That would require memory resizing, etc. which fromfile doesn't do. I suppose it could do it, but there seems no downside to me in reading via: data = raw_reading; arr = np.frombuffer(data, dtype='f') for binary data.
(For non-binary reads np.loadtxt, should be superior anyway.)

Well, it seems - naively - that reading raw, then calling frombuffer, would mean requiring twice the memory, at least temoprarily... but regardless - the point is that its not justified for NumPy to fail. That is, the fact that the user of numpy could have gone to the effort of checking the seekability of the file and writing something differently is not reason enough for refusing to perform the read within the library itself.

Well, it seems - naively - that reading raw, then calling frombuffer

Just for the record (not 100% sure you are aware): There is no additional copy involed.

Just for the record (not 100% sure you are aware): There is no additional copy involed.

Ok, that's a nice optimization, kudos... I'm not a Pythonista, so I wouldn't know.

However, in that case - why not just write the few lines of raw reading and a frombuffer invocation, in case you've discovered that the file is unseekable? It seems like little effort on your part, based on what you've said; it rounds out the API; it doesn't fill your codebase with only-used-once functionality... seems like a win.

Why not just improve the documentation by clearly stating the file is a raw binary file and not a buffered file?

I think the same issue occurs with numpy.rec.fromfile, but there documentation clearly states that "The file object must
support random access (i.e. it must have tell and seek methods).", as indicated below:

This would avoid confusion and close the issue. If a maintainer agree with that I can do the PR.

Otherwise, it would be better to tag this issue as an enhancement.

Why not just improve the documentation by clearly stating the file is a raw binary file and not a buffered file?

I think that sounds great, it doesn't mean we can't ever extend it (although if we point to alternatives, I am not sure I think that it is necessary).