Re: [PATCH v10 6/9] convert: add 'working-tree-encoding' attribute

[Junio C Hamano <gitster@pobox.com>]

lars.schneider@autodesk.com writes:

> +static const char *git_path_check_encoding(struct attr_check_item *check)
> +{
> +	const char *value = check->value;
> +
> +	if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value) ||
> +	    !strlen(value))
> +		return NULL;

This means that having "* working-tree-encoding" (without "=value")
in your .gitattributes file is silently ignored.  

Thinking aloud.  I wonder if that is something we would want to warn
about so that the project can fix it, perhaps?  Or would that become
too noisy to use an older version of Git that includes this series
when a newer version that defines new meanings to boolean
(set/unset) values for w-t-e attribute becomes available and
projects adopt it?  On the other hand, with this code as-is or with
an additional warning, such an "older version" of Git by definition
behaves differently from such a "newer version" that does something
different when the attribute is not a non-empty string, so it is
quite likely that we won't be able to redefine or extend the meaning
of w-t-e in a way like that.

Which means that the only sensible way to make sure we _could_ later
extend the meaning of w-t-e and make it behave differently when set
to a non-empty string is to make it an error in _this_ "older"
version.  That way, of course people cannot use the "older" version
on a newer project that depends on the way how "newer" Git treats
w-t-e that is set to, say, "true", but we won't silently (or loudly)
do wrong things to the project's data.