summaryrefslogtreecommitdiff
path: root/src/strgen
diff options
context:
space:
mode:
Diffstat (limited to 'src/strgen')
-rw-r--r--src/strgen/strgen.cpp128
1 files changed, 64 insertions, 64 deletions
diff --git a/src/strgen/strgen.cpp b/src/strgen/strgen.cpp
index 0cb2d02eb..04a59b387 100644
--- a/src/strgen/strgen.cpp
+++ b/src/strgen/strgen.cpp
@@ -93,12 +93,12 @@ static uint8 _lang_newgrflangid;
static char _genders[MAX_NUM_GENDER][16];
static uint _numgenders;
-// contains the name of all cases.
+/* contains the name of all cases. */
#define MAX_NUM_CASES 50
static char _cases[MAX_NUM_CASES][16];
static uint _numcases;
-// for each plural value, this is the number of plural forms.
+/* for each plural value, this is the number of plural forms. */
static const byte _plural_form_counts[] = { 2, 1, 2, 3, 3, 3, 3, 3, 4, 2 };
static const char *_cur_ident;
@@ -114,7 +114,7 @@ struct ParsedCommandStruct {
const CmdStruct *cmd[32]; // ordered by param #
};
-// Used when generating some advanced commands.
+/* Used when generating some advanced commands. */
static ParsedCommandStruct _cur_pcs;
static int _cur_argidx;
@@ -271,11 +271,11 @@ static void EmitSetXY(char *buf, int value)
PutByte((byte)y);
}
-// The plural specifier looks like
-// {NUM} {PLURAL -1 passenger passengers} then it picks either passenger/passengers depending on the count in NUM
+/* The plural specifier looks like
+ * {NUM} {PLURAL -1 passenger passengers} then it picks either passenger/passengers depending on the count in NUM */
-// This is encoded like
-// CommandByte <ARG#> <NUM> {Length of each string} {each string}
+/* This is encoded like
+ * CommandByte <ARG#> <NUM> {Length of each string} {each string} */
bool ParseRelNum(char **buf, int *value)
{
@@ -300,7 +300,7 @@ bool ParseRelNum(char **buf, int *value)
return true;
}
-// Parse out the next word, or NULL
+/* Parse out the next word, or NULL */
char *ParseWord(char **buf)
{
char *s = *buf, *r;
@@ -310,7 +310,7 @@ char *ParseWord(char **buf)
if (*s == '"') {
r = ++s;
- // parse until next " or NUL
+ /* parse until next " or NUL */
for (;;) {
if (*s == '\0') break;
if (*s == '"') {
@@ -320,7 +320,7 @@ char *ParseWord(char **buf)
s++;
}
} else {
- // proceed until whitespace or NUL
+ /* proceed until whitespace or NUL */
r = s;
for (;;) {
if (*s == '\0') break;
@@ -335,7 +335,7 @@ char *ParseWord(char **buf)
return r;
}
-// Forward declaration
+/* Forward declaration */
static int TranslateArgumentIdx(int arg);
static void EmitWordList(const char * const *words, uint nw)
@@ -356,10 +356,10 @@ static void EmitPlural(char *buf, int value)
const char *words[5];
int nw = 0;
- // Parse out the number, if one exists. Otherwise default to prev arg.
+ /* Parse out the number, if one exists. Otherwise default to prev arg. */
if (!ParseRelNum(&buf, &argidx)) argidx--;
- // Parse each string
+ /* Parse each string */
for (nw = 0; nw < 5; nw++) {
words[nw] = ParseWord(&buf);
if (words[nw] == NULL) break;
@@ -398,19 +398,19 @@ static void EmitGender(char *buf, int value)
if (buf[0] == '=') {
buf++;
- // This is a {G=DER} command
+ /* This is a {G=DER} command */
for (nw = 0; ; nw++) {
if (nw >= 8) error("G argument '%s' invalid", buf);
if (strcmp(buf, _genders[nw]) == 0) break;
}
- // now nw contains the gender index
+ /* now nw contains the gender index */
PutUtf8(SCC_GENDER_INDEX);
PutByte(nw);
} else {
const char *words[8];
- // This is a {G 0 foo bar two} command.
