Clearer error messages when write() is called with incorrectly formatted channels

[mcclure]

Resolves issue #404

[mcclure]

Tests

Mono/interleaved when stereo expected:

import soundfile
s = soundfile.SoundFile("outtest.wav", mode = 'w', samplerate=48000, channels=2, subtype='PCM_16')
s.write([0,1,2,3])
print(s)

Output

Invalid shape: (4,) (Expected 2 channels, got 1)

Gave data in row-order rather than column-order (or other way around? idk)

import soundfile
import numpy
s = soundfile.SoundFile("outtest.wav", mode = 'w', samplerate=48000, channels=2, subtype='PCM_16')
s.write(numpy.array([[0,1,2,3,4,5], [6,7,8,9,10,11]], dtype=numpy.int16)) # Oops made 5x2 instead of 2x5
print(s)

Output:

ValueError: Invalid shape: (2, 6) (Expected 2 channels, got 6)

Channel count was bad to begin with:

import soundfile
s = soundfile.SoundFile("outtest.wav", mode = 'w', samplerate=48000, channels=4, subtype='PCM_16')
s.write([[0,0,0,0], [2**32,-2**32,2**32,-2**32]])
print(s)

output:

ValueError: Invalid shape: (4,) (Expected 2 channels, got 1)

If it's confusing that parenthesis are used for both the shape and the note afterward, I could change the parenthetical from round to square brackets.

[bastibe]

The new error messages look good, that's a great improvement.

But the code logic is incorrect, I believe. It should follow the original logic.

[bastibe]

This seems wrong. Previously, we were checking if the input array had one or two dimensions. Now we're checking whether we have one or two channels. Those are different things, right?

We support arbitrary numbers of channels (depending on the file type), but only one- or two-dimensional data.

[mcclure]

I misunderstood what the old code was doing but I see the problem now. Will fix

[mcclure]

With corrections, new tests

import soundfile
s = soundfile.SoundFile("outtest.wav", mode = 'w', samplerate=48000, channels=4, subtype='PCM_16')
s.write([[0,0,0,0], [1,1,1,1]])
print(s)

Output: Successfully writes 4-channel file

import soundfile
s = soundfile.SoundFile("outtest.wav", mode = 'w', samplerate=48000, channels=4, subtype='PCM_16')
s.write([[[1,2],[3,4]],[[5,6],[7,8]]])
print(s)

Output: ValueError: Invalid shape: (2, 2, 2) (too many dimensions)

import soundfile
import numpy
s = soundfile.SoundFile("outtest.wav", mode = 'w', samplerate=48000, channels=2, subtype='PCM_16')
q = numpy.array(9, dtype=numpy.int16)
print(q)
print(q.ndim)
s.write(q) # Oops it's a scalar
print(s)

Output:

9
0
[snip]
ValueError: Invalid shape: (1,) (Expected 2 channels, got 1)

If I change channels= to 1, this successfully writes a 1-channel file.

It appears something earlier in the code is converting scalars (0-dimensional arrays) to 1-dimensional arrays. This is an okay behavior, and I recommend leaving the check in _array_io as submitted so the code works standalone.

[bastibe]

Thank you very much! The error messages are indeed much improved!

Is this ready to merge from your end?

[mcclure]

Is this ready to merge from your end?

Yes.

[bastibe]

Wait, why is _soundfile_data updated in this pull request? This breaks all the tests, too.