summaryrefslogtreecommitdiffhomepage
path: root/miscutils
diff options
context:
space:
mode:
Diffstat (limited to 'miscutils')
-rw-r--r--miscutils/bc.c443
1 files changed, 210 insertions, 233 deletions
diff --git a/miscutils/bc.c b/miscutils/bc.c
index f4e499f13..5789869a7 100644
--- a/miscutils/bc.c
+++ b/miscutils/bc.c
@@ -713,7 +713,6 @@ dc_LEX_to_INST[] = { // starts at XC_LEX_OP_POWER // corresponding XC/DC_L
typedef struct BcParse {
smallint lex; // was BcLexType // first member is most used
smallint lex_last; // was BcLexType
- bool lex_newline;
size_t lex_line;
const char *lex_inbuf;
const char *lex_end;
@@ -2741,147 +2740,6 @@ static BC_STATUS zbc_num_parse(BcNum *n, const char *val, unsigned base_t)
}
#define zbc_num_parse(...) (zbc_num_parse(__VA_ARGS__) COMMA_SUCCESS)
-static void bc_lex_lineComment(void)
-{
- BcParse *p = &G.prs;
- // Try: echo -n '#foo' | bc
- p->lex = XC_LEX_WHITESPACE;
- while (p->lex_inbuf < p->lex_end && *p->lex_inbuf != '\n')
- p->lex_inbuf++;
-}
-
-static void bc_lex_whitespace(void)
-{
- BcParse *p = &G.prs;
- p->lex = XC_LEX_WHITESPACE;
- for (;;) {
- char c = *p->lex_inbuf;
- if (c == '\n') // this is XC_LEX_NLINE, not XC_LEX_WHITESPACE
- break;
- if (!isspace(c))
- break;
- p->lex_inbuf++;
- }
-}
-
-static BC_STATUS zbc_lex_number(char start)
-{
- BcParse *p = &G.prs;
- const char *buf = p->lex_inbuf;
- size_t len, i, ccnt;
- bool pt;
-
- pt = (start == '.');
- p->lex = XC_LEX_NUMBER;
- ccnt = i = 0;
- for (;;) {
- char c = buf[i];
- if (c == '\0')
- break;
- if (c == '\\' && buf[i + 1] == '\n') {
- i += 2;
- //number_of_backslashes++ - see comment below
- continue;
- }
- if (!isdigit(c) && (c < 'A' || c > 'F')) {
- if (c != '.') break;
- // if '.' was already seen, stop on second one:
- if (pt) break;
- pt = true;
- }
- // buf[i] is one of "0-9A-F."
- i++;
- if (c != '.')
- ccnt = i;
- }
- //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.
- p->lex_inbuf += i;
-
- // This might overestimate the size, if there are "\<NL>"'s
- // in the number. Subtracting number_of_backslashes*2 correctly
- // is not that easy: consider that in the case of "NNN.\<NL>"
- // loop above will count "\<NL>" before it realizes it is not
- // in fact *inside* the number:
- len = ccnt + 1; // +1 byte for NUL termination
-
- // This check makes sense only if size_t is (much) larger than BC_MAX_NUM.
- if (SIZE_MAX > (BC_MAX_NUM | 0xff)) {
- if (len > BC_MAX_NUM)
- RETURN_STATUS(bc_error("number too long: must be [1,"BC_MAX_NUM_STR"]"));
- }
-
- bc_vec_pop_all(&p->lex_strnumbuf);
- bc_vec_expand(&p->lex_strnumbuf, 1 + len);
- bc_vec_push(&p->lex_strnumbuf, &start);
-
- while (ccnt != 0) {
- // If we have hit a backslash, skip it. We don't have
- // to check for a newline because it's guaranteed.
