tags: Use a perfect hash for lookups

The previous version using `strcasecmp` over an array was a bottleneck
on the library.

This version uses a simple, minimal perfect hash table (computed via
`mph`) to convert tag names into strings. Since we're now hashing tag
names, we can pass in the length of the tag name explicitly, and avoid
the superfluous allocations that the tokenizer was performing in order
to NULL-terminate the tag. This is implemented on the new
`gumbo_tagn_enum` API.

The old `gumbo_tag_enum` API has been left as a thin wrapper to keep
backwards compatibility -- it is not used internally by the library.

`mph` was chosen for the perfect hash function because it generates
hashes that are slightly slower than GPerf but significantly simpler,
and occuppying an order of magnitude less memory (as they don't
need a full copy of all the strings in the set for hashing).

If the tag lookup function proves to be a bottleneck, this decision can
be re-evaluated in the future.

Author: Vicent Marti

src/gumbo.h

@@ -365,9 +365,10 @@ const char* gumbo_normalize_svg_tagname(const GumboStringPiece* tagname);
 
 /**
  * Converts a tag name string (which may be in upper or mixed case) to a tag
- * enum.
+ * enum. The `tag` version expects `tagname` to be NULL-terminated
  */
 GumboTag gumbo_tag_enum(const char* tagname);
+GumboTag gumbo_tagn_enum(const char* tagname, int length);
 
 /**
  * Attribute namespaces.

src/tag.c

@@ -19,6 +19,7 @@
 #include <assert.h>
 #include <ctype.h>
 #include <strings.h>    // For strcasecmp.
+#include <string.h>    // For strcasecmp.
 
 // NOTE(jdtang): Keep this in sync with the GumboTag enum in the header.
 // TODO(jdtang): Investigate whether there're efficiency benefits to putting the
@@ -213,6 +214,7 @@ void gumbo_tag_from_original_text(GumboStringPiece* text) {
   }
 }
 
+#ifdef SLOW_TAG_LOOKUP
 GumboTag gumbo_tag_enum(const char* tagname) {
   for (int i = 0; i < GUMBO_TAG_LAST; ++i) {
     // TODO(jdtang): strcasecmp is non-portable, so if we want to support
@@ -224,3 +226,114 @@ GumboTag gumbo_tag_enum(const char* tagname) {
   }
   return GUMBO_TAG_UNKNOWN;
 }
+#else
+
+/*
+ * Generated with `mph`
+ * ./mph -d2 -m2 -c1.33 < tag.in | emitc -s -l
+ */
+static int hash_tag(const unsigned char *kp, int len)
+{
+  static short g[] = {
+    87, -1, -1, 54, 37, -1, 0, 63, -1, 4,
+    87, 132, 149, -1, 43, 103, 78, 89, 126, 74,
+    9, -1, 32, 68, 46, 132, 14, -1, -1, 147,
+    77, 120, 101, 138, 38, -1, 135, 24, 94, -1,
+    36, 88, 101, 29, -1, 83, 122, -1, 126, 148,
+    145, 46, 90, 94, 83, 140, -1, 4, -1, 103,
+    25, 0, 0, 129, 138, 0, 138, 53, -1, 0,
+    77, 43, 0, -1, 90, 22, 30, 109, 71, 1,
+    -1, 94, 20, -1, 27, 56, 0, 21, 72, 122,
+    -1, -1, 0, 142, 72, 5, 11, 7, 43, 111,
+    89, 96, 81, 48, 65, 27, 5, 73, -1, 57,
+    137, 52, 0, 60, -1, 3, -1, 100, 149, 41,
+    98, 118, 81, 0, 50, 30, -1, -1, 83, 10,
+    20, 25, 2, 0, 118, 9, 39, 94, 35, 42,
+    23, 75, 89, 31, 0, 148, 86, 6, 115, -1,
+    49, 107, 5, 90, 4, 12, -1, 21, 16, -1,
+    29, 39, -1, 96, 111, 96, 43, 43, 120, -1,
+    46, 84, -1, 0, 146, 126, 24, -1, 28, 110,
+    82, 42, 12, 84, -1, -1, -1, 0, 33, 12,
+    86, 93, -1, 147, 95, 58, 90, 145, -1, -1,
+  };
+
+  static unsigned char T0[] = {
+    196, 103, 27, 185, 60, 0, 58, 36, 180, 118,
+    101, 180, 61, 125, 144, 167, 140, 104, 131, 195,
+    176, 62, 79, 175, 195, 103, 116, 194, 122, 73,
+    44, 119, 128, 23, 56, 188, 23, 114, 24, 156,
+    32, 78, 136, 46, 3, 32, 165, 95, 136, 97,
+    90, 65, 111, 121, 40, 106, 25, 108, 53, 99,
+    181, 49, 18, 110, 72, 74, 50, 48, 141, 27,
+    4, 125, 105, 92, 171, 60, 124, 1, 72, 96,
+    178, 59, 58, 61, 0, 185, 12, 176, 111, 121,
+    49, 170, 70, 48, 43, 82, 178, 157, 34, 62,
+    137, 148, 110, 160, 96, 11, 50, 22, 12, 74,
+    71, 143, 133, 129, 4, 86, 67, 168, 62, 130,
+    41, 63, 101, 63, 112, 96, 146, 90, 5, 132,
+    153, 95, 32, 15, 7, 80, 26, 57, 103, 191,
+    83, 126, 134, 169, 55, 90, 55, 74, 58, 69,
+    5, 99, 132, 58,
+  };
+
+  static unsigned char T1[] = {
+    87, 14, 91, 162, 194, 198, 131, 1, 89, 2,
+    154, 17, 98, 25, 7, 121, 145, 178, 28, 70,
+    94, 135, 77, 129, 134, 137, 69, 128, 88, 126,
+    114, 175, 92, 5, 89, 87, 3, 20, 88, 44,
+    174, 194, 14, 73, 171, 21, 194, 117, 151, 175,
+    139, 45, 110, 17, 127, 196, 106, 148, 124, 194,
+    26, 190, 169, 118, 195, 59, 157, 150, 31, 197,
+    147, 6, 143, 161, 79, 67, 134, 68, 163, 61,
+    104, 124, 56, 39, 115, 99, 140, 101, 63, 91,
+    124, 4, 134, 110, 132, 61, 150, 96, 116, 167,
+    80, 174, 115, 169, 14, 184, 24, 47, 4, 188,
+    60, 109, 64, 68, 148, 179, 168, 41, 80, 183,
+    84, 156, 187, 18, 18, 119, 79, 169, 168, 148,
+    88, 0, 122, 3, 169, 88, 139, 146, 88, 144,
+    86, 148, 5, 150, 17, 105, 81, 137, 98, 113,
+    120, 182, 69, 107, 
+  };
+
+	int i, n;
+	unsigned int f0, f1;
+
+	if (len < 1 || len > 14)
+		return -1;
+
+	for (i=-45, f0=f1=0, n=0; n < len; ++n) {
+    int c = tolower(kp[n]);
+    if (c < 45 || c > 121)
+			return -1;
+		f0 += T0[i + c];
+		f1 += T1[i + c];
+		i += 77;
+		if (i >= 109)
+			i = -45;
+	}
+	return (g[f0 % 200] + g[f1 % 200]) % 150;
+}
+
+static int
+case_memcmp(const char *s1, const char *s2, int n)
+{
+	while (n--) {
+		unsigned char c1 = tolower(*s1++);
+		unsigned char c2 = tolower(*s2++);
+		if (c1 != c2)
+			return (int)c1 - (int)c2;
+	}
+	return 0;
+}
+
+GumboTag gumbo_tagn_enum(const char* tagname, int length) {
+  int position = hash_tag((const unsigned char *)tagname, length);
+  if (position >= 0 && !case_memcmp(tagname, kGumboTagNames[position], length))
+    return (GumboTag)position;
+  return GUMBO_TAG_UNKNOWN;
+}
+
+GumboTag gumbo_tag_enum(const char* tagname) {
+  return gumbo_tagn_enum(tagname, strlen(tagname));
+}
+#endif

