summaryrefslogtreecommitdiff
path: root/miscutils
diff options
context:
space:
mode:
authorDenys Vlasenko2018-12-25 21:44:10 +0100
committerDenys Vlasenko2018-12-25 21:44:10 +0100
commit0b0e8d0509bdd45aa600d3ea2bf9043630809cd2 (patch)
treeddf18f4181e26cbf6973024b0c108fab89bba31a /miscutils
parent3f8752c33f29d4cb4cc18fbff1f0da555f04c3ce (diff)
downloadbusybox-0b0e8d0509bdd45aa600d3ea2bf9043630809cd2.zip
busybox-0b0e8d0509bdd45aa600d3ea2bf9043630809cd2.tar.gz
bc: rename some members and macros, no code changes
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
Diffstat (limited to 'miscutils')
-rw-r--r--miscutils/bc.c258
1 files changed, 129 insertions, 129 deletions
diff --git a/miscutils/bc.c b/miscutils/bc.c
index 8fa312a..7a6278c 100644
--- a/miscutils/bc.c
+++ b/miscutils/bc.c
@@ -518,31 +518,31 @@ typedef enum BcLexType {
struct BcLexKeyword {
char name8[8];
};
-#define BC_LEX_KW_ENTRY(a, b) \
+#define LEX_KW_ENTRY(a, b) \
{ .name8 = a /*, .posix = b */ }
static const struct BcLexKeyword bc_lex_kws[20] = {
- BC_LEX_KW_ENTRY("auto" , 1), // 0
- BC_LEX_KW_ENTRY("break" , 1), // 1
- BC_LEX_KW_ENTRY("continue", 0), // 2 note: this one has no terminating NUL
- BC_LEX_KW_ENTRY("define" , 1), // 3
- BC_LEX_KW_ENTRY("else" , 0), // 4
- BC_LEX_KW_ENTRY("for" , 1), // 5
- BC_LEX_KW_ENTRY("halt" , 0), // 6
- BC_LEX_KW_ENTRY("ibase" , 1), // 7
- BC_LEX_KW_ENTRY("obase" , 1), // 8
- BC_LEX_KW_ENTRY("if" , 1), // 9
- BC_LEX_KW_ENTRY("last" , 0), // 10
- BC_LEX_KW_ENTRY("length" , 1), // 11
- BC_LEX_KW_ENTRY("limits" , 0), // 12
- BC_LEX_KW_ENTRY("print" , 0), // 13
- BC_LEX_KW_ENTRY("quit" , 1), // 14
- BC_LEX_KW_ENTRY("read" , 0), // 15
- BC_LEX_KW_ENTRY("return" , 1), // 16
- BC_LEX_KW_ENTRY("scale" , 1), // 17
- BC_LEX_KW_ENTRY("sqrt" , 1), // 18
- BC_LEX_KW_ENTRY("while" , 1), // 19
+ LEX_KW_ENTRY("auto" , 1), // 0
+ LEX_KW_ENTRY("break" , 1), // 1
+ LEX_KW_ENTRY("continue", 0), // 2 note: this one has no terminating NUL
+ LEX_KW_ENTRY("define" , 1), // 3
+ LEX_KW_ENTRY("else" , 0), // 4
+ LEX_KW_ENTRY("for" , 1), // 5
+ LEX_KW_ENTRY("halt" , 0), // 6
+ LEX_KW_ENTRY("ibase" , 1), // 7
+ LEX_KW_ENTRY("obase" , 1), // 8
+ LEX_KW_ENTRY("if" , 1), // 9
+ LEX_KW_ENTRY("last" , 0), // 10
+ LEX_KW_ENTRY("length" , 1), // 11
+ LEX_KW_ENTRY("limits" , 0), // 12
+ LEX_KW_ENTRY("print" , 0), // 13
+ LEX_KW_ENTRY("quit" , 1), // 14
+ LEX_KW_ENTRY("read" , 0), // 15
+ LEX_KW_ENTRY("return" , 1), // 16
+ LEX_KW_ENTRY("scale" , 1), // 17
+ LEX_KW_ENTRY("sqrt" , 1), // 18
+ LEX_KW_ENTRY("while" , 1), // 19
};
-#undef BC_LEX_KW_ENTRY
+#undef LEX_KW_ENTRY
#define STRING_else (bc_lex_kws[4].name8)
#define STRING_for (bc_lex_kws[5].name8)
#define STRING_if (bc_lex_kws[9].name8)
@@ -570,7 +570,7 @@ enum {
| (1 << 18) // 18
| (1 << 19) // 19
};
-#define bc_lex_kws_POSIX(i) ((1 << (i)) & POSIX_KWORD_MASK)
+#define keyword_is_POSIX(i) ((1 << (i)) & POSIX_KWORD_MASK)
// This is a bit array that corresponds to token types. An entry is
// true if the token is valid in an expression, false otherwise.
