Consistently add wildcards for completion-pcm-leading-wildcard

completion-pcm--find-all-completions has two different phases:
First we turn the minibuffer text into a regex and matches
completion alternatives against it.  If that finds no matches,
then we strip some text off the completions and minibuffer text
and call ourselves recursively to find completions, then filter
the results with the removed text (converted into a regex).

Because of this, completion-pcm-leading-wildcard had
inconsistent behavior: in the second phase, the filter created
from the removed text would have a leading wildcard.  That
effectively adds wildcards in the middle of the minibuffer text
at the start of each "word".  But the first phrase created a
regex which had no such wildcards.  Thus, the two phases could
get substantially different results.

We fix this by changing completion-pcm-leading-wildcard to
consistently add a leading wildcard for each word.  This was
always my intention.

* lisp/minibuffer.el (completion-pcm--string->pattern): Include
a wildcard after each delimter with
completion-pcm-leading-wildcard. (bug#74772)
* lisp/minibuffer.el (completion-pcm-leading-wildcard): Update
docs.
* doc/emacs/mini.texi (Completion Styles): Update docs.

2 files changed, 12 insertions, 11 deletions

diff --git a/doc/emacs/mini.texi b/doc/emacs/mini.texi
index 0fcd24ed79d..8e0d58d0f7c 100644
--- a/doc/emacs/mini.texi
+++ b/doc/emacs/mini.texi
@@ -577,9 +577,9 @@ corresponding position in the completion alternative.
 
 @vindex completion-pcm-leading-wildcard
 If @code{completion-pcm-leading-wildcard} is set to @code{t}, this style
-always acts as if a @dfn{wildcard} is present at the start of the
-minibuffer text, similar to the @code{substring} style.  For example,
-@samp{l-m} will complete to @samp{emacs-lisp-mode}.
+always acts as if a @dfn{wildcard} is present at the start of each word
+in the minibuffer text, similar to the @code{substring} style.  For
+example, @samp{l-ode} will complete to @samp{emacs-lisp-mode}.
 
 @item emacs22
 @cindex @code{emacs22}, completion style
diff --git a/lisp/minibuffer.el b/lisp/minibuffer.el
index 253c26a84ea..39e7a399404 100644
--- a/lisp/minibuffer.el
+++ b/lisp/minibuffer.el
@@ -3951,17 +3951,18 @@ the commands start with a \"-\" or a SPC."
          trivial)))
 
 (defcustom completion-pcm-leading-wildcard nil
-  "If non-nil, partial-completion completes as if there's a leading wildcard.
+  "If non-nil, partial-completion behaves as if each word is preceded by wildcard.
 
-If nil (the default), partial-completion requires a matching completion
-alternative to have the same beginning as the first \"word\" in the
-minibuffer text, where \"word\" is determined by
+If nil (the default), partial-completion requires each word in a
+matching completion alternative to have the same beginning as each
+\"word\" in the minibuffer text, where \"word\" is determined by
 `completion-pcm-word-delimiters'.
 
 If non-nil, partial-completion allows any string of characters to occur
-at the beginning of a completion alternative, as if a wildcard such as
-\"*\" was present at the beginning of the minibuffer text.  This makes
-partial-completion behave more like the substring completion style."
+at the beginning of each word in a completion alternative, as if a
+wildcard such as \"*\" was present at the beginning of each word.  This
+makes partial-completion behave more like the substring completion
+style."
   :version "31.1"
   :type 'boolean)
 
@@ -4008,7 +4009,7 @@ or a symbol, see `completion-pcm--merge-completions'."
               (setq p0 p)
             (push (substring string p (match-end 0)) pattern)
             ;; `any-delim' is used so that "a-b" also finds "array->beginning".
-            (setq pending 'any-delim)
+            (setq pending (if completion-pcm-leading-wildcard 'prefix 'any-delim))
             (setq p0 (match-end 0))))
         (setq p p0))