src/tag.in

@@ -0,0 +1,150 @@
+html
+head
+title
+base
+link
+meta
+style
+script
+noscript
+template
+body
+article
+section
+nav
+aside
+h1
+h2
+h3
+h4
+h5
+h6
+hgroup
+header
+footer
+address
+p
+hr
+pre
+blockquote
+ol
+ul
+li
+dl
+dt
+dd
+figure
+figcaption
+main
+div
+a
+em
+strong
+small
+s
+cite
+q
+dfn
+abbr
+data
+time
+code
+var
+samp
+kbd
+sub
+sup
+i
+b
+u
+mark
+ruby
+rt
+rtc
+rp
+bdi
+bdo
+span
+br
+wbr
+ins
+del
+image
+img
+iframe
+embed
+object
+param
+video
+audio
+source
+track
+canvas
+map
+area
+math
+mi
+mo
+mn
+ms
+mtext
+mglyph
+malignmark
+annotation-xml
+svg
+foreignobject
+desc
+table
+caption
+colgroup
+col
+tbody
+thead
+tfoot
+tr
+td
+th
+form
+fieldset
+legend
+label
+input
+button
+select
+datalist
+optgroup
+option
+textarea
+keygen
+output
+progress
+meter
+details
+summary
+menu
+menuitem
+applet
+acronym
+bgsound
+dir
+frame
+frameset
+noframes
+isindex
+listing
+xmp
+nextid
+noembed
+plaintext
+rb
+strike
+basefont
+big
+blink
+center
+font
+marquee
+multicol
+nobr
+spacer
+tt

src/tokenizer.c

@@ -742,11 +742,9 @@ static void finish_tag_name(GumboParser* parser) {
   GumboTokenizerState* tokenizer = parser->_tokenizer_state;
   GumboTagState* tag_state = &tokenizer->_tag_state;
 
-  const char* temp;
-  copy_over_tag_buffer(parser, &temp);
-  tag_state->_tag = gumbo_tag_enum(temp);
+  tag_state->_tag = gumbo_tagn_enum(
+		tag_state->_buffer.data, tag_state->_buffer.length);
   reinitialize_tag_buffer(parser);
-  gumbo_free((void*) temp);
 }
 
 // Adds an ERR_DUPLICATE_ATTR parse error to the parser's error struct.
@@ -832,13 +830,9 @@ static void finish_attribute_value(GumboParser* parser) {
 static bool is_appropriate_end_tag(GumboParser* parser) {
   GumboTagState* tag_state = &parser->_tokenizer_state->_tag_state;
   assert(!tag_state->_is_start_tag);
-  // Null terminate the current string buffer, so it can be passed to
-  // gumbo_tag_enum, but don't increment the length in case we need to dump the
-  // buffer as character tokens.
-  gumbo_string_buffer_append_codepoint('\0', &tag_state->_buffer);
-  --tag_state->_buffer.length;
   return tag_state->_last_start_tag != GUMBO_TAG_LAST &&
-      tag_state->_last_start_tag == gumbo_tag_enum(tag_state->_buffer.data);
+      tag_state->_last_start_tag == 
+	  gumbo_tagn_enum(tag_state->_buffer.data, tag_state->_buffer.length);
 }
 
 void gumbo_tokenizer_state_init(