Rule alignment fixes for non compliant platforms / Fix for emscripten demo

emscripten/Makefile

@@ -100,7 +100,7 @@ CXXFLAGS+=-I ./ -I ../ -I ../src/include/espeak-ng
 #CXXFLAGS+=-s USE_PTHREADS=1
 
 # NOTE: extra flags for emscripten
-EM_CXXFLAGS=-s RESERVED_FUNCTION_POINTERS=2 --memory-init-file 0 -s FORCE_FILESYSTEM=1 -s WASM=0
+EM_CXXFLAGS=-s RESERVED_FUNCTION_POINTERS=2 --memory-init-file 0 -s FORCE_FILESYSTEM=1 -s WASM=0 -s ERROR_ON_UNDEFINED_SYMBOLS=0
 
 
 

emscripten/post.js

@@ -120,6 +120,8 @@ eSpeakNGWorker.prototype.synthesize_ipa = function (aText, aCallback) {
 if (typeof WorkerGlobalScope !== 'undefined') {
   var worker;
   
+  Module.postRun = Module.postRun || [];
+
   Module.postRun.push(function () {
     worker = new eSpeakNGWorker();
     postMessage('ready');

src/libespeak-ng/dictionary.c

@@ -142,7 +142,7 @@ static void InitGroups(Translator *tr)
 	// a RULE_GROUP_END.
 	if (*p != RULE_GROUP_END) while (*p != 0) {
 		if (*p != RULE_GROUP_START) {
-			fprintf(stderr, "Bad rules data in '%s_dict' at 0x%x (%c)\n", dictionary_name, (unsigned int)(p - tr->data_dictrules), p);
+			fprintf(stderr, "Bad rules data in '%s_dict' at 0x%x (%c)\n", dictionary_name, (unsigned int)(p - tr->data_dictrules), *p);
 			break;
 		}
 		p++;
@@ -150,8 +150,12 @@ static void InitGroups(Translator *tr)
 		if (p[0] == RULE_REPLACEMENTS) {
 			p = (char *)(((intptr_t)p+4) & ~3); // advance to next word boundary
 			tr->langopts.replace_chars = (unsigned char *)p;
-			while (*(unsigned int *)p != 0)
+
+			// Don't cast to (unsigned int), result may not be garanted depending on compiler
+			while ( !is_str_fully_empty(p, 4) ) {
 				p++;
+			}
+
 			while (*p != RULE_GROUP_END) p++;
 			p++;
 			continue;

src/libespeak-ng/readclause.c

@@ -132,6 +132,14 @@ int clause_type_from_codepoint(uint32_t c)
 	return CLAUSE_NONE;
 }
 
+int is_str_fully_empty(const char* str, int size) {
+	// Tests if all bytes of str are null uf to size
+	// This should never be reimplemented with integers, because
+	// this function has to work with unaligned char*
+	// (casting to int when unaligned may result in ungaranteed behaviors)
+	return *str || memcmp(str, str+1, size-1);
+}
+
 int towlower2(unsigned int c, Translator *translator)
 {
 	// check for non-standard upper to lower case conversions

src/libespeak-ng/readclause.h

@@ -34,6 +34,9 @@ typedef struct {
 
 extern PARAM_STACK param_stack[];
 
+// Tests if all bytes of str up to size are null
+int is_str_fully_empty(const char* str, int size);
+
 int clause_type_from_codepoint(uint32_t c);
 int towlower2(unsigned int c, Translator *translator); // Supports Turkish I
 int Eof(void);
@@ -52,6 +55,7 @@ int ReadClause(Translator *tr,
 
 
 
+
 #ifdef __cplusplus
 }
 #endif

src/libespeak-ng/translate.c

@@ -1793,7 +1793,7 @@ static int EmbeddedCommand(unsigned int *source_index_out)
 static const char *FindReplacementChars(Translator *tr, const char **pfrom, unsigned int c, const char *next, int *ignore_next_n)
 {
 	const char *from = *pfrom;
-	while (*(unsigned int *)from != 0) {
+	while ( !is_str_fully_empty(from, 4) ) {
 		unsigned int fc = 0; // from character
 		unsigned int nc = c; // next character
 		const char *match_next = next;