summaryrefslogtreecommitdiffstats
path: root/parser/htmlparser/nsHTMLEntities.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'parser/htmlparser/nsHTMLEntities.cpp')
-rw-r--r--parser/htmlparser/nsHTMLEntities.cpp205
1 files changed, 205 insertions, 0 deletions
diff --git a/parser/htmlparser/nsHTMLEntities.cpp b/parser/htmlparser/nsHTMLEntities.cpp
new file mode 100644
index 000000000..e8365c21f
--- /dev/null
+++ b/parser/htmlparser/nsHTMLEntities.cpp
@@ -0,0 +1,205 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "mozilla/ArrayUtils.h"
+
+#include "nsHTMLEntities.h"
+
+#include "nsString.h"
+#include "nsCRT.h"
+#include "PLDHashTable.h"
+
+using namespace mozilla;
+
+struct EntityNode {
+ const char* mStr; // never owns buffer
+ int32_t mUnicode;
+};
+
+struct EntityNodeEntry : public PLDHashEntryHdr
+{
+ const EntityNode* node;
+};
+
+static bool matchNodeString(const PLDHashEntryHdr* aHdr, const void* key)
+{
+ const EntityNodeEntry* entry = static_cast<const EntityNodeEntry*>(aHdr);
+ const char* str = static_cast<const char*>(key);
+ return (nsCRT::strcmp(entry->node->mStr, str) == 0);
+}
+
+static bool matchNodeUnicode(const PLDHashEntryHdr* aHdr, const void* key)
+{
+ const EntityNodeEntry* entry = static_cast<const EntityNodeEntry*>(aHdr);
+ const int32_t ucode = NS_PTR_TO_INT32(key);
+ return (entry->node->mUnicode == ucode);
+}
+
+static PLDHashNumber hashUnicodeValue(const void* key)
+{
+ // key is actually the unicode value
+ return PLDHashNumber(NS_PTR_TO_INT32(key));
+}
+
+
+static const PLDHashTableOps EntityToUnicodeOps = {
+ PLDHashTable::HashStringKey,
+ matchNodeString,
+ PLDHashTable::MoveEntryStub,
+ PLDHashTable::ClearEntryStub,
+ nullptr,
+};
+
+static const PLDHashTableOps UnicodeToEntityOps = {
+ hashUnicodeValue,
+ matchNodeUnicode,
+ PLDHashTable::MoveEntryStub,
+ PLDHashTable::ClearEntryStub,
+ nullptr,
+};
+
+static PLDHashTable* gEntityToUnicode;
+static PLDHashTable* gUnicodeToEntity;
+static nsrefcnt gTableRefCnt = 0;
+
+#define HTML_ENTITY(_name, _value) { #_name, _value },
+static const EntityNode gEntityArray[] = {
+#include "nsHTMLEntityList.h"
+};
+#undef HTML_ENTITY
+
+#define NS_HTML_ENTITY_COUNT ((int32_t)ArrayLength(gEntityArray))
+
+nsresult
+nsHTMLEntities::AddRefTable(void)
+{
+ if (!gTableRefCnt) {
+ gEntityToUnicode = new PLDHashTable(&EntityToUnicodeOps,
+ sizeof(EntityNodeEntry),
+ NS_HTML_ENTITY_COUNT);
+ gUnicodeToEntity = new PLDHashTable(&UnicodeToEntityOps,
+ sizeof(EntityNodeEntry),
+ NS_HTML_ENTITY_COUNT);
+ for (const EntityNode *node = gEntityArray,
+ *node_end = ArrayEnd(gEntityArray);
+ node < node_end; ++node) {
+
+ // add to Entity->Unicode table
+ auto entry = static_cast<EntityNodeEntry*>
+ (gEntityToUnicode->Add(node->mStr, fallible));
+ NS_ASSERTION(entry, "Error adding an entry");
+ // Prefer earlier entries when we have duplication.
