which allows shortening R1 and R2 to different lengths.
* A too long minimum overlap is now automatically decreased to the length of
the adapter. That is, you can use ``-a 'ACGT;o=99'`` to force full-length
matches.
* :issue:`784`: Fixed some unexpected trimming results for anchored 5'
adapters (due to an incorrectly initialized alignment score matrix).

-------------------

* :issue:`772`: Restored ability to provide input files via
process substitution.
* :issue:`773`: Fixed crash when writing interleaved data to stdout
(that is, when no ``-o`` option is provided).

-----------------

* :issue:`767`: Error messages are always sent to stderr. They were
previously sent to stdout when using `-o`.
* Cutadapt can now read single-end data from unaligned BAM files (uBAM).
* Dropped support for Python 3.7.

-------------------

* :issue:`744`, :issue:`561`: Make ``--revcomp`` work with paired-end data.

-----------------

* :pr:`725`: Added a ``--max-average-error-rate``/``--max-aer`` option to add a filter
that checks if the number of expected errors divided by read length is above a
certain threshold. The expected errors are calculated the same as in
``--max-expected-errors``, and dividing by read length helps for reads that
have varying lengths.
* :issue:`696`: Added a histogram of the lengths of removed poly-A tails to
the report.
* :issue:`696`: For paired-end data, ``--poly-a`` was changed to trim poly-T
"heads" on R2 (this is still experimental as it is unclear whether that is
the desired behavior. Please give feedback!)
* A poly-A tail is only removed if it is at least three nucleotides long.
* :issue:`734`: Fixed misassignments during demultiplexing that would sometimes
happen when there are collisions between adapter sequences
(when the warning "sequence ... cannot be assigned uniquely" was printed).
Previously, sequences could incorrectly be assigned to an adapter that is not
actually the best match.

-----------------

* :issue:`695`: Fixed very slow k-mer heuristic initialization (hours instead
of seconds) for degenerate adapter sequences such as ``A{100}`` as used
when doing poly-A trimming.
* :issue:`696`: Added option ``--poly-a`` for trimming poly-A tails. This is
more accurate and multiple times faster than using ``-a A{100}`` as previously
recommended. This is currently experimental (that is, the algorithm may change
in the next one or two releases).
* :issue:`685`: Sped up index generation somewhat. This is most noticable when
demultiplexing using thousands or more adapters. The speedup is
different depending on whether indels are allowed or not because
different algorithms are used.
* :issue:`685`: Sped up demultiplexing (when using an index) for the case when
the read contains ``N`` bases within the region where the adapter matches.
Previously, any ``N`` would disable the index for that read and trigger a
fallback to the slow method of matching each adapter one-by-one. Now the index
is used even in those cases.
* :pr:`700`: Sped up ``--max-expected-errors``. Thanks rhpvorderman.