lineedit: invalid unicode characters are replaced with CONFIG_SUBST_WCHAR

function                                             old     new   delta
read_key_ungets                                        -      50     +50
lineedit_read_key                                    223     252     +29

Signed-off-by: Tomas Heinrich <heinrich.tomas@gmail.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

include/libbb.h

@@ -1277,6 +1277,7 @@ enum {
  * on first call.
  */
 int64_t read_key(int fd, char *buffer) FAST_FUNC;
+void read_key_ungets(char *buffer, const char *str, unsigned len) FAST_FUNC;
 
 
 #if ENABLE_FEATURE_EDITING

libbb/lineedit.c

@@ -1700,18 +1700,34 @@ static int lineedit_read_key(char *read_key_buffer)
 #endif
 
 #if ENABLE_FEATURE_ASSUME_UNICODE
-		{
+		if (unicode_status == UNICODE_ON) {
 			wchar_t wc;
 
 			if ((int32_t)ic < 0) /* KEYCODE_xxx */
 				return ic;
+			// TODO: imagine sequence like: 0xff, <left-arrow>: we are currently losing 0xff...
+
 			unicode_buf[unicode_idx++] = ic;
 			unicode_buf[unicode_idx] = '\0';
-			if (mbstowcs(&wc, unicode_buf, 1) != 1 && unicode_idx < MB_CUR_MAX) {
+			if (mbstowcs(&wc, unicode_buf, 1) != 1) {
-				delay = 50;
+				/* Not (yet?) a valid unicode char */
-				goto poll_again;
+				if (unicode_idx < MB_CUR_MAX) {
+					delay = 50;
+					goto poll_again;
+				}
+				/* Invalid sequence. Save all "bad bytes" except first */
+				read_key_ungets(read_key_buffer, unicode_buf + 1, MB_CUR_MAX - 1);
+				/*
+				 * ic = unicode_buf[0] sounds even better, but currently
+				 * this does not work: wchar_t[] -> char[] conversion
+				 * when lineedit finishes mangles such "raw bytes"
+				 * (by misinterpreting them as unicode chars):
+				 */
+				ic = CONFIG_SUBST_WCHAR;
+			} else {
+				/* Valid unicode char, return its code */
+				ic = wc;
 			}
-			ic = wc;
 		}
 #endif
 	} while (errno == EAGAIN);

libbb/read_key.c

@@ -246,3 +246,12 @@ int64_t FAST_FUNC read_key(int fd, char *buffer)
 	buffer[-1] = 0;
 	goto start_over;
 }
+
+void FAST_FUNC read_key_ungets(char *buffer, const char *str, unsigned len)
+{
+	unsigned cur_len = (unsigned char)buffer[0];
+	if (len > KEYCODE_BUFFER_SIZE-1 - cur_len)
+		len = KEYCODE_BUFFER_SIZE-1 - cur_len;
+	memcpy(buffer + 1 + cur_len, str, len);
+	buffer[0] += cur_len + len;
+}

testsuite/ash.tests

@@ -0,0 +1,42 @@
+#!/bin/sh
+#
+# These are not ash tests, we use ash as a way to test lineedit!
+#
+# Copyright 2010 by Denys Vlasenko
+# Licensed under GPL v2, see file LICENSE for details.
+
+. ./testing.sh
+
+# testing "test name" "options" "expected result" "file input" "stdin"
+
+testing "One byte which is not valid unicode char followed by valid input" \
+	"script -q -c 'ash' /dev/null >/dev/null; cat output; rm output" \
+	"\
+00000000  3f 2d 0a                                          |?-.|
+00000003
+" \
+	"" \
+	"echo \xff- | hexdump -C >output; exit; exit; exit; exit\n" \
+
+testing "30 bytes which are not valid unicode chars followed by valid input" \
+	"script -q -c 'ash' /dev/null >/dev/null; cat output; rm output" \
+	"\
+00000000  3f 3f 3f 3f 3f 3f 3f 3f  3f 3f 3f 3f 3f 3f 3f 3f  |????????????????|
+00000010  3f 3f 3f 3f 3f 3f 3f 3f  3f 3f 3f 3f 3f 3f 2d 0a  |??????????????-.|
+00000020
+" \
+	"" \
+	"echo \xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff- | hexdump -C >output; exit; exit; exit; exit\n" \
+
+# Not sure this behavior is perfect: we lose all invalid input which precedes
+# arrow keys and such. In this example, \xff\xff are lost
+testing "2 bytes which are not valid unicode chars followed by left arrow key" \
+	"script -q -c 'ash' /dev/null >/dev/null; cat output; rm output" \
+	"\
+00000000  3d 2d 0a                                          |=-.|
+00000003
+" \
+	"" \
+	"echo =+\xff\xff\x1b\x5b\x44- | hexdump -C >output; exit; exit; exit; exit\n" \
+
+exit $FAILCOUNT