- if (*buf == '\\') {
- buf += 2;
- ccnt -= 2;
- continue;
- }
- bc_vec_push(&p->lex_strnumbuf, buf);
- buf++;
- ccnt--;
- }
-
- bc_vec_pushZeroByte(&p->lex_strnumbuf);
-
- RETURN_STATUS(BC_STATUS_SUCCESS);
-}
-#define zbc_lex_number(...) (zbc_lex_number(__VA_ARGS__) COMMA_SUCCESS)
-
-static void bc_lex_name(void)
-{
- BcParse *p = &G.prs;
- size_t i;
- const char *buf;
-
- p->lex = XC_LEX_NAME;
-
- i = 0;
- buf = p->lex_inbuf - 1;
- for (;;) {
- char c = buf[i];
- if ((c < 'a' || c > 'z') && !isdigit(c) && c != '_') break;
- i++;
- }
-
-#if 0 // We do not protect against people with gigabyte-long names
- // This check makes sense only if size_t is (much) larger than BC_MAX_STRING.
- if (SIZE_MAX > (BC_MAX_STRING | 0xff)) {
- if (i > BC_MAX_STRING)
- return bc_error("name too long: must be [1,"BC_MAX_STRING_STR"]");
- }
-#endif
- bc_vec_string(&p->lex_strnumbuf, i, buf);
-
- // Increment the index. We minus 1 because it has already been incremented.
- p->lex_inbuf += i - 1;
-
- //return BC_STATUS_SUCCESS;
-}
-
-static void bc_lex_init(void)
-{
- bc_char_vec_init(&G.prs.lex_strnumbuf);
-}
-
-static void bc_lex_free(void)
-{
- bc_vec_free(&G.prs.lex_strnumbuf);
-}
-
-static void bc_lex_file(void)
-{
- G.err_line = G.prs.lex_line = 1;
- G.prs.lex_newline = false;
-}
-
static bool bc_lex_more_input(void)
{
BcParse *p = &G.prs;
@@ -2996,6 +2854,147 @@ static bool bc_lex_more_input(void)
return G.input_buffer.len > 1;
}
+// p->lex_inbuf points to the current string to be parsed.
+// if p->lex_inbuf points to '\0', it's either EOF or it points after
+// last processed line's terminating '\n' (and more reading needs to be done
+// to get next character).
+//
+// If you are in a situation where that is a possibility, call peek_inbuf().
+// If necessary, it performs more reading and changes p->lex_inbuf,
+// then it returns *p->lex_inbuf (which will be '\0' only if it's EOF).
+// After it, just referencing *p->lex_inbuf is valid, and if it wasn't '\0',
+// it's ok to do p->lex_inbuf++ once without end-of-buffer checking.
+//
+// eat_inbuf() is equvalent to "peek_inbuf(); if (c) p->lex_inbuf++":
+// it returns current char and advances the pointer (if not EOF).
+// After eat_inbuf(), referencing p->lex_inbuf[-1] and *p->lex_inbuf is valid.
+//
+// In many cases, you can use fast *p->lex_inbuf instead of peek_inbuf():
+// unless prev char might have been '\n', *p->lex_inbuf is '\0' ONLY
+// on real EOF, not end-of-buffer.
+static char peek_inbuf(void)
+{
+ if (G.prs.lex_inbuf == G.prs.lex_end) {
+ if (G.prs.lex_input_fp)
+ if (!bc_lex_more_input())
+ G.prs.lex_input_fp = NULL;
+ }
+ return *G.prs.lex_inbuf;
+}
+static char eat_inbuf(void)
+{
+ char c = peek_inbuf();
+ if (c) G.prs.lex_inbuf++;
+ return c;
+}
+
+static void bc_lex_lineComment(void)
+{
+ BcParse *p = &G.prs;
+ char c;
+
+ // Try: echo -n '#foo' | bc
+ p->lex = XC_LEX_WHITESPACE;
+
+ // We depend here on input being done in whole lines:
+ // '\0' which isn't the EOF can only be seen after '\n'.
