history-delete-duplicates and keep-all

history-delete-duplicates and keep-all

[Juri Linkov]

Adding the arg `keep-all' to `read-from-minibuffer' a few months ago
caused a bug for the case when `history-delete-duplicates' is t.

While looking at fixing this bug, I realized that instead of using the
new arg `keep-all', a cleaner solution would be to reuse the existing
variable `history-delete-duplicates'.  Its purpose is exactly the
same: to control how input from the minibuffer is added to the
history.  So let-binding `history-delete-duplicates' in necessary
places will make the arg `keep-all' obsolete.  This arg also has
another drawback: it is available only to one completion function
`read-from-minibuffer' and is not available to all other completion
functions.

Currently `keep-all' is used only in `query-replace'.  So it is not
too late to let-bind `history-delete-duplicates' in this place instead.

Thus `history-delete-duplicates' should support three different values:

1. to put all inputs in the history list, even empty and duplicate ones

2. to put input in the history list, unless it is identical to the most
   recent history item (this is the current default behavior)

3. to delete duplicate inputs from the history list (this happens when
   `history-delete-duplicates' is t)

Also currently `call-interactively' doesn't respect the value of
`history-delete-duplicates', and always puts all duplicate inputs in the
`command-history'.  It works as if `history-delete-duplicates' had the
value's type 1 from the list above.  This could be fixed as well.

-- 
Juri Linkov
http://www.jurta.org/emacs/

Re: history-delete-duplicates and keep-all

[Richard M. Stallman]

    Currently `keep-all' is used only in `query-replace'.  So it is not
    too late to let-bind `history-delete-duplicates' in this place instead.
    ...
    Thus `history-delete-duplicates' should support three different values:

We could do that, or make a separate variable.
I think a separate variable would be better, because
these two different activities are for different purposes.

Which means it is really a question of whether to use
an argument for this, or a new global variable.

      This arg also has
    another drawback: it is available only to one completion function
    `read-from-minibuffer' and is not available to all other completion
    functions.

I think that is a reason to switch to a variable.
Would you please make that change?

Please update NEWS and the Lisp manual, too.