SeqRecord broken on Debian packages when -sql package is not installed

[kblin]

Setup

I am reporting a problem with Biopython version, Python version, and operating
system as follows:

2.7.13 (default, Jan 19 2017, 14:48:08) 
[GCC 6.3.0 20170118]
CPython
Linux-4.4.0-93-generic-x86_64-with-debian-9.1
1.68

This is a Debian 9 (stretch), with python-biopython installed and python-biopython-sql not installed.

Expected behaviour

Calling extract on a SeqFeature that's coming from a SeqRecord loaded from a GenBank file on disk works.

Actual behaviour

__getitem__ in SeqRecord imports BioSQL.BioSeq without a try/except block, causing the script to crash.

Traceback (most recent call last):
  File "/usr/local/bin/run_antismash.py", line 1210, in <module>
    main()
  File "/usr/local/bin/run_antismash.py", line 556, in main
    run_analyses(seq_record, options, plugins)
  File "/usr/local/bin/run_antismash.py", line 640, in run_analyses
    cluster_specific_analysis(plugins, seq_record, options)
  File "/usr/local/bin/run_antismash.py", line 1191, in cluster_specific_analysis
    plugin.specific_analysis(seq_record, options)
  File "/antismash-dev/antismash/specific_modules/nrpspks/specific_analysis.py", line 81, in specific_analysis
    pksnrpsvars = annotate_pksnrps(pksnrpsvars, seq_record, options)
  File "/antismash-dev/antismash/specific_modules/nrpspks/nrpspksdomains.py", line 28, in annotate_pksnrps
    name_nrpspks(seq_record, pksnrpsvars, withinclustergenes, options)
  File "/antismash-dev/antismash/specific_modules/nrpspks/nrpspksdomains.py", line 163, in name_nrpspks
    quals['translation'] = [str(motifFeature.extract(seq_record).seq.translate(table=transl_table))]
  File "/usr/lib/python2.7/dist-packages/Bio/SeqFeature.py", line 329, in extract
    return self.location.extract(parent_sequence)
  File "/usr/lib/python2.7/dist-packages/Bio/SeqFeature.py", line 936, in extract
    f_seq = parent_sequence[self.nofuzzy_start:self.nofuzzy_end]
  File "/usr/lib/python2.7/dist-packages/Bio/SeqRecord.py", line 439, in __getitem__
    from BioSQL.BioSeq import DBSeqRecord
ImportError: No module named BioSQL.BioSeq

Steps to reproduce

Set up a pristine Debian install like this:

docker run --rm -it debian:stretch
apt update && apt install -y python-biopython

Then run the following python script:

from Bio.Seq import Seq
from Bio.SeqRecord import SeqRecord
seq = Seq("ATGGCAGGCATCTGCTAG")
rec = SeqRecord(seq)
rec[:5]

[kblin]

Just to confirm, installing python-biopython-sql in addition to python-biopython makes this problem go away, at the cost of pulling in client-side MariaDB.

[peterjc]

As discussed on Twitter, due to change in Biopython 1.68 onwards 48cd6bc1 - CC @ctSkennerton

Ideally all the Bio code would work without importing from BioSQL (as per the Debian split packaging approach).

We could implement __getitem__ on the DBSeqRecord object (more code duplication for all the annotation handling), or perhaps make that isinstance check pass if the import fails (if the import fails, it can't be a DBSeqRecord object anyway)?

[peterjc]

Cross reference https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=883860 (filed by @kblin)

[kblin]

I'm a bit stumped on how to write a meaningful test for this, otherwise I'd propose a patch that'd only do the DBSeqRecord logic if the import worked.

[kblin]

Not sure how to test if this works, though

[peterjc]

Closed via #1472 on the Biopython side, fix will be in our next release, i.e. Biopython 1.71

I suggest for Debian they back-port Kai's patch to whichever of the Biopython 1.68, 1.69 and 1.70 packages are still in active use.

[peterjc]

In terms of testing this (outside of Debian), one could hack the setup.py and/or delete the installed BioSQL/*.py files from the Python path, and then try test_SeqRecord.py or Kai's minimal test case before and after the patch.

I have not done that (its bed time here), but perhaps I should have to be 100% sure we've fixed this.