+ while ((c = *p->lex_inbuf) != '\n' && c != '\0')
+ p->lex_inbuf++;
+}
+
+static void bc_lex_whitespace(void)
+{
+ BcParse *p = &G.prs;
+
+ p->lex = XC_LEX_WHITESPACE;
+ for (;;) {
+ // We depend here on input being done in whole lines:
+ // '\0' which isn't the EOF can only be seen after '\n'.
+ char c = *p->lex_inbuf;
+ if (c == '\n') // this is XC_LEX_NLINE, not XC_LEX_WHITESPACE
+ break;
+ if (!isspace(c))
+ break;
+ p->lex_inbuf++;
+ }
+}
+
+static BC_STATUS zbc_lex_number(char last)
+{
+ BcParse *p = &G.prs;
+ bool pt;
+
+ bc_vec_pop_all(&p->lex_strnumbuf);
+ bc_vec_pushByte(&p->lex_strnumbuf, last);
+
+ pt = (last == '.');
+ p->lex = XC_LEX_NUMBER;
+ for (;;) {
+ // We depend here on input being done in whole lines:
+ // '\0' which isn't the EOF can only be seen after '\n'.
+ char c = *p->lex_inbuf;
+ check_c:
+ if (c == '\0')
+ break;
+ if (c == '\\' && p->lex_inbuf[1] == '\n') {
+ p->lex_inbuf += 2;
+ p->lex_line++;
+ c = peek_inbuf(); // force next line to be read
+ goto check_c;
+ }
+ if (!isdigit(c) && (c < 'A' || c > 'F')) {
+ if (c != '.') break;
+ // if '.' was already seen, stop on second one:
+ if (pt) break;
+ pt = true;
+ }
+ // c is one of "0-9A-F."
+ last = c;
+ bc_vec_push(&p->lex_strnumbuf, p->lex_inbuf);
+ p->lex_inbuf++;
+ }
+ if (last == '.') // remove trailing '.' if any
+ bc_vec_pop(&p->lex_strnumbuf);
+ bc_vec_pushZeroByte(&p->lex_strnumbuf);
+
+ G.err_line = G.prs.lex_line;
+ RETURN_STATUS(BC_STATUS_SUCCESS);
+}
+#define zbc_lex_number(...) (zbc_lex_number(__VA_ARGS__) COMMA_SUCCESS)
+
+static void bc_lex_name(void)
+{
+ BcParse *p = &G.prs;
+ size_t i;
+ const char *buf;
+
+ p->lex = XC_LEX_NAME;
+
+ // Since names can't cross lines with \<newline>,
+ // we depend on the fact that whole line is in the buffer
+ i = 0;
+ buf = p->lex_inbuf - 1;
+ for (;;) {
+ char c = buf[i];
+ if ((c < 'a' || c > 'z') && !isdigit(c) && c != '_') break;
+ i++;
+ }
+
+#if 0 // We do not protect against people with gigabyte-long names
+ // This check makes sense only if size_t is (much) larger than BC_MAX_STRING.
+ if (SIZE_MAX > (BC_MAX_STRING | 0xff)) {
+ if (i > BC_MAX_STRING)
+ return bc_error("name too long: must be [1,"BC_MAX_STRING_STR"]");
+ }
+#endif
+ bc_vec_string(&p->lex_strnumbuf, i, buf);
+
+ // Increment the index. We minus 1 because it has already been incremented.
