tweaks to ANSI sequence (OSC) handling (#34964)

Fixes: #34604

Prompted by that I realized we do not correctly recognize both "ST"
sequences we want to recognize, fix that.

src/basic/string-util.c

@@ -698,7 +698,8 @@ char* strip_tab_ansi(char **ibuf, size_t *_isz, size_t highlight[2]) {
                 STATE_OTHER,
                 STATE_ESCAPE,
                 STATE_CSI,
-                STATE_CSO,
+                STATE_OSC,
+                STATE_OSC_CLOSING,
         } state = STATE_OTHER;
         _cleanup_(memstream_done) MemStream m = {};
         size_t isz, shift[2] = {}, n_carriage_returns = 0;
@@ -711,7 +712,7 @@ char* strip_tab_ansi(char **ibuf, size_t *_isz, size_t highlight[2]) {
          *
          * 1. Replaces TABs by 8 spaces
          * 2. Strips ANSI color sequences (a subset of CSI), i.e. ESC '[' … 'm' sequences
-         * 3. Strips ANSI operating system sequences (CSO), i.e. ESC ']' … BEL sequences
+         * 3. Strips ANSI operating system sequences (OSC), i.e. ESC ']' … ST sequences
          * 4. Strip trailing \r characters (since they would "move the cursor", but have no
          *    other effect).
          *
@@ -719,7 +720,7 @@ char* strip_tab_ansi(char **ibuf, size_t *_isz, size_t highlight[2]) {
          * are any other special characters. Truncated ANSI sequences are left-as is too. This call is
          * supposed to suppress the most basic formatting noise, but nothing else.
          *
-         * Why care for CSO sequences? Well, to undo what terminal_urlify() and friends generate. */
+         * Why care for OSC sequences? Well, to undo what terminal_urlify() and friends generate. */
 
         isz = _isz ? *_isz : strlen(*ibuf);
 
@@ -766,8 +767,8 @@ char* strip_tab_ansi(char **ibuf, size_t *_isz, size_t highlight[2]) {
                         } else if (*i == '[') { /* ANSI CSI */
                                 state = STATE_CSI;
                                 begin = i + 1;
-                        } else if (*i == ']') { /* ANSI CSO */
-                                state = STATE_CSO;
+                        } else if (*i == ']') { /* ANSI OSC */
+                                state = STATE_OSC;
                                 begin = i + 1;
                         } else {
                                 fputc('\x1B', f);
@@ -793,17 +794,35 @@ char* strip_tab_ansi(char **ibuf, size_t *_isz, size_t highlight[2]) {
 
                         break;
 
-                case STATE_CSO:
+                case STATE_OSC:
                         assert(n_carriage_returns == 0);
 
+                        /* There are three kinds of OSC terminators: \x07, \x1b\x5c or \x9c. We only support
+                         * the first two, because the last one is a valid UTF-8 codepoint and hence creates
+                         * an ambiguity (many Terminal emulators refuse to support it as well). */
                         if (i >= *ibuf + isz || /* EOT … */
-                            (*i != '\a' && (uint8_t) *i < 32U) || (uint8_t) *i > 126U) { /* … or invalid chars in sequence */
+                            (!IN_SET(*i, '\x07', '\x1b') && (uint8_t) *i < 32U) || (uint8_t) *i > 126U) { /* … or invalid chars in sequence */
                                 fputc('\x1B', f);
                                 fputc(']', f);
                                 advance_offsets(i - *ibuf, highlight, shift, 2);
                                 state = STATE_OTHER;
                                 i = begin-1;
-                        } else if (*i == '\a')
+                        } else if (*i == '\x07') /* Single character ST */
+                                state = STATE_OTHER;
+                        else if (*i == '\x1B')
+                                state = STATE_OSC_CLOSING;
+
+                        break;
+
+                case STATE_OSC_CLOSING:
+                        if (i >= *ibuf + isz || /* EOT … */
+                            *i != '\x5c') { /* … or incomplete two-byte ST in sequence */
+                                fputc('\x1B', f);
+                                fputc(']', f);
+                                advance_offsets(i - *ibuf, highlight, shift, 2);
+                                state = STATE_OTHER;
+                                i = begin-1;
+                        } else if (*i == '\x5c')
                                 state = STATE_OTHER;
 
                         break;

src/basic/terminal-util.c

@@ -1923,12 +1923,12 @@ int get_default_background_color(double *ret_red, double *ret_green, double *ret
         if (tcsetattr(STDIN_FILENO, TCSADRAIN, &new_termios) < 0)
                 return -errno;
 
-        r = loop_write(STDOUT_FILENO, "\x1B]11;?\x07", SIZE_MAX);
+        r = loop_write(STDOUT_FILENO, ANSI_OSC "11;?" ANSI_ST, SIZE_MAX);
         if (r < 0)
                 goto finish;
 
         usec_t end = usec_add(now(CLOCK_MONOTONIC), 333 * USEC_PER_MSEC);
-        char buf[STRLEN("\x1B]11;rgb:0/0/0\x07")]; /* shortest possible reply */
+        char buf[STRLEN(ANSI_OSC "11;rgb:0/0/0" ANSI_ST)]; /* shortest possible reply */
         size_t buf_full = 0;
         BackgroundColorContext context = {};
 

src/basic/terminal-util.h

@@ -27,6 +27,16 @@
 #define ANSI_WINDOW_TITLE_PUSH "\x1b[22;2t"
 #define ANSI_WINDOW_TITLE_POP "\x1b[23;2t"
 
+/* ANSI "string terminator" character ("ST"). Terminal emulators typically allow three different ones: 0x07,
+ * 0x9c, and 0x1B 0x5C. We'll avoid 0x07 (BEL, aka ^G) since it might trigger unexpected TTY signal
+ * handling. And we'll avoid 0x9c since that's also valid regular codepoint in UTF-8 and elsewhere, and
+ * creates ambiguities. Because of that some terminal emulators explicitly choose not to support it. Hence we
+ * use 0x1B 0x5c */
+#define ANSI_ST "\e\\"
+
+/* The "operating system command" ("OSC") start sequence */
+#define ANSI_OSC "\e]"
+
 bool isatty_safe(int fd);
 
 int terminal_reset_defensive(int fd, bool switch_to_text);