From 69560f42da3deb5c87a9ccdbcaae26a5ff0cabaf Mon Sep 17 00:00:00 2001 From: Denys Vlasenko Date: Mon, 24 Dec 2018 14:14:23 +0100 Subject: bc: rename several BC_LEX_OPs to XC_LEX_OPs. Signed-off-by: Denys Vlasenko --- miscutils/bc.c | 143 +++++++++++++++++++++++++++++---------------------------- 1 file changed, 72 insertions(+), 71 deletions(-) (limited to 'miscutils/bc.c') diff --git a/miscutils/bc.c b/miscutils/bc.c index 3ae88f1a5..f9d654afb 100644 --- a/miscutils/bc.c +++ b/miscutils/bc.c @@ -388,27 +388,27 @@ typedef struct BcInstPtr { IF_BC(size_t results_len_before_call;) } BcInstPtr; -// BC_LEX_NEG is not used in lexing; it is only for parsing. +// XC_LEX_NEG is not used in lexing; it is only for parsing. typedef enum BcLexType { XC_LEX_EOF, XC_LEX_INVALID, - BC_LEX_1st_op, - BC_LEX_NEG = BC_LEX_1st_op, // order + XC_LEX_1st_op, + XC_LEX_NEG = XC_LEX_1st_op, // order - BC_LEX_OP_POWER, // should - BC_LEX_OP_MULTIPLY, // match - BC_LEX_OP_DIVIDE, // INST - BC_LEX_OP_MODULUS, // constants - BC_LEX_OP_PLUS, // for - BC_LEX_OP_MINUS, // these + XC_LEX_OP_POWER, // should + XC_LEX_OP_MULTIPLY, // match + XC_LEX_OP_DIVIDE, // INST + XC_LEX_OP_MODULUS, // constants + XC_LEX_OP_PLUS, // for + XC_LEX_OP_MINUS, // these - BC_LEX_OP_REL_EQ, // opeartions - BC_LEX_OP_REL_LE, // | - BC_LEX_OP_REL_GE, // | - BC_LEX_OP_REL_NE, // | - BC_LEX_OP_REL_LT, // | - BC_LEX_OP_REL_GT, // | + XC_LEX_OP_REL_EQ, // opeartions + XC_LEX_OP_REL_LE, // | + XC_LEX_OP_REL_GE, // | + XC_LEX_OP_REL_NE, // | + XC_LEX_OP_REL_LT, // | + XC_LEX_OP_REL_GT, // | BC_LEX_OP_BOOL_NOT, // | BC_LEX_OP_BOOL_OR, // | @@ -593,7 +593,7 @@ static ALWAYS_INLINE long bc_parse_exprs(unsigned i) } // This is an array of data for operators that correspond to -// [BC_LEX_1st_op...BC_LEX_last_op] token types. +// [XC_LEX_1st_op...] token types. static const uint8_t bc_parse_ops[] = { #define OP(p,l) ((int)(l) * 0x10 + (p)) OP(1, false), // neg @@ -617,18 +617,18 @@ static const //BcLexType - should be this type uint8_t dc_char_to_LEX[] = { /* %&'( */ - BC_LEX_OP_MODULUS, XC_LEX_INVALID, XC_LEX_INVALID, BC_LEX_LPAREN, + XC_LEX_OP_MODULUS, XC_LEX_INVALID, XC_LEX_INVALID, BC_LEX_LPAREN, /* )*+, */ - XC_LEX_INVALID, BC_LEX_OP_MULTIPLY, BC_LEX_OP_PLUS, XC_LEX_INVALID, + XC_LEX_INVALID, XC_LEX_OP_MULTIPLY, XC_LEX_OP_PLUS, XC_LEX_INVALID, /* -./ */ - BC_LEX_OP_MINUS, XC_LEX_INVALID, BC_LEX_OP_DIVIDE, + XC_LEX_OP_MINUS, XC_LEX_INVALID, XC_LEX_OP_DIVIDE, /* 0123456789 */ XC_LEX_INVALID, XC_LEX_INVALID, XC_LEX_INVALID, XC_LEX_INVALID, XC_LEX_INVALID, XC_LEX_INVALID, XC_LEX_INVALID, XC_LEX_INVALID, XC_LEX_INVALID, XC_LEX_INVALID, /* :;<=>?