+ p->lex_inbuf += i - 1;
+
+ //return BC_STATUS_SUCCESS;
+}
+
IF_BC(static BC_STATUS zbc_lex_token(void);)
IF_DC(static BC_STATUS zdc_lex_token(void);)
#define zbc_lex_token(...) (zbc_lex_token(__VA_ARGS__) COMMA_SUCCESS)
@@ -3007,26 +3006,18 @@ static BC_STATUS zbc_lex_next(void)
BcStatus s;
p->lex_last = p->lex;
- if (p->lex_last == XC_LEX_EOF) RETURN_STATUS(bc_error("end of file"));
-
- p->lex_line += p->lex_newline;
- G.err_line = p->lex_line;
- p->lex_newline = false;
+ if (p->lex_last == XC_LEX_EOF)
+ RETURN_STATUS(bc_error("end of file"));
// 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 (p->lex_inbuf == p->lex_end) {
+ if (*p->lex_inbuf == '\0') {
p->lex = XC_LEX_EOF;
- if (!G.prs.lex_input_fp)
+ if (peek_inbuf() == '\0')
RETURN_STATUS(BC_STATUS_SUCCESS);
- if (!bc_lex_more_input()) {
- G.prs.lex_input_fp = NULL;
- RETURN_STATUS(BC_STATUS_SUCCESS);
- }
- // here it's guaranteed that p->lex_ibuf is below p->lex_end
}
p->lex_next_at = p->lex_inbuf;
dbg_lex("next string to parse:'%.*s'",
@@ -3126,80 +3117,70 @@ static BC_STATUS zbc_lex_identifier(void)
static BC_STATUS zbc_lex_string(void)
{
BcParse *p = &G.prs;
- size_t len, nls, i;
p->lex = XC_LEX_STR;
-
- nls = 0;
- i = 0;
+ bc_vec_pop_all(&p->lex_strnumbuf);
for (;;) {
- char c = p->lex_inbuf[i];
+ char c = peek_inbuf(); // strings can cross lines
if (c == '\0') {
- p->lex_inbuf += i;
- RETURN_STATUS(bc_error("unterminated string"));
+ RETURN_STATUS(bc_error("unterminated string1"));
}
if (c == '"')
break;
- nls += (c == '\n');
- i++;
- }
-
- len = 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"]"));
+ if (c == '\n')
+ p->lex_line++;
+ bc_vec_push(&p->lex_strnumbuf, p->lex_inbuf);
+ p->lex_inbuf++;
}
- bc_vec_string(&p->lex_strnumbuf, len, p->lex_inbuf);
+ bc_vec_pushZeroByte(&p->lex_strnumbuf);
+ p->lex_inbuf++;
- p->lex_inbuf += i + 1;
- p->lex_line += nls;
G.err_line = p->lex_line;
-
RETURN_STATUS(BC_STATUS_SUCCESS);
}
#define zbc_lex_string(...) (zbc_lex_string(__VA_ARGS__) COMMA_SUCCESS)
-static void bc_lex_assign(unsigned with_and_without)
+static void parse_lex_by_checking_eq_sign(unsigned with_and_without)
{
BcParse *p = &G.prs;
if (*p->lex_inbuf == '=') {
+ // ^^^ not using peek_inbuf() since '==' etc can't be split across lines
p->lex_inbuf++;
with_and_without >>= 8; // store "with" value
} // else store "without" value
p->lex = (with_and_without & 0xff);
}
-#define bc_lex_assign(with, without) \
- bc_lex_assign(((with)<<8)|(without))
+#define parse_lex_by_checking_eq_sign(with, without) \
+ parse_lex_by_checking_eq_sign(((with)<<8)|(without))
static BC_STATUS zbc_lex_comment(void)
{
BcParse *p = &G.prs;
- size_t i, nls = 0;
- const char *buf = p->lex_inbuf;
p->lex = XC_LEX_WHITESPACE;
- i = 0; /* here lex_inbuf[0] is the '*' of opening comment delimiter */
+ // here lex_inbuf is at '*' of opening comment delimiter
for (;;) {
- char c = buf[++i];
+ char c;
+
+ p->lex_inbuf++;
+ c = peek_inbuf();
check_star:
if (c == '*') {
- c = buf[++i];
+ p->lex_inbuf++;
+ c = peek_inbuf();
if (c == '/')
break;
goto check_star;
}
if (c == '\0') {
- p->lex_inbuf += i;
RETURN_STATUS(bc_error("unterminated comment"));
}
- nls += (c == '\n');
+ if (c == '\n')
+ p->lex_line++;
}
+ p->lex_inbuf++; // skip trailing '/'
- p->lex_inbuf += i + 1;
- p->lex_line += nls;
G.err_line = p->lex_line;
-
RETURN_STATUS(BC_STATUS_SUCCESS);
}
#define zbc_lex_comment(...) (zbc_lex_comment(__VA_ARGS__) COMMA_SUCCESS)
@@ -3209,7 +3190,7 @@ static BC_STATUS zbc_lex_token(void)
{
BcParse *p = &G.prs;
BcStatus s = BC_STATUS_SUCCESS;
- char c = *p->lex_inbuf++;
+ char c = eat_inbuf();
char c2;
// This is the workhorse of the lexer.
