X-Git-Url: https://ocean-lang.org/code/?a=blobdiff_plain;f=csrc%2Fscanner.mdc;h=6b706411f5010e3da61b6214742f3db39e91de77;hb=850a39a0a761e0af89c15253f075ecd9e9ecc6ee;hp=9a1ea7144b612a6624e39d799921c37682d46c99;hpb=5513fc2e3fb56bdf2292e834077e0c33f9a5c2a9;p=ocean diff --git a/csrc/scanner.mdc b/csrc/scanner.mdc index 9a1ea71..6b70641 100644 --- a/csrc/scanner.mdc +++ b/csrc/scanner.mdc @@ -69,13 +69,15 @@ The scanner is not completely general, yet not completely specified. There are a fixed set of token types, though particular tokens within those types can be distinguish via configuration. -Most token types may be explicitly ignored, as typically comments -would be. The exact consequence of ignoring each token type varies -from token to token. +Most token types may be explicitly ignored, so they aren't parsed. +Comments typically parsed but not returned, but an option is provided to +return comments for further processing. The exact consequence of +ignoring each token type varies from token to token. ###### public types struct token_config { int ignored; // bit set of ignored tokens. + int return_comments; ## token config parameters }; @@ -85,7 +87,6 @@ from token to token. ###### token_next init int ignored = state->conf->ignored; - The different tokens are numbers, words, marks, strings, comments, newlines, EOF, and indents, each of which is examined in detail below. @@ -120,7 +121,11 @@ To make matters worse, our language designer has decided to experiment with allowing commas to be used as the decimal indicator, and spaces to be used to separate groups of digits in large numbers. Both of these can reasonably be restricted to appear between two digits, so we -have to add that condition to our tests. +have to add that condition to our tests. For consistency we require +every non-alpha-numeric to appear between two hex digits, with the +exception that a sign can appear only after a 'p' or 'e', and a space +can only appear between decimal digits. Allowing a space before a +letter easily leads to confusion, such a in `a < 3 and b < 4`. So we cannot just treat numbers as starting with a digit and being followed by some set of characters. We need more structure than that. @@ -128,13 +133,16 @@ followed by some set of characters. We need more structure than that. So: - Numbers must start with a digit. -- If the first digit is zero, the next character must be a base - signifier (one of `xob`) or a decimal marker (`.` or `,`). - In the first case the first `p` or `P` may be followed by a sign. +- If the first digit is zero, the next character should be a base + signifier (one of `xob`) or a decimal marker (`.` or `,`) (though this isn't + enforced at this stage) + In the first case the only first `p` or `P` may be followed by a sign. - If the number doesn't start with `0` followed by one of `xob`, the first `e` may be followed by a sign. -- Any digit or hex digit may be followed by a space or underscore - providing that the subsequence character is also a (hex) digit. +- A sign must always be followed by a digit. +- Any digit may be followed by a space or underscore and any hex digit + maybe followed by an underscore, providing that the subsequence character + is also a digit (for space) or hex digit (for underscore). This rule will require an extra level of 'unget' to be supported when handling characters. - Otherwise any digits or ASCII letters are allowed. We do not at @@ -164,7 +172,7 @@ are declared to be a start character for words. ###### parse number if (iswdigit(ch) && !(ignored & (1<conf->number_chars, ch)) { + /* non-number char */ + break; } if (ch == '+' || ch == '-') { + /* previous must be 'e' or 'p' in appropraite context */ if (!sign_ok) break; expect_p = -1; + } else if (ch == ' ') { + /* previous must be a digit */ + if (!iswdigit(prev)) + break; + } else { + /* previous must be a hex digit */ + if (!iswxdigit(prev)) + break; } if (ch == '.' || ch == ',') { + /* only one of these permitted */ if (decimal_mark) break; decimal_mark = 1; } - if (prev_special) { - /* Don't allow that special char, - * need two 'ungets' - */ - restore_unget_state(state); - break; - } - if (strchr(state->conf->number_chars, ch)) { - prev_special = 1; - continue; - } - /* non-number char */ - break; } /* We seem to have a "number" token */ unget_char(state); @@ -329,10 +356,10 @@ immediately before a string is handled correctly. If the first character of a comment marker (i.e. '/') is a known mark, the above rules would suggest that the start of a comment would be -parsed as that mark, which is not what is wanted. So