Cleaned up Reader classes

[richardjgowers]

In preparation for many Reader improvements ( #239 #314 ), did a clean up of how Reader.__iter__ works

Removed custom __iter__ from all but coordinates.xdrfile.core (XTC & TRR formats)

Removed many custom _sliced_iter implementations, can use base version

base.Reader now has __iter__ method

base.Reader.rewind now doesn't require seeking access, instead
_reopens and calls next

[richardjgowers]

I've just noticed this... does DCD skip do anything?

[orbeckst]

Looking at the DCDReader code, we just set self.skip = 1.
I think it really ought be called "step" but in the underlying dcd.c code, it's called skip. In principle the DCD C code should allow frame skipping at the C level with skip_dcdstep() but I just found the comment "XXX this needs to be changed"....

    // Figure out how many steps to skip
    numskip = skip - (dcd->setsread % skip) - 1;
    /* XXX this needs to be changed */
    rc = skip_dcdstep(dcd->fd, dcd->natoms, dcd->nfixed, dcd->charmm, numskip);

It might be related to having the possibility of fixed atoms in a DCD or it might simply not work as expected and that's why we seem to have effectively disabled skipping at the C level. I'm sorry, I don't remember.

It seems to work, though, because:

import MDAnalysis; from MDAnalysis.tests.datafiles import PSF, DCD
u = MDAnalysis.Universe(PSF,DCD)
u.trajectory.skip = 10
frames = [ts.frame for ts in u.trajectory]
print(frames)

gives

[10, 20, 30, 40, 50, 60, 70, 80, 90]

as expected,

[orbeckst]

@richardjgowers , looks fine. I'll merge it.

We're not loosing anything with regard to "DCD skip" but we can see if C-level skip can be used to speed up special cases of the the sliced iterator (although: maybe benchmark first on something big if it is faster than the python based implementation that uses computed seeks, which is pretty much the same what the C-version does).

[richardjgowers]

Ok thanks for clarifying. I'd expect that a Python file seek will be comparable too.