@ */ - DC_LEX_COLON, BC_LEX_SCOLON, BC_LEX_OP_REL_GT, BC_LEX_OP_REL_EQ, - BC_LEX_OP_REL_LT, BC_LEX_KEY_READ, XC_LEX_INVALID, + DC_LEX_COLON, BC_LEX_SCOLON, XC_LEX_OP_REL_GT, XC_LEX_OP_REL_EQ, + XC_LEX_OP_REL_LT, BC_LEX_KEY_READ, XC_LEX_INVALID, /* ABCDEFGH */ XC_LEX_INVALID, XC_LEX_INVALID, XC_LEX_INVALID, XC_LEX_INVALID, XC_LEX_INVALID, XC_LEX_INVALID, DC_LEX_EQ_NO_REG, XC_LEX_INVALID, @@ -641,7 +641,7 @@ dc_char_to_LEX[] = { /* Z[\] */ BC_LEX_KEY_LENGTH, XC_LEX_INVALID, XC_LEX_INVALID, XC_LEX_INVALID, /* ^_` */ - BC_LEX_OP_POWER, BC_LEX_NEG, XC_LEX_INVALID, + XC_LEX_OP_POWER, XC_LEX_NEG, XC_LEX_INVALID, /* abcdefgh */ DC_LEX_ASCIIFY, XC_LEX_INVALID, DC_LEX_CLEAR_STACK, DC_LEX_DUPLICATE, DC_LEX_ELSE, DC_LEX_PRINT_STACK, XC_LEX_INVALID, XC_LEX_INVALID, @@ -660,7 +660,8 @@ static const //BcInst - should be this type. Using signed narrow type since DC_I int8_t dc_LEX_to_INST[] = { // (so many INVALIDs b/c dc parser does not generate these LEXs) // corresponding BC_LEX_xyz: DC_INST_INVALID, DC_INST_INVALID, // EOF INVALID - DC_INST_INVALID, XC_INST_POWER, XC_INST_MULTIPLY, XC_INST_DIVIDE, // NEG OP_POWER OP_MULTIPLY OP_DIVIDE + DC_INST_INVALID, // NEG + XC_INST_POWER, XC_INST_MULTIPLY, XC_INST_DIVIDE, // OP_POWER OP_MULTIPLY OP_DIVIDE XC_INST_MODULUS, XC_INST_PLUS, XC_INST_MINUS, // OP_MODULUS OP_PLUS OP_MINUS DC_INST_INVALID, DC_INST_INVALID, DC_INST_INVALID, DC_INST_INVALID, // OP_REL_EQ OP_REL_LE OP_REL_GE OP_REL_NE DC_INST_INVALID, DC_INST_INVALID, // OP_REL_LT OP_REL_GT @@ -3159,7 +3160,7 @@ static BC_STATUS zbc_lex_token(BcLex *l) bc_lex_whitespace(l); break; case '!': - bc_lex_assign(l, BC_LEX_OP_REL_NE, BC_LEX_OP_BOOL_NOT); + bc_lex_assign(l, XC_LEX_OP_REL_NE, BC_LEX_OP_BOOL_NOT); if (l->t.t == BC_LEX_OP_BOOL_NOT) { s = bc_POSIX_does_not_allow_bool_ops_this_is_bad("!"); IF_ERROR_RETURN_POSSIBLE(if (s) RETURN_STATUS(s)); @@ -3174,7 +3175,7 @@ static BC_STATUS zbc_lex_token(BcLex *l) bc_lex_lineComment(l); break; case '%': - bc_lex_assign(l, BC_LEX_OP_ASSIGN_MODULUS, BC_LEX_OP_MODULUS); + bc_lex_assign(l, BC_LEX_OP_ASSIGN_MODULUS, XC_LEX_OP_MODULUS); break; case '&': c2 = l->buf[l->i]; @@ -3193,7 +3194,7 @@ static BC_STATUS zbc_lex_token(BcLex *l) l->t.t = (BcLexType)(c - '(' + BC_LEX_LPAREN); break; case '*': - bc_lex_assign(l, BC_LEX_OP_ASSIGN_MULTIPLY, BC_LEX_OP_MULTIPLY); + bc_lex_assign(l, BC_LEX_OP_ASSIGN_MULTIPLY, XC_LEX_OP_MULTIPLY); break; case '+': c2 = l->buf[l->i]; @@ -3201,7 +3202,7 @@ static BC_STATUS zbc_lex_token(BcLex *l) ++l->i; l->t.t = BC_LEX_OP_INC; } else - bc_lex_assign(l, BC_LEX_OP_ASSIGN_PLUS, BC_LEX_OP_PLUS); + bc_lex_assign(l, BC_LEX_OP_ASSIGN_PLUS, XC_LEX_OP_PLUS); break; case ',': l->t.t = BC_LEX_COMMA; @@ -3212,7 +3213,7 @@ static BC_STATUS zbc_lex_token(BcLex *l) ++l->i; l->t.t = BC_LEX_OP_DEC; } else - bc_lex_assign(l, BC_LEX_OP_ASSIGN_MINUS, BC_LEX_OP_MINUS); + bc_lex_assign(l, BC_LEX_OP_ASSIGN_MINUS, XC_LEX_OP_MINUS); break; case '.': if (isdigit(l->buf[l->i])) @@ -3227,7 +3228,7 @@ static BC_STATUS zbc_lex_token(BcLex *l) if (c2 == '*') s = zbc_lex_comment(l); else - bc_lex_assign(l, BC_LEX_OP_ASSIGN_DIVIDE, BC_LEX_OP_DIVIDE); + bc_lex_assign(l, BC_LEX_OP_ASSIGN_DIVIDE, XC_LEX_OP_DIVIDE); break; case '0': case '1': @@ -3251,13 +3252,13 @@ static BC_STATUS zbc_lex_token(BcLex *l) l->t.t = BC_LEX_SCOLON; break; case '<': - bc_lex_assign(l, BC_LEX_OP_REL_LE, BC_LEX_OP_REL_LT); + bc_lex_assign(l, XC_LEX_OP_REL_LE, XC_LEX_OP_REL_LT); break; case '=': - bc_lex_assign(l, BC_LEX_OP_REL_EQ, BC_LEX_OP_ASSIGN); + bc_lex_assign(l, XC_LEX_OP_REL_EQ, BC_LEX_OP_ASSIGN); break; case '>': - bc_lex_assign(l, BC_LEX_OP_REL_GE, BC_LEX_OP_REL_GT); + bc_lex_assign(l, XC_LEX_OP_REL_GE, XC_LEX_OP_REL_GT); break; case '[': case ']': @@ -3271,7 +3272,7 @@ static BC_STATUS zbc_lex_token(BcLex *l) s = bc_error_bad_character(c); break; case '^': - bc_lex_assign(l, BC_LEX_OP_ASSIGN_POWER, BC_LEX_OP_POWER); + bc_lex_assign(l, BC_LEX_OP_ASSIGN_POWER, XC_LEX_OP_POWER); break; case 'a': case 'b': @@ -3395,8 +3396,8 @@ static BC_STATUS zdc_lex_token(BcLex *l) static const //BcLexType - should be this type, but narrower type saves size: uint8_t dc_lex_regs[] = { - BC_LEX_OP_REL_EQ, BC_LEX_OP_REL_LE, BC_LEX_OP_REL_GE, BC_LEX_OP_REL_NE, - BC_LEX_OP_REL_LT, BC_LEX_OP_REL_GT, BC_LEX_SCOLON, DC_LEX_COLON, + XC_LEX_OP_REL_EQ, XC_LEX_OP_REL_LE, XC_LEX_OP_REL_GE, XC_LEX_OP_REL_NE, + XC_LEX_OP_REL_LT, XC_LEX_OP_REL_GT, BC_LEX_SCOLON, DC_LEX_COLON, DC_LEX_ELSE, DC_LEX_LOAD, DC_LEX_LOAD_POP, BC_LEX_OP_ASSIGN, DC_LEX_STORE_PUSH, }; @@ -3446,11 +3447,11 @@ static BC_STATUS zdc_lex_token(BcLex *l) case '!': c2 = l->buf[l->i]; if (c2 == '=') - l->t.t = BC_LEX_OP_REL_NE; + l->t.t = XC_LEX_OP_REL_NE; else if (c2 == '<') - l->t.t = BC_LEX_OP_REL_LE; + l->t.t = XC_LEX_OP_REL_LE; else if (c2 == '>') - l->t.t = BC_LEX_OP_REL_GE; + l->t.t = XC_LEX_OP_REL_GE; else RETURN_STATUS(bc_error_bad_character(c)); ++l->i; @@ -3686,7 +3687,7 @@ static size_t bc_program_addFunc(char *name) // We can calculate the conversion between tokens and exprs by subtracting the // position of the first operator in the lex enum and adding the position of the // first in the expr enum. Note: This only works for