@@ -3217,11 +3198,10 @@ static BC_STATUS zbc_lex_token(void)
// case '\0': // probably never reached
// p->lex_inbuf--;
// p->lex = XC_LEX_EOF;
-// p->lex_newline = true;
// break;
case '\n':
+ p->lex_line++;
p->lex = XC_LEX_NLINE;
- p->lex_newline = true;
break;
case '\t':
case '\v':
@@ -3231,7 +3211,7 @@ static BC_STATUS zbc_lex_token(void)
bc_lex_whitespace();
break;
case '!':
- bc_lex_assign(XC_LEX_OP_REL_NE, BC_LEX_OP_BOOL_NOT);
+ parse_lex_by_checking_eq_sign(XC_LEX_OP_REL_NE, BC_LEX_OP_BOOL_NOT);
if (p->lex == BC_LEX_OP_BOOL_NOT) {
s = zbc_POSIX_does_not_allow_bool_ops_this_is_bad("!");
if (s) RETURN_STATUS(s);
@@ -3246,7 +3226,7 @@ static BC_STATUS zbc_lex_token(void)
bc_lex_lineComment();
break;
case '%':
- bc_lex_assign(BC_LEX_OP_ASSIGN_MODULUS, XC_LEX_OP_MODULUS);
+ parse_lex_by_checking_eq_sign(BC_LEX_OP_ASSIGN_MODULUS, XC_LEX_OP_MODULUS);
break;
case '&':
c2 = *p->lex_inbuf;
@@ -3265,7 +3245,7 @@ static BC_STATUS zbc_lex_token(void)
p->lex = (BcLexType)(c - '(' + BC_LEX_LPAREN);
break;
case '*':
- bc_lex_assign(BC_LEX_OP_ASSIGN_MULTIPLY, XC_LEX_OP_MULTIPLY);
+ parse_lex_by_checking_eq_sign(BC_LEX_OP_ASSIGN_MULTIPLY, XC_LEX_OP_MULTIPLY);
break;
case '+':
c2 = *p->lex_inbuf;
@@ -3273,7 +3253,7 @@ static BC_STATUS zbc_lex_token(void)
p->lex_inbuf++;
p->lex = BC_LEX_OP_INC;
} else
- bc_lex_assign(BC_LEX_OP_ASSIGN_PLUS, XC_LEX_OP_PLUS);
+ parse_lex_by_checking_eq_sign(BC_LEX_OP_ASSIGN_PLUS, XC_LEX_OP_PLUS);
break;
case ',':
p->lex = BC_LEX_COMMA;
@@ -3284,7 +3264,7 @@ static BC_STATUS zbc_lex_token(void)
p->lex_inbuf++;
p->lex = BC_LEX_OP_DEC;
} else
- bc_lex_assign(BC_LEX_OP_ASSIGN_MINUS, XC_LEX_OP_MINUS);
+ parse_lex_by_checking_eq_sign(BC_LEX_OP_ASSIGN_MINUS, XC_LEX_OP_MINUS);
break;
case '.':
if (isdigit(*p->lex_inbuf))
@@ -3299,7 +3279,7 @@ static BC_STATUS zbc_lex_token(void)
if (c2 == '*')
s = zbc_lex_comment();
else
- bc_lex_assign(BC_LEX_OP_ASSIGN_DIVIDE, XC_LEX_OP_DIVIDE);
+ parse_lex_by_checking_eq_sign(BC_LEX_OP_ASSIGN_DIVIDE, XC_LEX_OP_DIVIDE);
break;
case '0':
case '1':
@@ -3323,13 +3303,13 @@ static BC_STATUS zbc_lex_token(void)
p->lex = BC_LEX_SCOLON;
break;
case '<':
- bc_lex_assign(XC_LEX_OP_REL_LE, XC_LEX_OP_REL_LT);