@@ -711,15 +711,15 @@ dc_LEX_to_INST[] = { // starts at XC_LEX_OP_POWER // corresponding XC/DC_L
#endif // ENABLE_DC
typedef struct BcLex {
- const char *buf;
+ const char *lex_inbuf;
const char *lex_next_at; // last lex_next() was called at this string
- size_t i;
- size_t line;
- size_t len;
- bool newline;
+ size_t lex_i;
+ size_t lex_line;
+ size_t lex_len;
+ bool lex_newline;
smallint lex; // was BcLexType
smallint lex_last; // was BcLexType
- BcVec lex_buf;
+ BcVec lex_strnumbuf;
} BcLex;
#define BC_PARSE_STREND (0xff)
@@ -2755,10 +2755,10 @@ static void bc_lex_lineComment(void)
// Try: echo -n '#foo' | bc
size_t i;
l->lex = XC_LEX_WHITESPACE;
- i = l->i;
- while (i < l->len && l->buf[i] != '\n')
+ i = l->lex_i;
+ while (i < l->lex_len && l->lex_inbuf[i] != '\n')
i++;
- l->i = i;
+ l->lex_i = i;
}
static void bc_lex_whitespace(void)
@@ -2766,19 +2766,19 @@ static void bc_lex_whitespace(void)
BcLex *l = &G.prs.l;
l->lex = XC_LEX_WHITESPACE;
for (;;) {
- char c = l->buf[l->i];
+ char c = l->lex_inbuf[l->lex_i];
if (c == '\n') // this is XC_LEX_NLINE, not XC_LEX_WHITESPACE
break;
if (!isspace(c))
break;
- l->i++;
+ l->lex_i++;
}
}
static BC_STATUS zbc_lex_number(char start)
{
BcLex *l = &G.prs.l;
- const char *buf = l->buf + l->i;
+ const char *buf = l->lex_inbuf + l->lex_i;
size_t len, i, ccnt;
bool pt;
@@ -2808,7 +2808,7 @@ static BC_STATUS zbc_lex_number(char start)
//ccnt is the number of chars in the number string, excluding possible
//trailing "[\<newline>].[\<newline>]" (with any number of \<NL> repetitions).
//i is buf[i] index of the first not-yet-parsed char after that.
- l->i += i;
+ l->lex_i += i;
// This might overestimate the size, if there are "\<NL>"'s
// in the number. Subtracting number_of_backslashes*2 correctly
@@ -2823,9 +2823,9 @@ static BC_STATUS zbc_lex_number(char start)
RETURN_STATUS(bc_error("number too long: must be [1,"BC_MAX_NUM_STR"]"));
}
- bc_vec_pop_all(&l->lex_buf);
- bc_vec_expand(&l->lex_buf, 1 + len);
- bc_vec_push(&l->lex_buf, &start);
+ bc_vec_pop_all(&l->lex_strnumbuf);
+ bc_vec_expand(&l->lex_strnumbuf, 1 + len);
+ bc_vec_push(&l->lex_strnumbuf, &start);
while (ccnt != 0) {
// If we have hit a backslash, skip it. We don't have
@@ -2835,12 +2835,12 @@ static BC_STATUS zbc_lex_number(char start)
ccnt -= 2;
continue;
}
- bc_vec_push(&l->lex_buf, buf);
+ bc_vec_push(&l->lex_strnumbuf, buf);
buf++;
ccnt--;
}
- bc_vec_pushZeroByte(&l->lex_buf);
+ bc_vec_pushZeroByte(&l->lex_strnumbuf);
RETURN_STATUS(BC_STATUS_SUCCESS);
}
@@ -2855,7 +2855,7 @@ static void bc_lex_name(void)
l->lex = XC_LEX_NAME;
i = 0;
- buf = l->buf + l->i - 1;
+ buf = l->lex_inbuf + l->lex_i - 1;
for (;;) {
char c = buf[i];
if ((c < 'a' || c > 'z') && !isdigit(c) && c != '_') break;
@@ -2869,28 +2869,28 @@ static void bc_lex_name(void)
return bc_error("name too long: must be [1,"BC_MAX_STRING_STR"]");
}
#endif
- bc_vec_string(&l->lex_buf, i, buf);
+ bc_vec_string(&l->lex_strnumbuf, i, buf);
// Increment the index. We minus 1 because it has already been incremented.