the introductory -sequences for a comment ("//" and "/*") are treated as -partially-known. They prevent the leading "/" from being a mark by -itself, but do not actually constitute a stand-alone mark. +parsed as that mark, which is not what is wanted. So when comments are +not ignored, the introductory sequences for a comment ("//" and "/*") +are treated as partially-known. They prevent the leading "/" from being +a mark by itself, but do not actually constitute a stand-alone mark. If `TK_mark` is ignored, then unknown marks are returned as errors. @@ -354,8 +381,7 @@ Known marks are included in the same list as the list of known words. /* found a longest-known-mark, still need to * check for comments */ - if (tk.txt.len == 2 && tk.txt.txt[0] == '/' && - (ch == '/' || ch == '*')) { + if (is_comment(ignored, tk.txt)) { /* Yes, this is a comment, not a '/' */ restore_unget_state(state); tk.num = TK_error; @@ -368,22 +394,21 @@ Known marks are included in the same list as the list of known words. prev = ch; save_unget_state(state); ch = get_char(state); - if (!(ignored && (1<= 0) + /* No need to worry about other token types */ + continue; + if (!(ignored & (1<= 3 && !(ignored & (1 << TK_multi_string)) && is_quote(tk.txt.txt[0]) && memcmp(tk.txt.txt, tk.txt.txt+1, 2) == 0 && @@ -557,19 +573,29 @@ still parsed, but is discarded. TK_block_comment, ###### internal functions - static int is_line_comment(struct text txt) + static int is_line_comment(int ignored, struct text txt) { + if (ignored & (1 << TK_line_comment)) + return 0; return (txt.len >= 1 && txt.txt[0] == '#') || (txt.len >= 2 && txt.txt[0] == '/' && txt.txt[1] == '/'); } - static int is_block_comment(struct text txt) + static int is_block_comment(int ignored, struct text txt) { + if (ignored & (1 << TK_block_comment)) + return 0; return txt.len >= 2 && txt.txt[0] == '/' && txt.txt[1] == '*'; } + static int is_comment(int ignored, struct text txt) + { + return is_line_comment(ignored, txt) || + is_block_comment(ignored, txt); + } + #### Single line comments A single-line comment continues up to, but not including the newline @@ -577,14 +603,14 @@ or end of node. ###### parse comment - if (is_line_comment(tk.txt)) { + if (is_line_comment(ignored, tk.txt)) { while (!is_newline(ch) && !at_eon(state)) ch = get_char(state); if (is_newline(ch)) unget_char(state); close_token(state, &tk); tk.num = TK_line_comment; - if (ignored & (1 << TK_line_comment)) + if (!state->conf->return_comments) continue; return tk; } @@ -601,7 +627,7 @@ the unget state (explained later). ###### parse comment - if (is_block_comment(tk.txt)) { + if (is_block_comment(ignored, tk.txt)) { wchar_t prev; int newlines = 0; reset_token(state, &tk); @@ -639,8 +665,7 @@ the unget state (explained later). if (!is_newline(ch)) tk.num = TK_error; } - if (tk.num == TK_error || - !(ignored & (1 << TK_block_comment))) + if (tk.num == TK_error || state->conf->return_comments) return tk; continue; } @@ -700,6 +725,8 @@ ignored. int indent_level; int indent_sizes[20]; +`indent_sizes[0]` will always be zero - this simplifies some code. + #### Newlines Newlines can optionally be reported. Newlines within a block comment @@ -734,7 +761,7 @@ blank lines that have been skipped. When a Newline leads to the next block of code there is a question of whether the various Newline and OUT/IN tokens should appear to -pbelong to the earlier or later block. This is addressed by processing +belong to the earlier or later block. This is addressed by processing the tokens in two stages based on the relative indent levels of the two blocks (each block has a base indent to which the actual indents are added). @@ -771,19 +798,29 @@ For this we store one more than the number of blank lines as int delayed_lines; int out_next; -Generating these tokens involve two separate pieces of code. +Generating these tokens involves two separate pieces of code. Firstly we need to recognise white space and count the indents and newlines. These are recorded in the above state fields. -Separately we need, on each call to `token_next`, we need to check if +Separately we need, on each call to `token_next`, to check if there are some delayed tokens and if so we need to advance the state information and return one token. +###### internal functions + static int state_indent(struct token_state *state) + { + if (state->node == NULL) + return state->col; + return state->node->indent - state->node->needs_strip + state->col; + } + ###### white space + if (is_newline(ch)) + state_check_node(state); if (is_newline(ch) || (at_son(state) && ch <= ' ')) { int newlines = 0; - int was_son = at_son(state); + int was_nl = is_newline(ch); if (ignored & (1<node->next && - state->node->next->indent > state->node->indent) - state->col = state->node->next->indent; - else - state->col = state->node->indent; - } else + if (ch != WEOF) unget_char(state); state->delayed_lines = newlines; - state->out_next = was_son; + state->out_next = !was_nl; state->check_indent = 1; continue; } - ###### delayed tokens if (state->check_indent || state->delayed_lines) { - if (state->col < state->indent_sizes[state->indent_level]) { + if (state_indent(state) < state->indent_sizes[state->indent_level]) { if (!state->out_next && !