bug#20365: 24.5; all-completions returns duplicates for Info-read-node-name-1

[Drew Adams <drew.adams@oracle.com>]

> Just a first comment: it's not considered incorrect for
> `all-completions' to return a list with duplicate entries in it.
> More specifically, it's considered the completion UI's job to remove
> those duplicates.

Yes, absolutely.

> > (setq x (all-completions "(" 'Info-read-node-name-1))
> > x will contain many duplicates for each node, like "org" "org"
> > "org" "org" "org.info.gz" "org" "org.info.gz" "org" "org.info.gz".
> 
> Maybe the way these entries are generated could be reviewed to try
> and reduce the number of duplicates.  And we could call `delete-dups' on
> the result: while a completion-table shouldn't need to go out of its way
> to reduce the number of duplicates (since the UI is supposed to handle
> it anyway), it's probably good to avoid having such expected large
> number of duplicates, indeed.

Just to be sure -

I hope you mean to do this only in `Info-read-node-name-1' or
in the code that calls `completing-read' (or in a "completion UI").
I hope you don't mean to do any of that in `all-completions'.
I'm guessing that this is what you mean.

If my guess is correct, then you need not read any further,
and thanks for confirming the guess.

---

In case not, let me disagree that such things should be done in
`all-completions' and similar functions or in `completing-read'.
They should not delete duplicates.

In my context, for instance, the completion candidates are often
alist elements where the cdrs matter.  Or they are strings that
are `string=' but have different properties.  The cdrs or the
string properties matter to the code that calls `all-completions'
or `completing-read'.  The calling code reconstructs the alist
element from the chosen candidate(s) (it provides access to the
associated cdr information).

It is important in such contexts to allow "duplicates", as
defined from the point of view of comparing only the cars with
`string=' (which is what `all-completions' does).  Just
because two completion candidate strings are equal does not
mean that the alist elements of which they are the cars are
equal.

It should be entirely up to the caller to delete duplicates.
It is trivial for it to do so, and logical that this be the
place where that is done.  Only the caller (or the "completion
UI") knows whether duplicates make sense in the current context.

`all-completions' & compagnie are building blocks (and coded
in C, no less, so not tweakable in Lisp).  It is not their
job to either (a) filter out duplicates or (b) try not to
produce duplicates.  (a) is the job of their callers, which
are the consumers of the candidates.  (b) is the job of the
COLLECTION function or other producer of the candidates.

So please try to tackle any problem of duplicates in this
particular use case by fiddling with `Info-read-node-name-1',
and not by introducing a change in `all-completions' etc.