- l->i += i - 1;
+ l->lex_i += i - 1;
//return BC_STATUS_SUCCESS;
}
static void bc_lex_init(void)
{
- bc_char_vec_init(&G.prs.l.lex_buf);
+ bc_char_vec_init(&G.prs.l.lex_strnumbuf);
}
static void bc_lex_free(void)
{
- bc_vec_free(&G.prs.l.lex_buf);
+ bc_vec_free(&G.prs.l.lex_strnumbuf);
}
static void bc_lex_file(void)
{
- G.err_line = G.prs.l.line = 1;
- G.prs.l.newline = false;
+ G.err_line = G.prs.l.lex_line = 1;
+ G.prs.l.lex_newline = false;
}
static bool bc_lex_more_input(void)
@@ -2958,12 +2958,12 @@ static bool bc_lex_more_input(void)
break;
}
- l->buf = G.input_buffer.v;
- l->i = 0;
+ l->lex_inbuf = G.input_buffer.v;
+ l->lex_i = 0;
// bb_error_msg("G.input_buffer.len:%d '%s'", G.input_buffer.len, G.input_buffer.v);
- l->len = G.input_buffer.len - 1; // do not include NUL
+ l->lex_len = G.input_buffer.len - 1; // do not include NUL
- return l->len != 0;
+ return l->lex_len != 0;
}
IF_BC(static BC_STATUS zbc_lex_token(void);)
@@ -2979,16 +2979,16 @@ static BC_STATUS zbc_lex_next(void)
l->lex_last = l->lex;
if (l->lex_last == XC_LEX_EOF) RETURN_STATUS(bc_error("end of file"));
- l->line += l->newline;
- G.err_line = l->line;
- l->newline = false;
+ l->lex_line += l->lex_newline;
+ G.err_line = l->lex_line;
+ l->lex_newline = false;
// Loop until failure or we don't have whitespace. This
// is so the parser doesn't get inundated with whitespace.
// Comments are also XC_LEX_WHITESPACE tokens and eaten here.
s = BC_STATUS_SUCCESS;
do {
- if (l->i == l->len) {
+ if (l->lex_i == l->lex_len) {
l->lex = XC_LEX_EOF;
if (!G.prs.input_fp)
RETURN_STATUS(BC_STATUS_SUCCESS);
@@ -2996,9 +2996,9 @@ static BC_STATUS zbc_lex_next(void)
G.prs.input_fp = NULL;
RETURN_STATUS(BC_STATUS_SUCCESS);
}
- // here it's guaranteed that l->i is below l->len
+ // here it's guaranteed that l->lex_i is below l->lex_len
}
- l->lex_next_at = l->buf + l->i;
+ l->lex_next_at = l->lex_inbuf + l->lex_i;
dbg_lex("next string to parse:'%.*s'",
(int)(strchrnul(l->lex_next_at, '\n') - l->lex_next_at),
l->lex_next_at
@@ -3038,9 +3038,9 @@ static BC_STATUS zbc_lex_next_and_skip_NLINE(void)
static BC_STATUS zbc_lex_text_init(const char *text)
{
- G.prs.l.buf = text;
- G.prs.l.i = 0;
- G.prs.l.len = strlen(text);
+ G.prs.l.lex_inbuf = text;
+ G.prs.l.lex_i = 0;
+ G.prs.l.lex_len = strlen(text);
G.prs.l.lex = G.prs.l.lex_last = XC_LEX_INVALID;
RETURN_STATUS(zbc_lex_next());
}
@@ -3052,7 +3052,7 @@ static BC_STATUS zbc_lex_identifier(void)
BcLex *l = &G.prs.l;
BcStatus s;
unsigned i;
- const char *buf = l->buf + l->i - 1;
+ const char *buf = l->lex_inbuf + l->lex_i - 1;
for (i = 0; i < ARRAY_SIZE(bc_lex_kws); ++i) {
const char *keyword8 = bc_lex_kws[i].name8;
@@ -3068,20 +3068,20 @@ static BC_STATUS zbc_lex_identifier(void)
if (isalnum(buf[j]) || buf[j]=='_')
continue; // "ifz" does not match "if" keyword, "if." does
l->lex = BC_LEX_KEY_1st_keyword + i;
- if (!bc_lex_kws_POSIX(i)) {
+ if (!keyword_is_POSIX(i)) {
s = zbc_posix_error_fmt("%sthe '%.8s' keyword", "POSIX does not allow ", bc_lex_kws[i].name8);
if (s) RETURN_STATUS(s);
}
// We minus 1 because the index has already been incremented.
