Re: next-error-last-buffer

[Richard Stallman]

    > Indeed.  In order to get the right behavior in all cases, we'd need
    > to read the user's mind.  So the current approach at least has the
    > advantage of being simple for the user to understand (and thus predict),
    > while working correctly for the most common cases.

    The current approach is simple, but unusable in real situations.

I expect the same would be true if we make the change that you
propose.

I pointed out, when we first started to unify Occur with next-error,
that finding the right way to treat these issues would be the hardest
part of the job.  On such issues it is quite common to find that
all the straightforward options are clearly wrong in some cases.

    > If you switch to another source buffer that was last reached
    > through C-x ` from a compilation error, then C-x ` will get the
    > next compilation error, rather than the next occurrence.
    > At least, it sounds like your change would have that result.

    Exactly.  This is the most expected behavior, because the user
    associates the buffer visited through C-x ` with the compilation
    buffer from which it was visited

You're assuming one scenario, but there are others just as plausible.
For instance, the user still working on looking through all the
occurrences in the Occur buffer, and looks at this other buffer
because one of its functions was called from the code at one of those
occurrences.

    It seems nobody really tried yet to use two or more different
    compile/grep/occur processes at the same time.  Otherwise,
    it would be apparent that this behavior is the most useful.

It could not hurt for more people to try these cases,
and try playing with different behaviors, so that we can
find out what is really best.