[Numpy-discussion] whitespace in git repo

[Friedrich Romstedt]

attribute: 'true', 'false', '<value>', ''

attributes: text, eol, core.autocrlf
filters: left alone.
core.safecrlf: left alone.


results.
normalise: True, False
workingdir_fmt: 'lf', 'crlf'

# Apply text.
# Can be skipped on check-out.
if text == 'true':
    normalise = True
elif text == 'false':
    normalise = False
elif text == 'auto':
    normalise = file_is_text
elif text == '':
    if file_is_text and core.autocrlf:
        # depends on is_normalised, in_repo
        # 0 0: True
        # 0 1: False
        # 1 0: N/A (does not occur)
        # 1 1: True
        normalise = not (in_repo and not is_normalised)
        # Except it is unnormalised in repo, do normalise.
    else:
        normalise = False

# SO we should think about "text" as "normalise_preference"

# Two levels of indentation get necessarily prosaic ...

# Define default fmt.
if core.autocrlf:
    workingdir_fmt = 'crlf'
else:
    workingdir_fmt = core.eol  # defaults to native

# Apply eol attribute.
if eol == 'lf':
    normalise = True
    workingdir_fmt = 'lf'
elif eol == 'crlf':
    normalise = True
    workingdir_fmt = 'crlf'
else:
    pass

# Translation of crlf attribute:
# crlf == 'true'  =>  text = 'true'
# crlf == 'false'  =>  text = 'false'
# crlf == 'input'  =>  eol = 'lf'

CHECKING IN

if normalise:
    checkin(normalise(file))
else:
    checkin(file)

CHECKING OUT

checkout(file, workingdir_fmt)


NOTES

Files are not changed without a checkin.

So I think Darren's last pull req is just perfect.  I was just worried
about the mechanisms that lead to the changes, but when Darren says
they are all okay and sensible then I think they are.

Maybe the pseudo code would be useful for somehow git docs too, I mean
the people who use git are coders and not philologists.

I checked the code but I cannot fully exclude bugs, as usual ..

Friedrich