+ if (!entry->node)
+ entry->node = node;
+
+ // add to Unicode->Entity table
+ entry = static_cast<EntityNodeEntry*>
+ (gUnicodeToEntity->Add(NS_INT32_TO_PTR(node->mUnicode),
+ fallible));
+ NS_ASSERTION(entry, "Error adding an entry");
+ // Prefer earlier entries when we have duplication.
+ if (!entry->node)
+ entry->node = node;
+ }
+#ifdef DEBUG
+ gUnicodeToEntity->MarkImmutable();
+ gEntityToUnicode->MarkImmutable();
+#endif
+ }
+ ++gTableRefCnt;
+ return NS_OK;
+}
+
+void
+nsHTMLEntities::ReleaseTable(void)
+{
+ if (--gTableRefCnt != 0) {
+ return;
+ }
+
+ delete gEntityToUnicode;
+ delete gUnicodeToEntity;
+ gEntityToUnicode = nullptr;
+ gUnicodeToEntity = nullptr;
+}
+
+int32_t
+nsHTMLEntities::EntityToUnicode(const nsCString& aEntity)
+{
+ NS_ASSERTION(gEntityToUnicode, "no lookup table, needs addref");
+ if (!gEntityToUnicode) {
+ return -1;
+ }
+
+ //this little piece of code exists because entities may or may not have the terminating ';'.
+ //if we see it, strip if off for this test...
+
+ if(';'==aEntity.Last()) {
+ nsAutoCString temp(aEntity);
+ temp.Truncate(aEntity.Length()-1);
+ return EntityToUnicode(temp);
+ }
+
+ auto entry =
+ static_cast<EntityNodeEntry*>(gEntityToUnicode->Search(aEntity.get()));
+
+ return entry ? entry->node->mUnicode : -1;
+}
+
+
+int32_t
+nsHTMLEntities::EntityToUnicode(const nsAString& aEntity) {
+ nsAutoCString theEntity; theEntity.AssignWithConversion(aEntity);
+ if(';'==theEntity.Last()) {
+ theEntity.Truncate(theEntity.Length()-1);
+ }
+
+ return EntityToUnicode(theEntity);
+}
+
+
+const char*
+nsHTMLEntities::UnicodeToEntity(int32_t aUnicode)
+{
+ NS_ASSERTION(gUnicodeToEntity, "no lookup table, needs addref");
+ auto entry =
+ static_cast<EntityNodeEntry*>
+ (gUnicodeToEntity->Search(NS_INT32_TO_PTR(aUnicode)));
+
+ return entry ? entry->node->mStr : nullptr;
+}
+
+#ifdef DEBUG
+#include <stdio.h>
+
+class nsTestEntityTable {
+public:
+ nsTestEntityTable() {
+ int32_t value;
+ nsHTMLEntities::AddRefTable();
+
+ // Make sure we can find everything we are supposed to
+ for (int i = 0; i < NS_HTML_ENTITY_COUNT; ++i) {
+ nsAutoString entity; entity.AssignWithConversion(gEntityArray[i].mStr);
+
+ value = nsHTMLEntities::EntityToUnicode(entity);
+ NS_ASSERTION(value != -1, "can't find entity");
+ NS_ASSERTION(value == gEntityArray[i].mUnicode, "bad unicode value");
+
+ entity.AssignWithConversion(nsHTMLEntities::UnicodeToEntity(value));
+ NS_ASSERTION(entity.EqualsASCII(gEntityArray[i].mStr), "bad entity name");
+ }
+
+ // Make sure we don't find things that aren't there
+ value = nsHTMLEntities::EntityToUnicode(nsAutoCString("@"));
+ NS_ASSERTION(value == -1, "found @");
+ value = nsHTMLEntities::EntityToUnicode(nsAutoCString("zzzzz"));
+ NS_ASSERTION(value == -1, "found zzzzz");
+ nsHTMLEntities::ReleaseTable();
+ }
+};
+//nsTestEntityTable validateEntityTable;
+#endif
+