- // If no relative number exists, default to +0
+ /* This is a {G 0 foo bar two} command.
+ * If no relative number exists, default to +0 */
if (!ParseRelNum(&buf, &argidx)) {}
for (nw = 0; nw < 8; nw++) {
@@ -426,15 +426,15 @@ static void EmitGender(char *buf, int value)
static const CmdStruct _cmd_structs[] = {
- // Update position
+ /* Update position */
{"SETX", EmitSetX, SCC_SETX, 0, 0},
{"SETXY", EmitSetXY, SCC_SETXY, 0, 0},
- // Font size
+ /* Font size */
{"TINYFONT", EmitSingleChar, SCC_TINYFONT, 0, 0},
{"BIGFONT", EmitSingleChar, SCC_BIGFONT, 0, 0},
- // Colors
+ /* Colors */
{"BLUE", EmitSingleChar, SCC_BLUE, 0, 0},
{"SILVER", EmitSingleChar, SCC_SILVER, 0, 0},
{"GOLD", EmitSingleChar, SCC_GOLD, 0, 0},
@@ -487,7 +487,7 @@ static const CmdStruct _cmd_structs[] = {
{"STRING", EmitSingleChar, SCC_STRING, 1, C_CASE},
{"RAW_STRING", EmitSingleChar, SCC_RAW_STRING_POINTER, 1, 0},
- // Numbers
+ /* Numbers */
{"COMMA", EmitSingleChar, SCC_COMMA, 1, 0}, // Number with comma
{"NUM", EmitSingleChar, SCC_NUM, 1, 0}, // Signed number
{"BYTES", EmitSingleChar, SCC_BYTES, 1, 0}, // Unsigned number with "bytes", i.e. "1.02 MiB or 123 KiB"
@@ -565,8 +565,8 @@ static uint ResolveCaseName(const char *str, uint len)
}
-// returns NULL on eof
-// else returns command struct
+/* returns NULL on eof
+ * else returns command struct */
static const CmdStruct *ParseCommandString(const char **str, char *param, int *argno, int *casei)
{
const char *s = *str, *start;
@@ -576,7 +576,7 @@ static const CmdStruct *ParseCommandString(const char **str, char *param, int *a
*argno = -1;
*casei = -1;
- // Scan to the next command, exit if there's no next command.
+ /* Scan to the next command, exit if there's no next command. */
for (; *s != '{'; s++) {
if (*s == '\0') return NULL;
}
@@ -590,7 +590,7 @@ static const CmdStruct *ParseCommandString(const char **str, char *param, int *a
s = end + 1;
}
- // parse command name
+ /* parse command name */
start = s;
do {
c = *s++;
@@ -620,7 +620,7 @@ static const CmdStruct *ParseCommandString(const char **str, char *param, int *a
if (c != '}') {
if (c == '=') s--;
- // copy params
+ /* copy params */
start = s;
for (;;) {
c = *s++;
@@ -714,12 +714,12 @@ static void ExtractCommandString(ParsedCommandStruct *p, const char *s, bool war
memset(p, 0, sizeof(*p));
for (;;) {
- // read until next command from a.
+ /* read until next command from a. */
const CmdStruct *ar = ParseCommandString(&s, param, &argno, &casei);
if (ar == NULL) break;
- // Sanity checking
+ /* Sanity checking */
if (argno != -1 && ar->consumes == 0) error("Non consumer param can't have a paramindex");
if (ar->consumes) {
@@ -767,19 +767,19 @@ static bool CheckCommandsMatch(char *a, char *b, const char *name)
ExtractCommandString(&templ, b, true);
ExtractCommandString(&lang, a, true);
- // For each string in templ, see if we find it in lang
+ /* For each string in templ, see if we find it in lang */
if (templ.np != lang.np) {
strgen_warning("%s: template string and language string have a different # of commands", name);
result = false;
}
for (i = 0; i < templ.np; i++) {
- // see if we find it in lang, and zero it out
+ /* see if we find it in lang, and zero it out */
bool found = false;
for (j = 0; j < lang.np; j++) {
if (templ.pairs[i].a == lang.pairs[j].a &&
strcmp(templ.pairs[i].v, lang.pairs[j].v) == 0) {
- // it was found in both. zero it out from lang so we don't find it again
+ /* it was found in both. zero it out from lang so we don't find it again */
lang.pairs[j].a = NULL;
found = true;
break;
@@ -792,8 +792,8 @@ static bool CheckCommandsMatch(char *a, char *b, const char *name)
}
}
- // if we reach here, all non consumer commands match up.
