Hi Stuart,

On Tue, Jan 20, 2026 at 09:40:49AM +0000, Stuart Henderson wrote:
> If the existing unchecked el_set() calls fail it's not a big problem and
> there isn't really anything sftp(1) can do differently.
> 
> If el_get() fails you can't assume anything about the contents of the
> pointer so it's not safe to use it. (Also remember that this code will
> run with editline implementations other than OpenBSD's so being a bit
> careful with this is appropriate).
> 
> Note the return value of el_get() *is* checked in Complete() in sftp.c.
> 

I'd already understood why you suggested including the check.  I just
felt it was unnecessary and I answered the way I did so as not to come
across as pedantic.

Does this version convince you?


Index: sftp.c
===================================================================
RCS file: /cvs/src/usr.bin/ssh/sftp.c,v
diff -u -p -u -p -r1.247 sftp.c
--- sftp.c	13 Oct 2025 00:54:29 -0000	1.247
+++ sftp.c	20 Jan 2026 11:06:05 -0000
@@ -2204,6 +2204,7 @@ interactive_loop(struct sftp_conn *conn,
 	char *remote_path;
 	char *dir = NULL, *startdir = NULL;
 	char cmd[2048];
+	const char *editor = "emacs";
 	int err, interactive;
 	EditLine *el = NULL;
 	History *hl = NULL;
@@ -2220,7 +2221,7 @@ interactive_loop(struct sftp_conn *conn,
 		el_set(el, EL_HIST, history, hl);
 
 		el_set(el, EL_PROMPT, prompt);
-		el_set(el, EL_EDITOR, "emacs");
+		el_set(el, EL_EDITOR, editor);
 		el_set(el, EL_TERMINAL, NULL);
 		el_set(el, EL_SIGNAL, 1);
 		el_source(el, NULL);
@@ -2239,6 +2240,10 @@ interactive_loop(struct sftp_conn *conn,
 		el_set(el, EL_BIND, "\\e\\e[D", "ed-prev-word", NULL);
 		/* make ^w match ksh behaviour */
 		el_set(el, EL_BIND, "^w", "ed-delete-prev-word", NULL);
+
+		/* Vi command mode */
+		if (el_get(el, EL_EDITOR, &editor) == 0 && editor[0] == 'v')
+			el_set(el, EL_BIND, "^[", "vi-command-mode", NULL);
 	}
 
 	if ((remote_path = sftp_realpath(conn, ".")) == NULL)


-- 
Walter