(ignored & (1<out_next = 1; @@ -830,11 +861,12 @@ information and return one token. tk.num = TK_out; return tk; } - if (state->col > state->indent_sizes[state->indent_level] && + if (state_indent(state) > state->indent_sizes[state->indent_level] && state->indent_level < sizeof(state->indent_sizes)-1) { state->indent_level += 1; - state->indent_sizes[state->indent_level] = state->col; - state->delayed_lines -= 1; + state->indent_sizes[state->indent_level] = state_indent(state); + if (state->delayed_lines) + state->delayed_lines -= 1; tk.num = TK_in; return tk; } @@ -857,14 +889,8 @@ tokens will continue to return the same end-of-file token. ###### token types TK_eof, - ###### white space if (ch == WEOF) { - if (state->col) { - state->col = 0; - state->check_indent = 1; - continue; - } tk.num = TK_eof; return tk; } @@ -876,7 +902,21 @@ If the token we have is not empty and `TK_mark` is allowed, we have an unknown mark, otherwise this must be an error. ###### unknown mark - /* one unknown character */ + + /* one unknown mark character */ + if (tk.txt.len) { + close_token(state, &tk); + if (ignored & (1<node->needs_strip) { @@ -925,7 +966,28 @@ a flag that tells us whether or not we need to strip. n -= 4; } } - return indent; + } + + static void state_check_node(struct token_state *state) + { + if (!state->node) + return; + if (state->node->code.len > state->offset) + return; + + do + state->node = state->node->next; + while (state->node && state->node->code.txt == NULL); + state->offset = 0; + state->prev_offset = 0; + state->strip_offset = 0; + state->col = 0; + if (state->node == NULL) + return; + state->line = state->node->line_no; + do_strip(state); + state->col = state->node->needs_strip; + state->strip_offset = state->offset; } static wint_t get_char(struct token_state *state) @@ -934,18 +996,9 @@ a flag that tells us whether or not we need to strip. size_t n; mbstate_t mbstate; + state_check_node(state); if (state->node == NULL) return WEOF; - if (state->node->code.len <= state->offset) { - do - state->node = state->node->next; - while (state->node && state->node->code.txt == NULL); - state->offset = 0; - if (state->node == NULL) - return WEOF; - state->line = state->node->line_no; - state->col = do_strip(state); - } ## before get_char @@ -956,12 +1009,12 @@ a flag that tells us whether or not we need to strip. &mbstate); if (n == -2 || n == 0) { /* Not enough bytes - not really possible */ - next = '\n'; - state->offset = state->node->code.len; + next = '\n'; // NOTEST + state->offset = state->node->code.len; // NOTEST } else if (n == -1) { /* error */ - state->offset += 1; - next = 0x7f; // an illegal character + state->offset += 1; // NOTEST + next = 0x7f; // an illegal character // NOTEST } else state->offset += n; @@ -969,7 +1022,8 @@ a flag that tells us whether or not we need to strip. state->col += 1; } else if (is_newline(next)) { state->line += 1; - state->col = do_strip(state); + do_strip(state); + state->col = state->node->needs_strip; } else if (next == '\t') { state->col = indent_tab(state->col); } @@ -1088,8 +1142,7 @@ parsed too much already. For that there is `reset_token`. tok->txt.len = 0; } - -Tokens make not cross into the next `code_node`, and some tokens can +Tokens may not cross into the next `code_node`, and some tokens can include the newline at the and of a `code_node`, we must be able to easily check if we have reached the end. Equally we need to know if we are at the start of a node, as white space is treated a little @@ -1099,7 +1152,7 @@ differently there. static int at_son(struct token_state *state) { - return state->offset == 0; + return state->prev_offset <= state->strip_offset; } static int at_eon(struct token_state *state) @@ -1182,7 +1235,9 @@ As well as getting tokens, we need to be able to create the memset(state, 0, sizeof(*state)); state->node = code; state->line = code->line_no; - state->col = do_strip(state); + do_strip(state); + state->col = state->node->needs_strip; + state->strip_offset = state->offset; state->conf = conf; return state; } @@ -1309,7 +1364,7 @@ tokens. Now we just need C files to store them, and a mk file to make them. Converting a `TK_number` token to a numerical value is a slightly higher level task than lexical analysis, and slightly lower than -grammar parsing, so put it here - as an index if you like. +grammar parsing, so put it here - as an appendix if you like. Importantly it will be used by the same testing rig that is used for testing the token scanner. @@ -1334,10 +1389,10 @@ had never been initialised. int *placesp) { /* Accept digits up to 'base', ignore '_' and - * ' ' if they appear between two legal digits, - * and if `placesp` is not NULL, allow a single - * '.' or ',' and report the number of digits - * beyond there. + * (for base 10) ' ' if they appear between two + * legal digits, and if `placesp` is not NULL, + * allow a single '.' or ',' and report the number + * of digits beyond there. * Return number of characters processed (p), * or 0 if something illegal was found. */ @@ -1350,7 +1405,7 @@ had never been initialised. int dig; char c = tok.txt[p]; - if (c == '_' || c == ' ') { + if (c == '_' || (c == ' ' && base == 10)) { if (prev != Digit) goto bad; prev = Space; @@ -1470,7 +1525,7 @@ we need to record the number of places. We won't impose the number of places until we have the exponent as well. ###### number vars - int places =0; + int places = 0; mpz_t mant; int d; @@ -1520,7 +1575,6 @@ character `expc`. tok.txt += d; tok.len -= d; - Now that we have the mantissa and the exponent we can multiply them together, also allowing for the number of digits after the decimal mark. @@ -1584,7 +1638,6 @@ Now we are ready to parse a number: the base, mantissa, and exponent. If all goes well we check for the possible trailing letters and return. Return value is 1 for success and 0 for failure. - ###### number functions int number_parse(mpq_t num, char tail[3], struct text tok) { @@ -1626,7 +1679,7 @@ Number parsing goes in `libnumber.c` ## number includes ## number functions -###### File: number.h +###### File: parse_number.h int number_parse(mpq_t num, char tail[3], struct text tok); ###### File: scanner.mk @@ -1910,7 +1963,7 @@ String parsing goes in `libstring.c` ## string functions ## string main -###### File: string.h +###### File: parse_string.h int string_parse(struct token *tok, char escape, struct text *str, char tail[3]); @@ -1919,7 +1972,6 @@ String parsing goes in `libstring.c` libstring.o : libstring.c $(CC) $(CFLAGS) -c libstring.c - ## Testing As "untested code is buggy code" we need a program to easily test @@ -1940,8 +1992,8 @@ the tokens one per line. #include #include "mdcode.h" #include "scanner.h" - #include "number.h" - #include "string.h" + #include "parse_number.h" + #include "parse_string.h" static int errs; static void pr_err(char *msg) @@ -1994,13 +2046,17 @@ the tokens one per line. { "ignore-newline", 0, NULL, 'l'}, { "ignore-block-comment", 0, NULL, 'C'}, { "ignore-indent", 0, NULL, 'i'}, + { "return-comments", 0, NULL, 'r'}, { "file", 1, NULL, 'f'}, + { "section", 1, NULL, 's'}, { NULL, 0, NULL, 0}, }; - static const char options[] = "W:w:n:NIMSzclCif:"; + static const char options[] = "W:w:n:NIMSzclCirf:s:"; struct section *table, *s, *prev; int opt; + char *section_name = NULL; + int section_found = 0; setlocale(LC_ALL,""); while ((opt = getopt_long(argc, argv, options, long_options, NULL)) @@ -2018,7 +2074,9 @@ the tokens one per line. case 'C': conf.ignored |= 1 << TK_block_comment; break; case 'l': conf.ignored |= 1 << TK_newline; break; case 'i': conf.ignored |= 1 << TK_in; break; + case 'r': conf.return_comments = 1; break; case 'f': filename = optarg; break; + case 's': section_name = optarg; break; default: fprintf(stderr, "scanner: unknown option '%c'.\n", opt); exit(1); @@ -2055,6 +2113,12 @@ the tokens one per line. for (s = table; s; (code_free(s->code), prev = s, s = s->next, free(prev))) { + if (section_name && + (s->section.len != strlen(section_name) || + strncmp(s->section.txt, section_name, s->section.len) != 0)) + continue; + if (section_name) + section_found = 1; printf("Tokenizing: %.*s\n", s->section.len, s->section.txt); state = token_open(s->code, &conf); @@ -2097,6 +2161,10 @@ the tokens one per line. } if (conf.words_marks != known) free(conf.words_marks); + if (section_name && !section_found) { + fprintf(stderr, "scanner: section %s not found\n", section_name); + errs = 1; + } exit(!!errs); } ###### File: scanner.mk