- // Check if the non consumer commands match up also.
+ /* if we reach here, all non consumer commands match up.
+ * Check if the non consumer commands match up also. */
for (i = 0; i < lengthof(templ.cmd); i++) {
if (TranslateCmdForCompare(templ.cmd[i]) != TranslateCmdForCompare(lang.cmd[i])) {
strgen_warning("%s: Param idx #%d '%s' doesn't match with template command '%s'", name, i,
@@ -817,7 +817,7 @@ static void HandleString(char *str, bool master)
return;
}
- // Ignore comments & blank lines
+ /* Ignore comments & blank lines */
if (*str == ';' || *str == ' ' || *str == '\0') return;
s = strchr(str, ':');
@@ -826,8 +826,8 @@ static void HandleString(char *str, bool master)
return;
}
- // Trim spaces.
- // After this str points to the command name, and s points to the command contents
+ /* Trim spaces.
+ * After this str points to the command name, and s points to the command contents */
for (t = s; t > str && (t[-1] == ' ' || t[-1] == '\t'); t--);
*t = 0;
s++;
@@ -842,12 +842,12 @@ static void HandleString(char *str, bool master)
}
}
- // Check if the string has a case..
- // The syntax for cases is IDENTNAME.case
+ /* Check if the string has a case..
+ * The syntax for cases is IDENTNAME.case */
casep = strchr(str, '.');
if (casep) *casep++ = 0;
- // Check if this string already exists..
+ /* Check if this string already exists.. */
ent = HashFind(str);
if (master) {
@@ -867,7 +867,7 @@ static void HandleString(char *str, bool master)
return;
}
- // Allocate a new LangString
+ /* Allocate a new LangString */
ent = CallocT<LangString>(1);
_strings[_next_string_id] = ent;
ent->index = _next_string_id++;
@@ -900,10 +900,10 @@ static void HandleString(char *str, bool master)
}
if (s[0] == ':' && s[1] == '\0' && casep == NULL) {
- // Special syntax :: means we should just inherit the master string
+ /* Special syntax :: means we should just inherit the master string */
ent->translated = strdup(ent->english);
} else {
- // make sure that the commands match
+ /* make sure that the commands match */
if (!CheckCommandsMatch(s, ent->english, str)) return;
if (casep != NULL) {
@@ -942,8 +942,8 @@ static void ParseFile(const char *file, bool english)
_lang_textdir = TD_LTR;
_lang_winlangid = 0x0000; // neutral language code
_lang_newgrflangid = 0; // standard english
- // TODO:!! We can't reset the cases. In case the translated strings
- // derive some strings from english....
+ /* TODO:!! We can't reset the cases. In case the translated strings
+ * derive some strings from english.... */
in = fopen(file, "r");
if (in == NULL) error("Cannot open file");
@@ -971,7 +971,7 @@ static uint32 MyHashStr(uint32 hash, const char *s)
}
-// make a hash of the file to get a unique "version number"
+/* make a hash of the file to get a unique "version number" */
static void MakeHashOfStrings()
{
uint32 hash = 0;
@@ -1079,10 +1079,10 @@ static void WriteStringsH(const char *filename)
fclose(out);
if (CompareFiles("tmp.xxx", filename)) {
- // files are equal. tmp.xxx is not needed
+ /* files are equal. tmp.xxx is not needed */
unlink("tmp.xxx");
} else {
- // else rename tmp.xxx into filename
+ /* else rename tmp.xxx into filename */
#if defined(WIN32) || defined(WIN64)
unlink(filename);
#endif
@@ -1122,7 +1122,7 @@ static void PutCommandString(const char *str)
_cur_argidx = 0;
while (*str != '\0') {
- // Process characters as they are until we encounter a {
+ /* Process characters as they are until we encounter a { */
if (*str != '{') {
PutByte(*str++);
continue;
@@ -1135,15 +1135,15 @@ static void PutCommandString(const char *str)
PutByte(casei);
}
- // For params that consume values, we need to handle the argindex properly
+ /* For params that consume values, we need to handle the argindex properly */
if (cs->consumes > 0) {
- // Check if we need to output a move-param command
+ /* Check if we need to output a move-param command */
if (argno != -1 && argno != _cur_argidx) {
_cur_argidx = argno;
PutArgidxCommand();
}
- // Output the one from the master string... it's always accurate.
