check for off-by-1 errors

[daler]

check for off-by-1 errors

[daler]

good idea.

[brentp]

so all starts in bedFile.h are converted to 0-based (gff, vcf have 1 subtracted) and ends are unchanged. So everything will have BED-like coordinates.
Nowhere in cbedtools.pyx do we account for this. so,

import pybedtools
b = pybedtools.BedTool('pybedtools/test/data/c.gff')
d = iter(b).next()
print d
d.start = d.start
print d

gives:
chr1 ucb gene 465 805 . + . ID=thaliana_1_465_805;match=scaffold_801404.1;rname=thaliana_1_465_805
chr1 ucb gene 464 805 . + . ID=thaliana_1_465_805;match=scaffold_801404.1;rname=thaliana_1_465_805

so the start has changed. Thoughts on how to address this?

[daler]

Phew, nice catch.

It seems that the fundamental issue is that there are 2 different 'start' values -- one is in _bed.start and the other is in _bed.fields[idx].

import pybedtools
b = pybedtools.BedTool('pybedtools/test/data/c.gff')
d = iter(b).next()
print d.fields[3] # 465
print d.start     # 464

Is there a good reason for bedFile.h to subtract 1?

[brentp]

| Is there a good reason for bedFile.h to subtract 1?

then it makes things like .length work the same for all.
maybe it's easiest just to have the .start property setter do:

if self.file_type != "bed": start -= 1

will that solve everything?

[daler]

looks like it won't . . .
the start -= 1 will continuously decrement the start position of d if you do:

d.start = d.start
d.start = d.start
etc

i just added a test for starts to be able to check this, 50974ad

[daler]

OK, I think this is fixed now. I assumed these conventions:

• Interval.start always contains the 0-based coord. This makes things like len() internally consistent for all feature types, and leaves bedFile.h alone.
• even when setting Interval.start for a GFF feature, the user-provided value provided is still assumed to be 0-based
• Interval.fields always contains the "string representation" to make the Interval a valid line for whatever format it represents. So for GFF features, Interval.fields[3] will always contain the 1-based start position
• For GFF files, when you set .start, Interval.fields[3] is updated to be str(start + 1)

If a user tries to use .fields[3] for something, hopefully the fact that they have to do an extra int() on it will be a reminder that it's different than the already-as-an-int Interval.start.

(see c0163ce for this, which includes BED- and GFF-specific tests)

[brentp]

closing as this is tested and documented (thanks @daler).