- l->i += j - 1;
+ l->lex_i += j - 1;
RETURN_STATUS(BC_STATUS_SUCCESS);
}
bc_lex_name();
s = BC_STATUS_SUCCESS;
- if (l->lex_buf.len > 2) {
+ if (l->lex_strnumbuf.len > 2) {
// Prevent this:
// >>> qwe=1
// bc: POSIX only allows one character names; this is bad: 'qwe=1
@@ -3102,11 +3102,11 @@ static BC_STATUS zbc_lex_string(void)
l->lex = XC_LEX_STR;
nls = 0;
- i = l->i;
+ i = l->lex_i;
for (;;) {
- char c = l->buf[i];
+ char c = l->lex_inbuf[i];
if (c == '\0') {
- l->i = i;
+ l->lex_i = i;
RETURN_STATUS(bc_error("unterminated string"));
}
if (c == '"')
@@ -3115,17 +3115,17 @@ static BC_STATUS zbc_lex_string(void)
i++;
}
- len = i - l->i;
+ len = i - l->lex_i;
// This check makes sense only if size_t is (much) larger than BC_MAX_STRING.
if (SIZE_MAX > (BC_MAX_STRING | 0xff)) {
if (len > BC_MAX_STRING)
RETURN_STATUS(bc_error("string too long: must be [1,"BC_MAX_STRING_STR"]"));
}
- bc_vec_string(&l->lex_buf, len, l->buf + l->i);
+ bc_vec_string(&l->lex_strnumbuf, len, l->lex_inbuf + l->lex_i);
- l->i = i + 1;
- l->line += nls;
- G.err_line = l->line;
+ l->lex_i = i + 1;
+ l->lex_line += nls;
+ G.err_line = l->lex_line;
RETURN_STATUS(BC_STATUS_SUCCESS);
}
@@ -3134,8 +3134,8 @@ static BC_STATUS zbc_lex_string(void)
static void bc_lex_assign(unsigned with_and_without)
{
BcLex *l = &G.prs.l;
- if (l->buf[l->i] == '=') {
- ++l->i;
+ if (l->lex_inbuf[l->lex_i] == '=') {
+ l->lex_i++;
with_and_without >>= 8; // store "with" value
} // else store "without" value
l->lex = (with_and_without & 0xff);
@@ -3147,10 +3147,10 @@ static BC_STATUS zbc_lex_comment(void)
{
BcLex *l = &G.prs.l;
size_t i, nls = 0;
- const char *buf = l->buf;
+ const char *buf = l->lex_inbuf;
l->lex = XC_LEX_WHITESPACE;
- i = l->i; /* here buf[l->i] is the '*' of opening comment delimiter */
+ i = l->lex_i; /* here buf[l->lex_i] is the '*' of opening comment delimiter */
for (;;) {
char c = buf[++i];
check_star:
@@ -3161,15 +3161,15 @@ static BC_STATUS zbc_lex_comment(void)
goto check_star;
}
if (c == '\0') {
- l->i = i;
+ l->lex_i = i;
RETURN_STATUS(bc_error("unterminated comment"));
}
nls += (c == '\n');
}
- l->i = i + 1;
- l->line += nls;
- G.err_line = l->line;
+ l->lex_i = i + 1;
+ l->lex_line += nls;
+ G.err_line = l->lex_line;
RETURN_STATUS(BC_STATUS_SUCCESS);
}
@@ -3180,18 +3180,18 @@ static BC_STATUS zbc_lex_token(void)
{
BcLex *l = &G.prs.l;
BcStatus s = BC_STATUS_SUCCESS;
- char c = l->buf[l->i++], c2;
+ char c = l->lex_inbuf[l->lex_i++], c2;
// This is the workhorse of the lexer.
