lei: improve bash completion involving colons

This fixes completions of labels (`+L:' for `lei import' and
`L:' for `lei q') so they can appear anywhere in the
command-line.

I mainly wanted this for `lei import $URL +L:label', but
this also fixes `lei forget-external' completions for URLs
(which involve colons).

1 files changed, 13 insertions, 18 deletions

diff --git a/lib/PublicInbox/LeiExternal.pm b/lib/PublicInbox/LeiExternal.pm
index 3e2a2288..31b9bd1e 100644
--- a/lib/PublicInbox/LeiExternal.pm
+++ b/lib/PublicInbox/LeiExternal.pm
@@ -86,39 +86,34 @@ sub canonicalize_excludes {
 # returns an anonymous sub which returns an array of potential results
 sub complete_url_prepare {
         my $argv = $_[-1]; # $_[0] may be $lei
-        # Workaround bash word-splitting URLs to ['https', ':', '//' ...]
-        # Maybe there's a better way to go about this in
-        # contrib/completion/lei-completion.bash
-        my $re = '';
-        my $cur = pop(@$argv) // '';
+        # Workaround bash default COMP_WORDBREAKS splitting URLs to
+        # ['https', ':', '//', ...].  COMP_WORDBREAKS is global for all
+        # completions loaded, not just ours, so we can't change it.
+        # cf. contrib/completion/lei-completion.bash
+        my ($pfx, $cur)  = ('', pop(@$argv) // '');
         if (@$argv) {
                 my @x = @$argv;
-                if ($cur eq ':' && @x) {
+                if ($cur =~ /\A[:;=]\z/) { # COMP_WORDBREAKS + URL union
                         push @x, $cur;
                         $cur = '';
                 }
-                while (@x > 2 && $x[0] !~ /\A(?:http|nntp|imap)s?\z/i &&
-                                $x[1] ne ':') {
-                        shift @x;
+                while (@x && $pfx !~ m!\A(?: (?:[\+\-]?(?:L|kw):) |
+                                (?:(?:imap|nntp|http)s?:) |
+                                (?:--\w?\z)|(?:-\w?\z) )!x) {
+                        $pfx = pop(@x).$pfx;
                 }
-                if (@x >= 2) { # qw(https : hostname : 443) or qw(http :)
-                        $re = join('', @x);
-                } else { # just filter out the flags and hope for the best
-                        $re = join('', grep(!/^-/, @$argv));
-                }
-                $re = quotemeta($re);
         }
+        my $re = qr!\A\Q$pfx\E(\Q$cur\E.*)!;
         my $match_cb = sub {
                 # the "//;" here (for AUTH=ANONYMOUS) interacts badly with
                 # bash tab completion, strip it out for now since our commands
                 # work w/o it.  Not sure if there's a better solution...
                 $_[0] =~ s!//;AUTH=ANONYMOUS\@!//!i;
-                $_[0] =~ s!;!\\;!g;
                 # only return the part specified on the CLI
                 # don't duplicate if already 100% completed
-                $_[0] =~ /\A$re(\Q$cur\E.*)/ ? ($cur eq $1 ? () : $1) : ()
+                $_[0] =~ $re ? ($cur eq $1 ? () : $1) : ()
         };
-        wantarray ? ($re, $cur, $match_cb) : $match_cb;
+        wantarray ? ($pfx, $cur, $match_cb) : $match_cb;
 }
 
 1;