Add append(), overwrite() and accumulate()?

[bastibe]

I'm glad that __getitem__ and __setitem__ will be deprecated (see #75). I think that most of their functionality can be easily done with the other available functions and methods (and often in a better way).

There is only one set of functions where indexing could be quicker (but not necessarily better) than the currently available options:

• append(): open a file in 'r+' mode, seek to the end, write a given array of data, close the file.
• overwrite(): open a file in 'r+' mode, seek to a given position, write a given array of data, close the file.
• accumulate(): open a file in 'r+' mode, seek to a given position, accumulate a given array of data to the data which is already in the file, close the file.

Note that all of these only really make sense in 'r+' mode.
Note also, that append() is only a special case of overwrite().
accumulate() is probably too specialized to deserve its own function and it's also my least favorite of the 3.

I personally cannot imagine that I would ever need any of these, but if someone else needs them, I'm fine with supporting them.

[bastibe]

Very interesting idea! This would certainly solve many use cases very elegantly. I like it!

Let's do it after 0.6 though.

[faroit]

👍

[bastibe]

It turns out that you can write append as a special case of overwrite, but this requires opening the file twice (once for getting its length, and once for overwriting). I opted for not making it a special case instead.

[mgeier]

I think channels has to be removed here.
It should be obtained from data, like in the write() function.

[mgeier]

Same for overwrite() and accumulate(), of course.

[mgeier]

It turns out that you can write append as a special case of overwrite, but this requires opening the file twice (once for getting its length, and once for overwriting).

I think this is just a matter of the API of overwrite(). To be able to re-use it, it's necessary to expose the whence argument of seek().

I think those three "situations" are way too specialized (and I suspect too seldom used) to deserve three distinct top-level functions.

If we expose whence and add a accumulate argument (with default value False), we can get everything in one function:

• overwrite: overwrite(mydata, 'myfile.wav', start=27)
• append: overwrite(mydata, 'myfile.wav', start=0, whence=sf.SEEK_END)
• accumulate: overwrite(mydata, 'myfile.wav', start=44, accumulate=True)

There is still an inconsistency with the argument start, which I guess is motivated by the identically named argument in the read() function, but it (naturally) has no according frames and stop arguments.
OTOH, in truth, this is the argument frames from the seek() method, so there is some potential for confusion there.

Apart from all that, I'm still not convinced that any of this is necessary at all.

I think such specialized convenience functions are only justified if they are used often and if they are used in interactive sessions.
In "normal" Python code it's really easy to write a with statement and two to four more lines, and it may be even easier to understand and most likely easier to maintain than using one of the proposed functions!

@faroit: It's really helpful to get feedback, thanks for that!
But can you please elaborate a bit about how you would use those functions and share your opinion about if they are necessary or not?

[mgeier]

I don't like the repetition of those details. I would just refer to the function :func:write``.

[bastibe]

We agreed to not do this.