switch (c) {
// case '\0': // probably never reached
-// l->i--;
+// l->lex_i--;
// l->lex = XC_LEX_EOF;
-// l->newline = true;
+// l->lex_newline = true;
// break;
case '\n':
l->lex = XC_LEX_NLINE;
- l->newline = true;
+ l->lex_newline = true;
break;
case '\t':
case '\v':
@@ -3219,11 +3219,11 @@ static BC_STATUS zbc_lex_token(void)
bc_lex_assign(BC_LEX_OP_ASSIGN_MODULUS, XC_LEX_OP_MODULUS);
break;
case '&':
- c2 = l->buf[l->i];
+ c2 = l->lex_inbuf[l->lex_i];
if (c2 == '&') {
s = zbc_POSIX_does_not_allow_bool_ops_this_is_bad("&&");
if (s) RETURN_STATUS(s);
- ++l->i;
+ l->lex_i++;
l->lex = BC_LEX_OP_BOOL_AND;
} else {
l->lex = XC_LEX_INVALID;
@@ -3238,9 +3238,9 @@ static BC_STATUS zbc_lex_token(void)
bc_lex_assign(BC_LEX_OP_ASSIGN_MULTIPLY, XC_LEX_OP_MULTIPLY);
break;
case '+':
- c2 = l->buf[l->i];
+ c2 = l->lex_inbuf[l->lex_i];
if (c2 == '+') {
- ++l->i;
+ l->lex_i++;
l->lex = BC_LEX_OP_INC;
} else
bc_lex_assign(BC_LEX_OP_ASSIGN_PLUS, XC_LEX_OP_PLUS);
@@ -3249,15 +3249,15 @@ static BC_STATUS zbc_lex_token(void)
l->lex = BC_LEX_COMMA;
break;
case '-':
- c2 = l->buf[l->i];
+ c2 = l->lex_inbuf[l->lex_i];
if (c2 == '-') {
- ++l->i;
+ l->lex_i++;
l->lex = BC_LEX_OP_DEC;
} else
bc_lex_assign(BC_LEX_OP_ASSIGN_MINUS, XC_LEX_OP_MINUS);
break;
case '.':
- if (isdigit(l->buf[l->i]))
+ if (isdigit(l->lex_inbuf[l->lex_i]))
s = zbc_lex_number(c);
else {
l->lex = BC_LEX_KEY_LAST;
@@ -3265,7 +3265,7 @@ static BC_STATUS zbc_lex_token(void)
}
break;
case '/':
- c2 = l->buf[l->i];
+ c2 = l->lex_inbuf[l->lex_i];
if (c2 == '*')
s = zbc_lex_comment();
else
@@ -3306,9 +3306,9 @@ static BC_STATUS zbc_lex_token(void)
l->lex = (BcLexType)(c - '[' + BC_LEX_LBRACKET);
break;
case '\\':
- if (l->buf[l->i] == '\n') {
+ if (l->lex_inbuf[l->lex_i] == '\n') {
l->lex = XC_LEX_WHITESPACE;
- ++l->i;
+ l->lex_i++;
} else
s = bc_error_bad_character(c);
break;
@@ -3348,11 +3348,11 @@ static BC_STATUS zbc_lex_token(void)
l->lex = (BcLexType)(c - '{' + BC_LEX_LBRACE);
break;
case '|':
- c2 = l->buf[l->i];
+ c2 = l->lex_inbuf[l->lex_i];
if (c2 == '|') {
s = zbc_POSIX_does_not_allow_bool_ops_this_is_bad("||");
if (s) RETURN_STATUS(s);
- ++l->i;
+ l->lex_i++;
l->lex = BC_LEX_OP_BOOL_OR;
} else {
l->lex = XC_LEX_INVALID;