+ parse_lex_by_checking_eq_sign(XC_LEX_OP_REL_LE, XC_LEX_OP_REL_LT);
break;
case '=':
- bc_lex_assign(XC_LEX_OP_REL_EQ, BC_LEX_OP_ASSIGN);
+ parse_lex_by_checking_eq_sign(XC_LEX_OP_REL_EQ, BC_LEX_OP_ASSIGN);
break;
case '>':
- bc_lex_assign(XC_LEX_OP_REL_GE, XC_LEX_OP_REL_GT);
+ parse_lex_by_checking_eq_sign(XC_LEX_OP_REL_GE, XC_LEX_OP_REL_GT);
break;
case '[':
case ']':
@@ -3343,7 +3323,7 @@ static BC_STATUS zbc_lex_token(void)
s = bc_error_bad_character(c);
break;
case '^':
- bc_lex_assign(BC_LEX_OP_ASSIGN_POWER, XC_LEX_OP_POWER);
+ parse_lex_by_checking_eq_sign(BC_LEX_OP_ASSIGN_POWER, XC_LEX_OP_POWER);
break;
case 'a':
case 'b':
@@ -3422,43 +3402,30 @@ static BC_STATUS zdc_lex_register(void)
static BC_STATUS zdc_lex_string(void)
{
BcParse *p = &G.prs;
- size_t depth, nls, i;
+ size_t depth;
p->lex = XC_LEX_STR;
bc_vec_pop_all(&p->lex_strnumbuf);
- nls = 0;
depth = 1;
- i = 0;
for (;;) {
- char c = p->lex_inbuf[i];
+ char c = peek_inbuf();
if (c == '\0') {
- p->lex_inbuf += i;
- RETURN_STATUS(bc_error("string end could not be found"));
- }
- nls += (c == '\n');
- if (i == 0 || p->lex_inbuf[i - 1] != '\\') {
- if (c == '[') depth++;
- if (c == ']')
- if (--depth == 0)
- break;
+ RETURN_STATUS(bc_error("unterminated string"));
}
- bc_vec_push(&p->lex_strnumbuf, &p->lex_inbuf[i]);
- i++;
+ if (c == '[') depth++;
+ if (c == ']')
+ if (--depth == 0)
+ break;
+ if (c == '\n')
+ p->lex_line++;
+ bc_vec_push(&p->lex_strnumbuf, p->lex_inbuf);
+ p->lex_inbuf++;
}
- i++;
-
bc_vec_pushZeroByte(&p->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 > BC_MAX_STRING)
- RETURN_STATUS(bc_error("string too long: must be [1,"BC_MAX_STRING_STR"]"));
- }
+ p->lex_inbuf++; // skip trailing ']'
- p->lex_inbuf += i;
- p->lex_line += nls;
G.err_line = p->lex_line;
-
RETURN_STATUS(BC_STATUS_SUCCESS);
}
#define zdc_lex_string(...) (zdc_lex_string(__VA_ARGS__) COMMA_SUCCESS)
@@ -3486,7 +3453,7 @@ static BC_STATUS zdc_lex_token(void)
}
s = BC_STATUS_SUCCESS;
- c = *p->lex_inbuf++;
+ c = eat_inbuf();
if (c >= '%' && c <= '~'
&& (p->lex = dc_char_to_LEX[c - '%']) != XC_LEX_INVALID
) {
@@ -3507,15 +3474,14 @@ static BC_STATUS zdc_lex_token(void)
// commands are not executed on pressing <enter>).
