=================================================================== RCS file: /cvs/mandoc/roff.c,v retrieving revision 1.78 retrieving revision 1.86 diff -u -p -r1.78 -r1.86 --- mandoc/roff.c 2010/05/16 22:28:33 1.78 +++ mandoc/roff.c 2010/06/01 11:47:28 1.86 @@ -1,4 +1,4 @@ -/* $Id: roff.c,v 1.78 2010/05/16 22:28:33 kristaps Exp $ */ +/* $Id: roff.c,v 1.86 2010/06/01 11:47:28 kristaps Exp $ */ /* * Copyright (c) 2010 Kristaps Dzonsons * @@ -19,6 +19,7 @@ #endif #include +#include #include #include #include @@ -26,45 +27,51 @@ #include "mandoc.h" #include "roff.h" +#define RSTACK_MAX 128 + #define ROFF_CTL(c) \ ('.' == (c) || '\'' == (c)) -#if 0 -#define ROFF_MDEBUG(p, str) \ - fprintf(stderr, "%s: %s (%d:%d)\n", (str), \ - roffs[(p)->last->tok].name, \ - (p)->last->line, (p)->last->col) -#else -#define ROFF_MDEBUG(p, str) while (/* CONSTCOND */ 0) -#endif enum rofft { - ROFF_if, - ROFF_ig, - ROFF_cblock, - ROFF_ccond, -#if 0 ROFF_am, ROFF_ami, + ROFF_am1, ROFF_de, ROFF_dei, - ROFF_close, -#endif + ROFF_de1, + ROFF_ds, + ROFF_el, + ROFF_ie, + ROFF_if, + ROFF_ig, + ROFF_rm, + ROFF_tr, + ROFF_cblock, + ROFF_ccond, ROFF_MAX }; +enum roffrule { + ROFFRULE_ALLOW, + ROFFRULE_DENY +}; + struct roff { struct roffnode *last; /* leaf of stack */ mandocmsg msg; /* err/warn/fatal messages */ void *data; /* privdata for messages */ + enum roffrule rstack[RSTACK_MAX]; /* stack of !`ie' rules */ + int rstackpos; /* position in rstack */ }; struct roffnode { enum rofft tok; /* type of node */ struct roffnode *parent; /* up one in stack */ - char *end; /* end-token: custom */ int line; /* parse line */ int col; /* parse col */ - int endspan; + char *end; /* end-rules: custom token */ + int endspan; /* end-rules: next-line or infty */ + enum roffrule rule; /* current evaluation rule */ }; #define ROFF_ARGS struct roff *r, /* parse ctx */ \ @@ -80,33 +87,84 @@ typedef enum rofferr (*roffproc)(ROFF_ARGS); struct roffmac { const char *name; /* macro name */ - roffproc proc; - roffproc text; + roffproc proc; /* process new macro */ + roffproc text; /* process as child text of macro */ + roffproc sub; /* process as child of macro */ + int flags; +#define ROFFMAC_STRUCT (1 << 0) /* always interpret */ + struct roffmac *next; }; -static enum rofferr roff_if(ROFF_ARGS); -static enum rofferr roff_if_text(ROFF_ARGS); -static enum rofferr roff_ig(ROFF_ARGS); -static enum rofferr roff_ig_text(ROFF_ARGS); +static enum rofferr roff_block(ROFF_ARGS); +static enum rofferr roff_block_text(ROFF_ARGS); +static enum rofferr roff_block_sub(ROFF_ARGS); static enum rofferr roff_cblock(ROFF_ARGS); static enum rofferr roff_ccond(ROFF_ARGS); +static enum rofferr roff_cond(ROFF_ARGS); +static enum rofferr roff_cond_text(ROFF_ARGS); +static enum rofferr roff_cond_sub(ROFF_ARGS); +static enum rofferr roff_line(ROFF_ARGS); -const struct roffmac roffs[ROFF_MAX] = { - { "if", roff_if, roff_if_text }, - { "ig", roff_ig, roff_ig_text }, - { ".", roff_cblock, NULL }, - { "\\}", roff_ccond, NULL }, +/* See roff_hash_find() */ + +#define ASCII_HI 126 +#define ASCII_LO 33 +#define HASHWIDTH (ASCII_HI - ASCII_LO + 1) + +static struct roffmac *hash[HASHWIDTH]; + +static struct roffmac roffs[ROFF_MAX] = { + { "am", roff_block, roff_block_text, roff_block_sub, 0, NULL }, + { "ami", roff_block, roff_block_text, roff_block_sub, 0, NULL }, + { "am1", roff_block, roff_block_text, roff_block_sub, 0, NULL }, + { "de", roff_block, roff_block_text, roff_block_sub, 0, NULL }, + { "dei", roff_block, roff_block_text, roff_block_sub, 0, NULL }, + { "de1", roff_block, roff_block_text, roff_block_sub, 0, NULL }, + { "ds", roff_line, NULL, NULL, 0, NULL }, + { "el", roff_cond, roff_cond_text, roff_cond_sub, ROFFMAC_STRUCT, NULL }, + { "ie", roff_cond, roff_cond_text, roff_cond_sub, ROFFMAC_STRUCT, NULL }, + { "if", roff_cond, roff_cond_text, roff_cond_sub, ROFFMAC_STRUCT, NULL }, + { "ig", roff_block, roff_block_text, roff_block_sub, 0, NULL }, + { "rm", roff_line, NULL, NULL, 0, NULL }, + { "tr", roff_line, NULL, NULL, 0, NULL }, + { ".", roff_cblock, NULL, NULL, 0, NULL }, + { "\\}", roff_ccond, NULL, NULL, 0, NULL }, }; static void roff_free1(struct roff *); static enum rofft roff_hash_find(const char *); +static void roff_hash_init(void); static void roffnode_cleanscope(struct roff *); static int roffnode_push(struct roff *, enum rofft, int, int); static void roffnode_pop(struct roff *); static enum rofft roff_parse(const char *, int *); +/* See roff_hash_find() */ +#define ROFF_HASH(p) (p[0] - ASCII_LO) +static void +roff_hash_init(void) +{ + struct roffmac *n; + int buc, i; + + for (i = 0; i < (int)ROFF_MAX; i++) { + assert(roffs[i].name[0] >= ASCII_LO); + assert(roffs[i].name[0] <= ASCII_HI); + + buc = ROFF_HASH(roffs[i].name); + + if (NULL != (n = hash[buc])) { + for ( ; n->next; n = n->next) + /* Do nothing. */ ; + n->next = &roffs[i]; + } else + hash[buc] = &roffs[i]; + } +} + + /* * Look up a roff token by its name. Returns ROFF_MAX if no macro by * the nil-terminated string name could be found. @@ -114,14 +172,27 @@ static enum rofft roff_parse(const char *, int *); static enum rofft roff_hash_find(const char *p) { - int i; + int buc; + struct roffmac *n; - /* FIXME: make this be fast and efficient. */ + /* + * libroff has an extremely simple hashtable, for the time + * being, which simply keys on the first character, which must + * be printable, then walks a chain. It works well enough until + * optimised. + */ - for (i = 0; i < (int)ROFF_MAX; i++) - if (0 == strcmp(roffs[i].name, p)) - return((enum rofft)i); + if (p[0] < ASCII_LO || p[0] > ASCII_HI) + return(ROFF_MAX); + buc = ROFF_HASH(p); + + if (NULL == (n = hash[buc])) + return(ROFF_MAX); + for ( ; n; n = n->next) + if (0 == strcmp(n->name, p)) + return((enum rofft)(n - roffs)); + return(ROFF_MAX); } @@ -137,6 +208,11 @@ roffnode_pop(struct roff *r) assert(r->last); p = r->last; + + if (ROFF_el == p->tok) + if (r->rstackpos > -1) + r->rstackpos--; + r->last = r->last->parent; if (p->end) free(p->end); @@ -162,6 +238,7 @@ roffnode_push(struct roff *r, enum rofft tok, int line p->parent = r->last; p->line = line; p->col = col; + p->rule = p->parent ? p->parent->rule : ROFFRULE_DENY; r->last = p; return(1); @@ -206,6 +283,9 @@ roff_alloc(const mandocmsg msg, void *data) r->msg = msg; r->data = data; + r->rstackpos = -1; + + roff_hash_init(); return(r); } @@ -215,69 +295,44 @@ roff_parseln(struct roff *r, int ln, char **bufp, size_t *szp, int pos, int *offs) { enum rofft t; - int ppos, i, j, wtf; + int ppos; + /* + * First, if a scope is open and we're not a macro, pass the + * text through the macro's filter. If a scope isn't open and + * we're not a macro, just let it through. + */ + if (r->last && ! ROFF_CTL((*bufp)[pos])) { - /* - * If a scope is open and we're not a macro, pass it - * through our text detector and continue as quickly as - * possible. - */ t = r->last->tok; assert(roffs[t].text); return((*roffs[t].text) (r, t, bufp, szp, ln, pos, pos, offs)); } else if ( ! ROFF_CTL((*bufp)[pos])) - /* - * Don't do anything if we're free-form text. - */ return(ROFF_CONT); - /* A macro-ish line with a possibly-open macro context. */ + /* + * If a scope is open, go to the child handler for that macro, + * as it may want to preprocess before doing anything with it. + */ - wtf = 0; - - if (r->last && r->last->end) { - /* - * We