split-paired-reads.py has nonstandard behavior with orphaned reads

[camillescott]

By default, split-paired-reads.py mixes orphaned reads into their respective .1 / .2 files, which is a nonstandard behavior -- orphaned reads should be put into their own files. Tools like bowtie and Trinity fail when they encounter orphaned reads mixed into split paired files.

[ctb]

We talked about this in person, and the conclusion was that we cannot alter this until khmer 2.0; but the -p was added in #818 to force the expected behavior.

Two thoughts -

1. shall we update the documentation to make this clear?
2. @mr-c, what's the right way to punt these kinds of issues to khmer 2.0 release?

[ctb]

For 2.0,

• make -p default on split-paired-reads;
• change to using broken_paired_reader(..., require_paired=True) in split-paired-reads;
• upgrade extract-paired-reads to properly handle streaming input and specification of output files;
• in the error message that results from -p, mention that extract-paired-reads can be used to fix;

@camillescott @mr-c the alternative here is to add an option to split-paired-reads to sideline or trash orphans. I think this makes the script too complicated so am -0 on it but would appreciate your thoughts.

[ctb]

Closed by #1164.