Precision when saving data to ASCII about silx

[PiRK]

If we want to be conservative we could try to detect the data type for each column when the user doesn't specify which format he wants. My suggestion would be:

• %d for integers
• %.7g for float32
• %.18e for float64

If the user specifies a format we should assume he knows what he is doing and not try to check types.

We can do something similar to numpy.savetxt (http://docs.scipy.org/doc/numpy-1.10.0/reference/generated/numpy.savetxt.html): 

• fmt parameter:

single format (%10.5f), a sequence of formats, or a multi-format string, e.g. ‘Iteration %d – %10.5f’, in which case delimiter is ignored.

Except we could set fmt=None as default and assign our default format depending on the type.

Should we handle complex values ?

But again, I would not have any moral problem if we just assumed that a vast majority users who don't bother defining a custom format would be happy with %.7g for all values, because it looks nicer and uses less disk space. We would just need to specify in the documentation that the default behavior is not safe for extremely large integers and very precise floats.

[t20100]

Yes, that's why I asked the question.
Do we want a conservative default behavior or a simple one?

[vasole]

Simple one. If it turns out to be not enough we change. A bit of Extreme Programming should make no harm. If we start from "Cogito ergo sum" we are not going to get very far.