// IOW: typing "1p<enter>" should print "1" _at once_,
// not after some more input.
+ p->lex_line++;
p->lex = XC_LEX_NLINE;
- p->lex_newline = true;
break;
case '\t':
case '\v':
case '\f':
case '\r':
case ' ':
- p->lex_newline = 0; // was (c == '\n')
bc_lex_whitespace();
break;
case '!':
@@ -3702,7 +3668,7 @@ static void bc_parse_free(void)
IF_BC(bc_vec_free(&p->exits);)
IF_BC(bc_vec_free(&p->conds);)
IF_BC(bc_vec_free(&p->ops);)
- bc_lex_free();
+ bc_vec_free(&G.prs.lex_strnumbuf);
}
static void bc_parse_create(size_t fidx)
@@ -3710,7 +3676,7 @@ static void bc_parse_create(size_t fidx)
BcParse *p = &G.prs;
memset(p, 0, sizeof(BcParse));
- bc_lex_init();
+ bc_char_vec_init(&G.prs.lex_strnumbuf);
IF_BC(bc_vec_init(&p->exits, sizeof(size_t), NULL);)
IF_BC(bc_vec_init(&p->conds, sizeof(size_t), NULL);)
IF_BC(bc_vec_init(&p->ops, sizeof(BcLexType), NULL);)
@@ -4753,11 +4719,13 @@ static BcStatus bc_parse_expr_empty_ok(uint8_t flags)
switch (t) {
case BC_LEX_OP_INC:
case BC_LEX_OP_DEC:
+ dbg_lex("%s:%d LEX_OP_INC/DEC", __func__, __LINE__);
s = zbc_parse_incdec(&prev, &paren_expr, &nexprs, flags);
rprn = bin_last = false;
//get_token = false; - already is
break;
case XC_LEX_OP_MINUS:
+ dbg_lex("%s:%d LEX_OP_MINUS", __func__, __LINE__);
s = zbc_parse_minus(&prev, ops_bgn, rprn, &nexprs);
rprn = false;
//get_token = false; - already is
@@ -4770,6 +4738,7 @@ static BcStatus bc_parse_expr_empty_ok(uint8_t flags)
case BC_LEX_OP_ASSIGN_PLUS:
case BC_LEX_OP_ASSIGN_MINUS:
case BC_LEX_OP_ASSIGN:
+ dbg_lex("%s:%d LEX_ASSIGNxyz", __func__, __LINE__);
if (prev != XC_INST_VAR && prev != XC_INST_ARRAY_ELEM
&& prev != XC_INST_SCALE && prev != XC_INST_IBASE
&& prev != XC_INST_OBASE && prev != BC_INST_LAST
@@ -4794,6 +4763,7 @@ static BcStatus bc_parse_expr_empty_ok(uint8_t flags)
case BC_LEX_OP_BOOL_NOT:
case BC_LEX_OP_BOOL_OR:
case BC_LEX_OP_BOOL_AND:
+ dbg_lex("%s:%d LEX_OP_xyz", __func__, __LINE__);
if (((t == BC_LEX_OP_BOOL_NOT) != bin_last)
|| (t != BC_LEX_OP_BOOL_NOT && prev == XC_INST_BOOL_NOT)
) {
@@ -4808,6 +4778,7 @@ static BcStatus bc_parse_expr_empty_ok(uint8_t flags)
bin_last = (t != BC_LEX_OP_BOOL_NOT);
break;
case BC_LEX_LPAREN:
+ dbg_lex("%s:%d LEX_LPAREN", __func__, __LINE__);
if (BC_PARSE_LEAF(prev, rprn))
return bc_error_bad_expression();
bc_vec_push(&p->ops, &t);
@@ -4817,6 +4788,7 @@ static BcStatus bc_parse_expr_empty_ok(uint8_t flags)
rprn = bin_last = false;
break;
case BC_LEX_RPAREN:
+ dbg_lex("%s:%d LEX_RPAREN", __func__, __LINE__);
if (bin_last || prev == XC_INST_BOOL_NOT)
