Can you check what the parser did? It should be parsing /pseudo as a dictionary entry of feature.qualifiers['pseudo'] = None (not an empty list), which the writer handles as a special case:

https://github.com/biopython/biopython/blob/master/Bio/SeqIO/InsdcIO.py#L300

It's definitely an empty list in the example:

qualifiers:
    Key: codon_start, Value: ['1']
    ...
    Key: pseudo, Value: ['']
    ---

Edit: Technically, it's not even an empty list, but a list containing only an empty string.

Thanks - looks like something was broken in the parser, and we can't have any tests at the moment which check this explicitly.

OK many thanks, for now I will add a workaround. What do you mean with 'we can't have any tests which check this explicitly' -- are you saying there are no tests but there should be or that you aren't intending to test this at all?

Sorry, poor phrasing, I meant that fact this bug occurred implies we currently don't have test checking for this. We should have a test for this.

It looks like the parser was changed a while back. Switching the parser to use feature.qualifiers['pseudo'] =[None] here has some side effects related to the BioSQL tests, and the nasty special case of GenBank/empty_feature_qualifier.gb which does this:

                     /note="This is a correct note, the following one isn't"
                     /note

That is probably invalid, but might be stored as feature.qualifiers['note'] = ['This is a correct note, the following one isn't', None] and the more typical value-less example could be feature.qualifiers['pseudo'] =[None] which would be consistent with always being a list.

The good news is for the GenBank/EMBL output, you can use either feature.qualifiers['pseudo'] = None or feature.qualifiers['pseudo'] = [None] which is helpful.

Since this overlaps with BioSQL, I filed biosql/biosql#6

Well, this would work for setting new pseudo entries, but what I wanted was keeping the pseudo qualifier untouched in features read from EMBL. At the moment I am writing the EMBL output into a StringIO and then cleaning it up afterwards anyway (not just the /pseudos but also the ID line).

Pull request #905 makes the parser changes I outlined (so we can read and then write /pseudo correctly), but the BioSQL issue blocks this change as is. Further work needed...

Hi peterjc, I saw this is still handled like this in SeqIO.write and the code http://biopython.org/DIST/docs/api/Bio.GenBank-pysrc.html#_FeatureConsumer.feature_qualifier
Currently I'm fixing this with an additional sed routine, but on the longer term it would of course preferable to have this handled according the the NCBI specs. This will avoid other programs behave unpredictable when they rely on these genbank files. It's only an minor thing, but this is why the bug is easy to miss in the genbank file, too.
Thank you

@Carambakaracho Can you have a look at #905?