loadtxt and savetxt

[certik]

Implementation: https://github.com/certik/fortran-utils/blob/b43bd24cd421509a5bc6d3b9c3eeae8ce856ed88/src/utils.f90#L176

The interface is compatible with NumPy (e.g., you can do savetxt from Fortran and do loadtxt from NumPy and it just works, and vice versa).

The loadtxt has the argument as allocatable, intent(out), because you typically do not know the size of the matrix ahead of time.

NumPy: savetxt, loadtxt

[milancurcic]

I think these are great functions to start with and figure out the workflow. Good to go ahead as far as I'm concerened.

What module would this belong to? stdlib_io in io.f90?

[certik]

I think we agreed to prefix modules with stdlib_.

The io module (stdlib_io.f90) might be a good name for it. NumPy has this directly in the numpy namespace, but I think we should have things in modules, not the global stdlib namespace. Because we can always later expose some frequently used functionality in stdlib.f90 directly, but if we put it there now, it will be hard to remove later, as it would break people's code.

There is also scipy.io, with loadmat that loads Matlab's file.

So stdlib_io.f90 works.

[jvdp1]

@certik Nice examples. I think it could be easily combine with something like

real, allocatable :: a(:)
type(File) :: f
f = File('log.txt')
call f%read(a)
call f%close()

Float, double-precision,integer,... could be then supported.

[milancurcic]

@certik Correct, I just re-read #9. module stdlib_io in stdlib_io.f90 is good.

[milancurcic]

@jvdp1 Indeed, plain savetxt and loadtxt will be useful on their own, and the File derived type (a higher level abstraction) can later figure out if and how to use these more basic building blocks.

[certik]

I sent a PR in #23.

[jacobwilliams]

We can also get some inspiration from this csv read/write one of mine: https://github.com/jacobwilliams/fortran-csv-module

[certik]

@jacobwilliams good point, I created a standalone issue #24 for CSV. You are right that it is related, as in NumPy one can use genfromtxt to read CSV.

[epagone]

Very nice start. What do you think of adding some optional functionalities like the following?

• Header handling (we might need to define a "comment" character that introduces the header):
  • when reading simply detect it and store it somewhere;
  • when writing add it and number the columns automatically (very useful for writing plotting scripts).
• column sorting or indexing using, for example, ODERPACK (if there is no licence clash).

I found very useful (actually necessary) both them but I don't know how much general interest there is.

[certik]

@epagone, yes, we can add the header, footer, comments, newline, delimiter arguments for savetxt that NumPy has. We can also do the fmt argument and pass in Fortran formatting string.

All these arguments go after the fname and X arguments, so I think we can add them in subsequent PRs after #23 is merged.

Can you elaborate on the column sorting or indexing using? I don't think NumPy does that. What prevents the user to sort the data before calling savetxt?