bug#65137: 29.1; completion-substring-try-completion doesn't return the longest common substring

[Spencer Baugh <sbaugh@janestreet.com>]

[-- Attachment #1: Type: text/plain, Size: 540 bytes --]

Stefan Monnier <monnier@iro.umontreal.ca> writes:
>>   (completion-pcm--merge-completions '("ab" "ab") '(prefix "b"))
>>   -> ("b")
>
> Right this is a bug.
>
>     (completion-pcm--merge-completions '("ab" "sab") '(prefix "b"))
>
> returns (correctly)
>
>     ("b" "a" prefix)
>
> so
>
>     (completion-pcm--merge-completions '("ab" "ab") '(prefix "b"))
>
> should return either ("b" "a") or ("b" "a" prefix).
>
> Could you accompany your patch of a regression test using
> `completion-pcm--merge-completions` as above?

Patch with test:

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0001-Correctly-handle-common-prefixes-in-substring-comple.patch --]
[-- Type: text/x-patch, Size: 4128 bytes --]

From 364a10aa54acd8de71a10d1ab98d059a27f26460 Mon Sep 17 00:00:00 2001
From: Spencer Baugh <sbaugh@janestreet.com>
Date: Tue, 5 Sep 2023 15:40:06 -0400
Subject: [PATCH] Correctly handle common prefixes in substring completion

Substring completion would previously not complete the longest common
substring if that substring was a prefix of all the completion
alternatives.  Now it does.  An explanation of this bug

Substring completion is implemented by passing the `prefix' symbol as
part of the pattern passed to completion-pcm--merge-completions.  This
symbol is supposed to cause completion-pcm--merge-completions to
"grow" a completion of a common substring only from the "right" of the
symbol (a common suffix), not from the "left" of the symbol (a common
prefix).  Yes, this is the opposite of what the name `prefix' would
imply.

When processing a symbolic element of the pattern,
completion-pcm--merge-completions first finds the common prefix of all
the completions in that part of the pattern (using try-completion).
Then for `prefix' and other elements which want to complete a common
suffix, the common prefix is removed from each element and then the
common suffix is calculated with completion--common-suffix.

If the common prefix covers the entirety of all the alternatives
(i.e. when "unique" is true in the code), it's also a common suffix.
In that case, the common suffix calculation (if it runs) is basically
a no-op which will produce an empty string, since we removed the
common prefix before running it.

Before this change, `prefix' elements would unconditionally discard
the common prefix, which produced the wrong result in the case that
common prefix == common suffix.  For example:

  (completion-pcm--merge-completions '("ab" "ab") '(prefix "b"))
  -> ("b")

Now we detect this situation and include the common prefix in this
case for `prefix' elements.  Then we get:

  (completion-pcm--merge-completions '("ab" "ab") '(prefix "b"))
  -> ("b" "a")

which is correct.

* lisp/minibuffer.el (completion-pcm--merge-completions): Don't ignore
a common suffix in a `prefix' pattern element when it's also a common
prefix.
* test/lisp/minibuffer-tests.el (completion-substring-test-5): Add a
test.
---
 lisp/minibuffer.el            |  4 +++-
 test/lisp/minibuffer-tests.el | 13 +++++++++++++
 2 files changed, 16 insertions(+), 1 deletion(-)

diff --git a/lisp/minibuffer.el b/lisp/minibuffer.el
index 35b359a75e2..ac054da61d2 100644
--- a/lisp/minibuffer.el
+++ b/lisp/minibuffer.el
@@ -4038,7 +4038,9 @@ completion-pcm--merge-completions
                      (unique (or (and (eq prefix t) (setq prefix fixed))
                                  (and (stringp prefix)
                                       (eq t (try-completion prefix comps))))))
-                (unless (or (eq elem 'prefix)
+                ;; if the common prefix is unique, it also is a common
+                ;; suffix, so we should add it for `prefix' elements
+                (unless (or (and (eq elem 'prefix) (not unique))
                             (equal prefix ""))
                   (push prefix res))
                 ;; If there's only one completion, `elem' is not useful
diff --git a/test/lisp/minibuffer-tests.el b/test/lisp/minibuffer-tests.el
index ff58d35eb3e..4f92d7f841c 100644
--- a/test/lisp/minibuffer-tests.el
+++ b/test/lisp/minibuffer-tests.el
@@ -298,6 +298,19 @@ completion-substring-test-4
                   "jab" '("dabjabstabby" "many") nil 3)))
            6)))
 
+(ert-deftest completion-substring-test-5 ()
+  ;; merge-completions needs to work correctly when
+  (should (equal
+           (completion-pcm--merge-completions '("ab" "sab") '(prefix "b"))
+           '("b" "a" prefix)))
+  (should (equal
+           (completion-pcm--merge-completions '("ab" "ab") '(prefix "b"))
+           '("b" "a")))
+  ;; substring completion should successfully complete the entire string
+  (should (equal
+           (completion-substring-try-completion "b" '("ab" "ab") nil 0)
+           '("ab" . 2))))
+
 (ert-deftest completion-flex-test-1 ()
   ;; Fuzzy match
   (should (equal
-- 
2.39.3


[-- Attachment #3: Type: text/plain, Size: 1014 bytes --]


> Similarly, I see that
>
>     (completion-pcm--merge-completions '("abr" "absabr") '(prefix "br"))
>
> returns
>
>     ("br" prefix)
>
> whereas it should arguably return
>
>     ("br" "a" prefix)
>
> [ Tho this may have the side effect that after this completion, `absabr`
>   won't be considered any more, if the `basic` completion comes before
>   `substring` :-(  ]

I did notice this too.  I could try fixing/changing this too, but it
does seem annoying when basic comes before substring - as it does by
default in a number of completion categories.

I wonder if we should move basic to after substring in those categories
in completion-category-defaults?  Or just remove basic from them.  It
doesn't seem like having both basic and substring in those lists has
much point.  Plus it would be a nice improvement to defaults - buffer
completion in particular confused me for a long time before I understood
how basic and substring interacted, so having only substring would ease
understanding for new users.