+ /* Output the one from the master string... it's always accurate. */
cs = _cur_pcs.cmd[_cur_argidx++];
if (cs == NULL) {
error("%s: No argument exists at position %d", _cur_ident, _cur_argidx - 1);
@@ -1186,7 +1186,7 @@ static void WriteLangfile(const char *filename)
hdr.offsets[i] = TO_LE16(n);
}
- // see line 655: fprintf(..."\tLANGUAGE_PACK_IDENT = 0x474E414C,...)
+ /* see line 655: fprintf(..."\tLANGUAGE_PACK_IDENT = 0x474E414C,...) */
hdr.ident = TO_LE32(0x474E414C); // Big Endian value for 'LANG'
hdr.version = TO_LE32(_hash);
hdr.plural_form = _lang_pluralform;
@@ -1205,7 +1205,7 @@ static void WriteLangfile(const char *filename)
const Case *casep;
const char *cmdp;
- // For undefined strings, just set that it's an empty string
+ /* For undefined strings, just set that it's an empty string */
if (ls == NULL) {
WriteLength(f, 0);
continue;
@@ -1214,7 +1214,7 @@ static void WriteLangfile(const char *filename)
_cur_ident = ls->name;
_cur_line = ls->line;
- // Produce a message if a string doesn't have a translation.
+ /* Produce a message if a string doesn't have a translation. */
if (_show_todo > 0 && ls->translated == NULL) {
if ((_show_todo & 2) != 0) {
strgen_warning("'%s' is untranslated", ls->name);
@@ -1225,7 +1225,7 @@ static void WriteLangfile(const char *filename)
}
}
- // Extract the strings and stuff from the english command string
+ /* Extract the strings and stuff from the english command string */
ExtractCommandString(&_cur_pcs, ls->english, false);
if (ls->translated_case != NULL || ls->translated != NULL) {
@@ -1242,28 +1242,28 @@ static void WriteLangfile(const char *filename)
const Case *c;
uint num;
- // Need to output a case-switch.
- // It has this format
- // <0x9E> <NUM CASES> <CASE1> <LEN1> <STRING1> <CASE2> <LEN2> <STRING2> <CASE3> <LEN3> <STRING3> <STRINGDEFAULT>
- // Each LEN is printed using 2 bytes in big endian order.
+ /* Need to output a case-switch.
+ * It has this format
+ * <0x9E> <NUM CASES> <CASE1> <LEN1> <STRING1> <CASE2> <LEN2> <STRING2> <CASE3> <LEN3> <STRING3> <STRINGDEFAULT>
+ * Each LEN is printed using 2 bytes in big endian order. */
PutUtf8(SCC_SWITCH_CASE);
- // Count the number of cases
+ /* Count the number of cases */
for (num = 0, c = casep; c; c = c->next) num++;
PutByte(num);
- // Write each case
+ /* Write each case */
for (c = casep; c != NULL; c = c->next) {
int pos;
PutByte(c->caseidx);
- // Make some space for the 16-bit length
+ /* Make some space for the 16-bit length */
pos = _put_pos;
PutByte(0);
PutByte(0);
- // Write string
+ /* Write string */
PutCommandString(c->string);
PutByte(0); // terminate with a zero
- // Fill in the length
+ /* Fill in the length */
_put_buf[pos + 0] = GB(_put_pos - (pos + 2), 8, 8);
_put_buf[pos + 1] = GB(_put_pos - (pos + 2), 0, 8);
}