Speed up redisplay of binary files with long series of nulls

* src/bidi.c (bidi_resolve_weak): Avoid entering a loop searching
for a character needed for resolving the type of a series of BN
and ET characters, as required by rule W5 of UAX#9, if the results
of the resolution are known in advance, because we are at level
zero, and the previous strong character was L.
(bidi_resolve_neutral): Partially resurrect the optimization for a
long series of control characters in an otherwise strictly L2R
text.
(bidi_level_of_next_char): Don't enter the loop that searches for
a paragraph separator if the current character is already at base
embedding level.  (Bug#22739)

diff --git a/src/bidi.c b/src/bidi.c
index e7787054e236b4b8313c962961bb02fba35b49ea..9797517e420aec2cb22e4410776b8c2e57546cdc 100644 (file)
--- a/src/bidi.c
+++ b/src/bidi.c
@@ -2318,7 +2318,31 @@ bidi_resolve_weak (struct bidi_it *bidi_it)
              if (bidi_it->next_en_type == WEAK_EN) /* ET/BN with EN after it */
                type = WEAK_EN;
            }
-         else if (bidi_it->next_en_pos >=0)
+         else if (type == WEAK_BN
+                  /* This condition is for the following important case:
+
+                     . we are at level zero
+                     . either previous strong character was L,
+                        or we've seen no strong characters since sos
+                        and the base paragraph direction is L2R
+                     . this BN is NOT a bidi directional control
+
+                     For such a situation, either this BN will be
+                     converted to EN per W5, and then to L by virtue
+                     of W7; or it will become ON per W6, and then L
+                     because of N1/N2.  So we take a shortcut here
+                     and make it L right away, to avoid the
+                     potentially costly loop below.  This is
+                     important when the buffer has a long series of
+                     control characters, like binary nulls, and no
+                     R2L characters at all.  */
+                  && new_level == 0
+                  && !bidi_explicit_dir_char (bidi_it->ch)
+                  && ((bidi_it->last_strong.type == STRONG_L)
+                      || (bidi_it->last_strong.type == UNKNOWN_BT
+                          && bidi_it->sos == L2R)))
+           type = STRONG_L;
+         else if (bidi_it->next_en_pos >= 0)
            {
              /* We overstepped the last known position for ET
                 resolution but there could be other such characters
@@ -2981,9 +3005,10 @@ bidi_resolve_neutral (struct bidi_it *bidi_it)
         entering the expensive loop in the "else" clause.  */
       else if (current_level == 0
               && bidi_it->prev_for_neutral.type == STRONG_L
-              && type != WEAK_BN
-              && !bidi_explicit_dir_char (bidi_it->ch)
-              && !bidi_isolate_fmt_char (type))
+              && (ASCII_CHAR_P (bidi_it->ch)
+                  || (type != WEAK_BN
+                      && !bidi_explicit_dir_char (bidi_it->ch)
+                      && !bidi_isolate_fmt_char (type))))
        type = bidi_resolve_neutral_1 (bidi_it->prev_for_neutral.type,
                                       STRONG_L, current_level);
       else if (/* current level is 1 */
@@ -3163,7 +3188,7 @@ bidi_level_of_next_char (struct bidi_it *bidi_it)
        }
     }
 
-  /* Perhaps the character we want is already cached s fully resolved.
+  /* Perhaps the character we want is already cached as fully resolved.
      If it is, the call to bidi_cache_find below will return a type
      other than UNKNOWN_BT.  */
   if (bidi_cache_idx > bidi_cache_start && !bidi_it->first_elt)
@@ -3223,7 +3248,10 @@ bidi_level_of_next_char (struct bidi_it *bidi_it)
   if ((bidi_it->orig_type == NEUTRAL_WS
        || bidi_it->orig_type == WEAK_BN
        || bidi_isolate_fmt_char (bidi_it->orig_type))
-      && bidi_it->next_for_ws.charpos < bidi_it->charpos)
+      && bidi_it->next_for_ws.charpos < bidi_it->charpos
+      /* If this character is already at base level, we don't need to
+        reset it, so avoid the potentially costly loop below.  */
+      && level != bidi_it->level_stack[0].level)
     {
       int ch;
       ptrdiff_t clen = bidi_it->ch_len;