return bc_error_bad_expression();
if (nparens == 0) {
@@ -4833,6 +4805,7 @@ static BcStatus bc_parse_expr_empty_ok(uint8_t flags)
bin_last = false;
break;
case XC_LEX_NAME:
+ dbg_lex("%s:%d LEX_NAME", __func__, __LINE__);
if (BC_PARSE_LEAF(prev, rprn))
return bc_error_bad_expression();
s = zbc_parse_name(&prev, flags & ~BC_PARSE_NOCALL);
@@ -4842,6 +4815,7 @@ static BcStatus bc_parse_expr_empty_ok(uint8_t flags)
nexprs++;
break;
case XC_LEX_NUMBER:
+ dbg_lex("%s:%d LEX_NUMBER", __func__, __LINE__);
if (BC_PARSE_LEAF(prev, rprn))
return bc_error_bad_expression();
bc_parse_pushNUM();
@@ -4854,6 +4828,7 @@ static BcStatus bc_parse_expr_empty_ok(uint8_t flags)
case BC_LEX_KEY_IBASE:
case BC_LEX_KEY_LAST:
case BC_LEX_KEY_OBASE:
+ dbg_lex("%s:%d LEX_IBASE/LAST/OBASE", __func__, __LINE__);
if (BC_PARSE_LEAF(prev, rprn))
return bc_error_bad_expression();
prev = (char) (t - BC_LEX_KEY_IBASE + XC_INST_IBASE);
@@ -4865,6 +4840,7 @@ static BcStatus bc_parse_expr_empty_ok(uint8_t flags)
break;
case BC_LEX_KEY_LENGTH:
case BC_LEX_KEY_SQRT:
+ dbg_lex("%s:%d LEX_LEN/SQRT", __func__, __LINE__);
if (BC_PARSE_LEAF(prev, rprn))
return bc_error_bad_expression();
s = zbc_parse_builtin(t, flags, &prev);
@@ -4874,6 +4850,7 @@ static BcStatus bc_parse_expr_empty_ok(uint8_t flags)
nexprs++;
break;
case BC_LEX_KEY_READ:
+ dbg_lex("%s:%d LEX_READ", __func__, __LINE__);
if (BC_PARSE_LEAF(prev, rprn))
return bc_error_bad_expression();
s = zbc_parse_read();
@@ -4884,6 +4861,7 @@ static BcStatus bc_parse_expr_empty_ok(uint8_t flags)
nexprs++;
break;
case BC_LEX_KEY_SCALE:
+ dbg_lex("%s:%d LEX_SCALE", __func__, __LINE__);
if (BC_PARSE_LEAF(prev, rprn))
return bc_error_bad_expression();
s = zbc_parse_scale(&prev, flags);
@@ -5348,7 +5326,7 @@ static BC_STATUS zbc_program_read(void)
sv_parse = G.prs; // struct copy
bc_parse_create(BC_PROG_READ);
- //bc_lex_file(&G.prs.l); - not needed, error line info is not printed for read()
+ //G.err_line = G.prs.lex_line = 1; - not needed, error line info is not printed for read()
s = zbc_parse_text_init(buf.v);
if (s) goto exec_err;
@@ -6950,7 +6928,7 @@ static BC_STATUS zbc_vm_execute_FILE(FILE *fp, const char *filename)
G.prs.lex_filename = filename;
G.prs.lex_input_fp = fp;
- bc_lex_file();
+ G.err_line = G.prs.lex_line = 1;
do {
s = zbc_vm_process("");
@@ -7232,7 +7210,6 @@ static BC_STATUS zbc_vm_exec(void)
// We know that internal library is not buggy,
// thus error checking is normally disabled.
# define DEBUG_LIB 0
- bc_lex_file();
s = zbc_vm_process(bc_lib);
if (DEBUG_LIB && s) RETURN_STATUS(s);
}