@@ -3374,14 +3374,14 @@ static BC_STATUS zbc_lex_token(void)
static BC_STATUS zdc_lex_register(void)
{
BcLex *l = &G.prs.l;
- if (G_exreg && isspace(l->buf[l->i])) {
+ if (G_exreg && isspace(l->lex_inbuf[l->lex_i])) {
bc_lex_whitespace(); // eats whitespace (but not newline)
- l->i++; // bc_lex_name() expects this
+ l->lex_i++; // bc_lex_name() expects this
bc_lex_name();
} else {
- bc_vec_pop_all(&l->lex_buf);
- bc_vec_push(&l->lex_buf, &l->buf[l->i++]);
- bc_vec_pushZeroByte(&l->lex_buf);
+ bc_vec_pop_all(&l->lex_strnumbuf);
+ bc_vec_push(&l->lex_strnumbuf, &l->lex_inbuf[l->lex_i++]);
+ bc_vec_pushZeroByte(&l->lex_strnumbuf);
l->lex = XC_LEX_NAME;
}
@@ -3395,39 +3395,39 @@ static BC_STATUS zdc_lex_string(void)
size_t depth, nls, i;
l->lex = XC_LEX_STR;
- bc_vec_pop_all(&l->lex_buf);
+ bc_vec_pop_all(&l->lex_strnumbuf);
nls = 0;
depth = 1;
- i = l->i;
+ i = l->lex_i;
for (;;) {
- char c = l->buf[i];
+ char c = l->lex_inbuf[i];
if (c == '\0') {
- l->i = i;
+ l->lex_i = i;
RETURN_STATUS(bc_error("string end could not be found"));
}
nls += (c == '\n');
- if (i == l->i || l->buf[i - 1] != '\\') {
+ if (i == l->lex_i || l->lex_inbuf[i - 1] != '\\') {
if (c == '[') depth++;
if (c == ']')
if (--depth == 0)
break;
}
- bc_vec_push(&l->lex_buf, &l->buf[i]);
+ bc_vec_push(&l->lex_strnumbuf, &l->lex_inbuf[i]);
i++;
}
i++;
- bc_vec_pushZeroByte(&l->lex_buf);
+ bc_vec_pushZeroByte(&l->lex_strnumbuf);
// This check makes sense only if size_t is (much) larger than BC_MAX_STRING.
if (SIZE_MAX > (BC_MAX_STRING | 0xff)) {
- if (i - l->i > BC_MAX_STRING)
+ if (i - l->lex_i > BC_MAX_STRING)
RETURN_STATUS(bc_error("string too long: must be [1,"BC_MAX_STRING_STR"]"));
}
- l->i = i;
- l->line += nls;
- G.err_line = l->line;
+ l->lex_i = i;
+ l->lex_line += nls;
+ G.err_line = l->lex_line;
RETURN_STATUS(BC_STATUS_SUCCESS);
}
@@ -3456,7 +3456,7 @@ static BC_STATUS zdc_lex_token(void)
}
s = BC_STATUS_SUCCESS;
- c = l->buf[l->i++];
+ c = l->lex_inbuf[l->lex_i++];
if (c >= '%' && c <= '~'
&& (l->lex = dc_char_to_LEX[c - '%']) != XC_LEX_INVALID
) {
@@ -3478,18 +3478,18 @@ static BC_STATUS zdc_lex_token(void)
// IOW: typing "1p<enter>" should print "1" _at once_,
// not after some more input.
