mne.io.read_raw_brainvision reads incorrect number of datapoints

I am writing a simple time series visualization script basing on MNE. I need to read data from EDF and BrainVision txt files, I have a sample that was exported to both. I discovered a strange behaviour that probably is a bug in mne.io.read_raw_brainvision, it is illustrated by

#!/usr/bin/env python
import mne

edf = mne.io.read_raw_edf("datasets/example.edf")
bv = mne.io.read_raw_brainvision("datasets/dattxt/example.vhdr")

print("Number of datapoints per channel ", edf.n_times, bv.n_times)

which prints in my case ('Number of datapoints per channel ', 831488, 73). It can be found (together with sample data) in this repo. I would expect that edf.n_times and bv.n_times be equal.

Additionally, when I try to read data from bv like so

data, time = bv[:, :1000]

I get

Traceback (most recent call last):
  File "./test.py", line 9, in <module>
    data, time = bv[:, :1000]
  File "/home/arccha/.conda/envs/pfb/lib/python2.7/site-packages/mne/io/base.py", line 841, in __getitem__
    verbose=self.verbose)
  File "/home/arccha/.conda/envs/pfb/lib/python2.7/site-packages/mne/io/base.py", line 523, in _read_segment
    cals, mult)
  File "/home/arccha/.conda/envs/pfb/lib/python2.7/site-packages/mne/io/brainvision/brainvision.py", line 116, in _read_segment_file
    _read_segments_c(self, data, idx, fi, start, stop, cals, mult)
  File "/home/arccha/.conda/envs/pfb/lib/python2.7/site-packages/mne/io/brainvision/brainvision.py", line 174, in _read_segments_c
    dtype = _fmt_dtype_dict[raw.orig_format]
TypeError: unhashable type: 'dict'

Am I doing something wrong or is it indeed a bug?

EDIT: I’ve tested this against MNE 0.15.2 and master branch.

[ArcCha]

@cbrnr as this files are rather big I’ve used GIT LFS to put them in the repository. You need to have it installed in order to clone them.

[agramfort]

ok sounds like a plan

can you open a PR?