attributes_to_columns() raises ValueError on some valid GFFs

[joelb123]

• pandasgff version: 1.2.0
• Python version: 3.8
• Operating System: linux

Description

I got the following failure when I read
this GFF file:

File "gffpandas/gffpandas.py", line 133, in
lambda attributes: dict([key_value_pair.split('=') for
ValueError: dictionary update sequence element #6 has length 1; 2 is required

This lambda is one of three in gffpandas. In my opinion, all 3 should be
refactored into map() or stand-alone functions. Lambdas are hard to read
and even harder to make defensive.

This failure is because a handful of features in this file end with a ";". Yes, this is a somewhat rare thing, but I also see it on some NCBI gffs. Like many ugly things in GFF-land, I don't know if it's strictly valid, but plenty of GFFs that pass GFF validators (e.g. the one in genometools) have them. There's an additional (even more rare) problem that somebody can put an equal sign into the feature string.

Here's a function that can be substituted for the lambda expression that fixes the problem:

def _split_atts(atts):
"""Split a feature string into attributes."""
splits_list = [a.split("=") for a in atts.split(";") if "=" in a]
return {l[0]:"=".join(l[1:]) for l in splits_list}