l->lex = XC_LEX_NLINE;
- l->newline = true;
+ l->lex_newline = true;
break;
case '\t':
case '\v':
case '\f':
case '\r':
case ' ':
- l->newline = 0; // was (c == '\n')
+ l->lex_newline = 0; // was (c == '\n')
bc_lex_whitespace();
break;
case '!':
- c2 = l->buf[l->i];
+ c2 = l->lex_inbuf[l->lex_i];
if (c2 == '=')
l->lex = XC_LEX_OP_REL_NE;
else if (c2 == '<')
@@ -3498,13 +3498,13 @@ static BC_STATUS zdc_lex_token(void)
l->lex = XC_LEX_OP_REL_GE;
else
RETURN_STATUS(bc_error_bad_character(c));
- ++l->i;
+ l->lex_i++;
break;
case '#':
bc_lex_lineComment();
break;
case '.':
- if (isdigit(l->buf[l->i]))
+ if (isdigit(l->lex_inbuf[l->lex_i]))
s = zbc_lex_number(c);
else
s = bc_error_bad_character(c);
@@ -3593,7 +3593,7 @@ static void bc_parse_pushJUMP_ZERO(size_t idx)
static BC_STATUS zbc_parse_pushSTR(void)
{
BcParse *p = &G.prs;
- char *str = xstrdup(p->l.lex_buf.v);
+ char *str = xstrdup(p->l.lex_strnumbuf.v);
bc_parse_push(XC_INST_STR);
bc_parse_pushIndex(p->func->strs.len);
@@ -3607,7 +3607,7 @@ static BC_STATUS zbc_parse_pushSTR(void)
static void bc_parse_pushNUM(void)
{
BcParse *p = &G.prs;
- char *num = xstrdup(p->l.lex_buf.v);
+ char *num = xstrdup(p->l.lex_strnumbuf.v);
#if ENABLE_BC && ENABLE_DC
size_t idx = bc_vec_push(IS_BC ? &p->func->consts : &G.prog.consts, &num);
#elif ENABLE_BC
@@ -3656,7 +3656,7 @@ static void bc_parse_reset(void)
p->func = bc_program_func_BC_PROG_MAIN();
}
- p->l.i = p->l.len;
+ p->l.lex_i = p->l.lex_len;
p->l.lex = XC_LEX_EOF;
IF_BC(bc_vec_pop_all(&p->exits);)
@@ -3897,7 +3897,7 @@ static BC_STATUS zbc_parse_name(BcInst *type, uint8_t flags)
BcStatus s;
char *name;
- name = xstrdup(p->l.lex_buf.v);
+ name = xstrdup(p->l.lex_strnumbuf.v);
s = zbc_lex_next();
if (s) goto err;
@@ -4317,7 +4317,7 @@ static BC_STATUS zbc_parse_for(void)
// Set this for the next call to bc_parse_pushNUM().
// This is safe to set because the current token is a semicolon,
// which has no string requirement.
- bc_vec_string(&p->l.lex_buf, 1, "1");
+ bc_vec_string(&p->l.lex_strnumbuf, 1, "1");
bc_parse_pushNUM();
s = zbc_POSIX_does_not_allow_empty_X_expression_in_for("condition");
}
@@ -4418,7 +4418,7 @@ static BC_STATUS zbc_parse_funcdef(void)
if (p->l.lex != XC_LEX_NAME)
RETURN_STATUS(bc_error("bad function definition"));
- name = xstrdup(p->l.lex_buf.v);
+ name = xstrdup(p->l.lex_strnumbuf.v);
p->fidx = bc_program_addFunc(name);
p->func = bc_program_func(p->fidx);
@@ -4435,7 +4435,7 @@ static BC_STATUS zbc_parse_funcdef(void)
++p->func->nparams;
- name = xstrdup(p->l.lex_buf.v);
+ name = xstrdup(p->l.lex_strnumbuf.v);
s = zbc_lex_next();
if (s) goto err;
@@ -4517,7 +4517,7 @@ static BC_STATUS zbc_parse_auto(void)
if (p->l.lex != XC_LEX_NAME)
RETURN_STATUS(bc_error("bad 'auto' syntax"));
- name = xstrdup(p->l.lex_buf.v);
+ name = xstrdup(p->l.lex_strnumbuf.v);
s = zbc_lex_next();
if (s) goto err;
@@ -4925,7 +4925,7 @@ static BC_STATUS zdc_parse_register(void)
if (s) RETURN_STATUS(s);
if (p->l.lex != XC_LEX_NAME) RETURN_STATUS(bc_error_bad_token());
- bc_parse_pushName(p->l.lex_buf.v);
+ bc_parse_pushName(p->l.lex_strnumbuf.v);
RETURN_STATUS(s);
}
@@ -4939,7 +4939,7 @@ static void dc_parse_string(void)
dbg_lex_enter("%s:%d entered", __func__, __LINE__);
- str = xstrdup(p->l.lex_buf.v);
+ str = xstrdup(p->l.lex_strnumbuf.v);
bc_parse_push(XC_INST_STR);
bc_parse_pushIndex(len);
bc_vec_push(&G.prog.strs, &str);