binary operators. -#define BC_TOKEN_2_INST(t) ((char) ((t) - BC_LEX_OP_POWER + XC_INST_POWER)) +#define BC_TOKEN_2_INST(t) ((char) ((t) - XC_LEX_OP_POWER + XC_INST_POWER)) static BcStatus bc_parse_expr_empty_ok(BcParse *p, uint8_t flags); @@ -3726,19 +3727,19 @@ static BC_STATUS zbc_parse_stmt_allow_NLINE_before(BcParse *p, const char *after static void bc_parse_operator(BcParse *p, BcLexType type, size_t start, size_t *nexprs) { - char l, r = bc_parse_op_PREC(type - BC_LEX_1st_op); - bool left = bc_parse_op_LEFT(type - BC_LEX_1st_op); + char l, r = bc_parse_op_PREC(type - XC_LEX_1st_op); + bool left = bc_parse_op_LEFT(type - XC_LEX_1st_op); while (p->ops.len > start) { BcLexType t = BC_PARSE_TOP_OP(p); if (t == BC_LEX_LPAREN) break; - l = bc_parse_op_PREC(t - BC_LEX_1st_op); + l = bc_parse_op_PREC(t - XC_LEX_1st_op); if (l >= r && (l != r || !left)) break; bc_parse_push(p, BC_TOKEN_2_INST(t)); bc_vec_pop(&p->ops); - *nexprs -= (t != BC_LEX_OP_BOOL_NOT && t != BC_LEX_NEG); + *nexprs -= (t != BC_LEX_OP_BOOL_NOT && t != XC_LEX_NEG); } bc_vec_push(&p->ops, &type); @@ -3756,7 +3757,7 @@ static BC_STATUS zbc_parse_rightParen(BcParse *p, size_t ops_bgn, size_t *nexs) bc_parse_push(p, BC_TOKEN_2_INST(top)); bc_vec_pop(&p->ops); - *nexs -= (top != BC_LEX_OP_BOOL_NOT && top != BC_LEX_NEG); + *nexs -= (top != BC_LEX_OP_BOOL_NOT && top != XC_LEX_NEG); if (p->ops.len <= ops_bgn) RETURN_STATUS(bc_error_bad_expression()); @@ -4032,13 +4033,13 @@ static BC_STATUS zbc_parse_minus(BcParse *p, BcInst *prev, size_t ops_bgn, type = rparen || etype == BC_INST_INC_POST || etype == BC_INST_DEC_POST || (etype >= XC_INST_NUM && etype <= XC_INST_SQRT) ? - BC_LEX_OP_MINUS : - BC_LEX_NEG; + XC_LEX_OP_MINUS : + XC_LEX_NEG; *prev = BC_TOKEN_2_INST(type); // We can just push onto the op stack because this is the largest // precedence operator that gets pushed. Inc/dec does not. - if (type != BC_LEX_OP_MINUS) + if (type != XC_LEX_OP_MINUS) bc_vec_push(&p->ops, &type); else bc_parse_operator(p, type, ops_bgn, nexprs); @@ -4523,9 +4524,9 @@ static BC_STATUS zbc_parse_stmt_possibly_auto(BcParse *p, bool auto_allowed) dbg_lex("%s:%d p->l.t.t:%d", __func__, __LINE__, p->l.t.t); switch (p->l.t.t) { + case XC_LEX_OP_MINUS: case BC_LEX_OP_INC: case BC_LEX_OP_DEC: - case BC_LEX_OP_MINUS: case BC_LEX_OP_BOOL_NOT: case BC_LEX_LPAREN: case BC_LEX_NAME: @@ -4642,7 +4643,7 @@ static BcStatus bc_parse_expr_empty_ok(BcParse *p, uint8_t flags) s = zbc_parse_incdec(p, &prev, &paren_expr, &nexprs, flags); rprn = get_token = bin_last = false; break; - case BC_LEX_OP_MINUS: + case XC_LEX_OP_MINUS: s = zbc_parse_minus(p, &prev, ops_bgn, rprn, &nexprs); rprn = get_token = false; bin_last = prev == XC_INST_MINUS; @@ -4665,17 +4666,17 @@ static BcStatus bc_parse_expr_empty_ok(BcParse *p, uint8_t flags) break; } // Fallthrough. - case BC_LEX_OP_POWER: - case BC_LEX_OP_MULTIPLY: - case BC_LEX_OP_DIVIDE: - case BC_LEX_OP_MODULUS: - case BC_LEX_OP_PLUS: - case BC_LEX_OP_REL_EQ: - case BC_LEX_OP_REL_LE: - case BC_LEX_OP_REL_GE: - case BC_LEX_OP_REL_NE: - case BC_LEX_OP_REL_LT: - case BC_LEX_OP_REL_GT: + case XC_LEX_OP_POWER: + case XC_LEX_OP_MULTIPLY: + case XC_LEX_OP_DIVIDE: + case XC_LEX_OP_MODULUS: + case XC_LEX_OP_PLUS: + case XC_LEX_OP_REL_EQ: + case XC_LEX_OP_REL_LE: + case XC_LEX_OP_REL_GE: + case XC_LEX_OP_REL_NE: + case XC_LEX_OP_REL_LT: + case XC_LEX_OP_REL_GT: case BC_LEX_OP_BOOL_NOT: case BC_LEX_OP_BOOL_OR: case BC_LEX_OP_BOOL_AND: @@ -4684,12 +4685,12 @@ static BcStatus bc_parse_expr_empty_ok(BcParse *p, uint8_t flags) ) { return bc_error_bad_expression(); } - nrelops += t >= BC_LEX_OP_REL_EQ && t <= BC_LEX_OP_REL_GT; + nrelops += (t >= XC_LEX_OP_REL_EQ && t <= XC_LEX_OP_REL_GT); prev = BC_TOKEN_2_INST(t); bc_parse_operator(p, t, ops_bgn, &nexprs); s = zbc_lex_next(&p->l); rprn = get_token = false; - bin_last = t != BC_LEX_OP_BOOL_NOT; + bin_last = (t != BC_LEX_OP_BOOL_NOT); break; case BC_LEX_LPAREN: if (BC_PARSE_LEAF(prev, rprn)) @@ -4793,7 +4794,7 @@ static BcStatus bc_parse_expr_empty_ok(BcParse *p, uint8_t flags) bc_parse_push(p, BC_TOKEN_2_INST(top)); - nexprs -= (top != BC_LEX_OP_BOOL_NOT && top != BC_LEX_NEG); + nexprs -= (top != BC_LEX_OP_BOOL_NOT && top != XC_LEX_NEG); bc_vec_pop(&p->ops); } @@ -4912,14 +4913,14 @@ static BC_STATUS zdc_parse_token(BcParse *p, BcLexType t) s = BC_STATUS_SUCCESS; get_token = true; switch (t) { - case BC_LEX_OP_REL_EQ: - case BC_LEX_OP_REL_LE: - case BC_LEX_OP_REL_GE: - case BC_LEX_OP_REL_NE: - case BC_LEX_OP_REL_LT: - case BC_LEX_OP_REL_GT: + case XC_LEX_OP_REL_EQ: + case XC_LEX_OP_REL_LE: + case XC_LEX_OP_REL_GE: + case XC_LEX_OP_REL_NE: + case XC_LEX_OP_REL_LT: + case XC_LEX_OP_REL_GT: dbg_lex("%s:%d LEX_OP_REL_xyz", __func__, __LINE__); - s = zdc_parse_cond(p, t - BC_LEX_OP_REL_EQ + XC_INST_REL_EQ); + s = zdc_parse_cond(p, t - XC_LEX_OP_REL_EQ + XC_INST_REL_EQ); get_token = false; break; case BC_LEX_SCOLON: @@ -4931,7 +4932,7 @@ static BC_STATUS zdc_parse_token(BcParse *p, BcLexType t) dbg_lex("%s:%d LEX_STR", __func__, __LINE__); dc_parse_string(p); break; - case BC_LEX_NEG: + case XC_LEX_NEG: dbg_lex("%s:%d LEX_NEG", __func__, __LINE__); s = zbc_lex_next(&p->l); if (s) RETURN_STATUS(s); -- cgit v1.2.3