Re: ediff feature request: diffing line by line

[Carlo Traverso]

>>>>> "Michael" == Michael Kifer <address@hidden> writes:

>>>>> "CT" == Carlo Traverso <of Sun, 17 Mar 2002 17:26:46 +0100> writes:

    CT> I had missed ediff-regions-wordwise and
    CT> ediff-windows-wordwise, that solve a lot of my problems;
    CT> however these three enhancements would help:

    CT> 1 - switching from ediff-buffers to ediff-regions-wordwise: a
    CT> key could be defined to select the current ediff regions in
    CT> both buffers and enter an ediff-regions-wordwise on them; the
    CT> same for ediff-windows-wordwise. This is currently possible,
    CT> but not with one key (this should be extremely easy to
    CT> implement).

    Michael> I didn't understand the original problem, but when Alex
    Michael> Schroeder explained it I also thought about
    Michael> ediff-regions-wordwise. If I understand you and him
    Michael> correctly, all that is needed is to be able to
    Michael> conveniently invoke this function on the currently
    Michael> highlighted regions.  In fact this key already exists
    Michael> (=), but it asks you to select a region instead of taking
    Michael> the currently highlighted diffs.  I felt that having this
    Michael> key is not very useful, because one can simply run
    Michael> ediff-regions-* from command line or from the menu, and
    Michael> this won't be any more difficult. So, I am thinking of
    Michael> repurposing this key to run ediff-regions-wordwise on the
    Michael> selected diff regions.

Please, don't. I hate when a key to which I am used changes; there are
other unused keys, e.g. + and -, to run ediff-*-wordwise on the current *.



    CT> 2 - the highlighting scheme should be revised, since entering
    CT> ediff-regions-wordwise from ediff-buffers removes highlighting
    CT> from the current word (i.e. the current region in ediff and
    CT> the current word in ediff-regions-wordwise are highlited in
    CT> the same color...)  ediff-windows-wordwise inside of
    CT> ediff-buffers is even worse....  (this should be very easy
    CT> too)

    Michael> I don't understand. Are you saying that the highlighting
    Michael> of the current diff is not removed when you invoke
    Michael> ediff-regions-*? This is a bug, which I noticed recently.

Yes, non removing the highlighting makes the highlighting of the new
session ineffective.


    CT> 3 - enhancing ediff-regions-wordwise (ediff-windows-wordwise)
    CT> allowing to discover and reconcile whitespace "substantial"
    CT> differences: I consider "substantial" these differences:

    CT> - additional blank lines - space between words vs no space
    CT> between words (e.g. "one=1" vs "one = 1"

    CT> The amount of whitespace (e.g. "  " vs " ") or the type
    CT> (space, tab, newline) is inessential (but two consecutive
    CT> newlines is not the same as one newline...)

    Michael> What you are saying is that for word-wise operations the
    Michael> meaning of ediff-word should be different from line-wise
    Michael> operations. This makes sense.  If somebody comes up with
    Michael> a better definition, I can incorporate it.  Ediff is
    Michael> using a simple heuristic to determine what should
    Michael> constitute a word for the purpose of diffing. Take a look
    Michael> at ediff-forward-word in ediff-diff.el.  I found it to
    Michael> work very well for line-wise diffing, but I don't use
    Michael> word-wise diffing much and have no opinion about it.  If
    Michael> you can come up with a good (and simple) heuristic for
    Michael> word-wise diffing, I can incorporate it.

I'll look at that. 

I have also remarked some strange behaviour in ediff-*-wordwise; in
particular, if you accept one version of all the differences, then you
find a new set of differences; this is mainly due to the handling of
whitespace. I'll prepare a report on what I think is wrong (and maybe
a patch...).

Try for example a file consisting of the line

"one two three"

and one with the line

"one three"

(the result is "one twothree")


Carlo