have a scope open that has a custom end-macro - * handler. Try to match it against the input. - */ - i = pos + 1; - while (' ' == (*bufp)[i] || '\t' == (*bufp)[i]) - i++; - - for (j = 0; r->last->end[j]; j++, i++) - if ((*bufp)[i] != r->last->end[j]) - break; - - if ('\0' == r->last->end[j] && - ('\0' == (*bufp)[i] || - ' ' == (*bufp)[i] || - '\t' == (*bufp)[i])) { - roffnode_pop(r); - roffnode_cleanscope(r); - wtf = 1; - } + if (r->last) { + t = r->last->tok; + assert(roffs[t].sub); + return((*roffs[t].sub) + (r, t, bufp, szp, ln, pos, pos, offs)); } + /* + * Lastly, as we've no scope open, try to look up and execute + * the new macro. If no macro is found, simply return and let + * the compilers handle it. + */ + ppos = pos; - if (ROFF_MAX == (t = roff_parse(*bufp, &pos))) { - /* - * This is some of groff's stranger behaviours. If we - * encountered a custom end-scope tag and that tag also - * happens to be a "real" macro, then we need to try - * interpreting it again as a real macro. If it's not, - * then return ignore. Else continue. - */ - if (wtf) - return(ROFF_IGN); - else if (NULL == r->last) - return(ROFF_CONT); + if (ROFF_MAX == (t = roff_parse(*bufp, &pos))) + return(ROFF_CONT); - /* FIXME: this assumes that we ignore!? */ - return(ROFF_IGN); - } - assert(roffs[t].proc); return((*roffs[t].proc) (r, t, bufp, szp, ln, ppos, pos, offs)); @@ -318,7 +373,7 @@ roff_parse(const char *buf, int *pos) for (j = 0; j < 4; j++, (*pos)++) if ('\0' == (mac[j] = buf[*pos])) break; - else if (' ' == buf[*pos]) + else if (' ' == buf[*pos] || (j && '\\' == buf[*pos])) break; if (j == 4 || j < 1) @@ -341,13 +396,33 @@ static enum rofferr roff_cblock(ROFF_ARGS) { + /* + * A block-close `..' should only be invoked as a child of an + * ignore macro, otherwise raise a warning and just ignore it. + */ + if (NULL == r->last) { if ( ! (*r->msg)(MANDOCERR_NOSCOPE, r->data, ln, ppos, NULL)) return(ROFF_ERR); return(ROFF_IGN); } - if (ROFF_ig != r->last->tok) { + switch (r->last->tok) { + case (ROFF_am): + /* FALLTHROUGH */ + case (ROFF_ami): + /* FALLTHROUGH */ + case (ROFF_am1): + /* FALLTHROUGH */ + case (ROFF_de): + /* FALLTHROUGH */ + case (ROFF_dei): + /* FALLTHROUGH */ + case (ROFF_de1): + /* FALLTHROUGH */ + case (ROFF_ig): + break; + default: if ( ! (*r->msg)(MANDOCERR_NOSCOPE, r->data, ln, ppos, NULL)) return(ROFF_ERR); return(ROFF_IGN); @@ -357,7 +432,6 @@ roff_cblock(ROFF_ARGS) if ( ! (*r->msg)(MANDOCERR_ARGSLOST, r->data, ln, pos, NULL)) return(ROFF_ERR); - ROFF_MDEBUG(r, "closing ignore block"); roffnode_pop(r); roffnode_cleanscope(r); return(ROFF_IGN); @@ -372,7 +446,6 @@ roffnode_cleanscope(struct roff *r) while (r->last) { if (--r->last->endspan < 0) break; - ROFF_MDEBUG(r, "closing implicit scope"); roffnode_pop(r); } } @@ -389,7 +462,14 @@ roff_ccond(ROFF_ARGS) return(ROFF_IGN); } - if (ROFF_if != r->last->tok) { + switch (r->last->tok) { + case (ROFF_el): + /* FALLTHROUGH */ + case (ROFF_ie): + /* FALLTHROUGH */ + case (ROFF_if): + break; + default: if ( ! (*r->msg)(MANDOCERR_NOSCOPE, r->data, ln, ppos, NULL)) return(ROFF_ERR); return(ROFF_IGN); @@ -405,7 +485,6 @@ roff_ccond(ROFF_ARGS) if ( ! (*r->msg)(MANDOCERR_ARGSLOST, r->data, ln, pos, NULL)) return(ROFF_ERR); - ROFF_MDEBUG(r, "closing explicit scope"); roffnode_pop(r); roffnode_cleanscope(r); return(ROFF_IGN); @@ -414,18 +493,27 @@ roff_ccond(ROFF_ARGS) /* ARGSUSED */ static enum rofferr -roff_ig(ROFF_ARGS) +roff_block(ROFF_ARGS) { int sv; size_t sz; + if (ROFF_ig != tok && '\0' == (*bufp)[pos]) { + if ( ! (*r->msg)(MANDOCERR_NOARGS, r->data, ln, ppos, NULL)) + return(ROFF_ERR); + return(ROFF_IGN); + } else if (ROFF_ig != tok) { + while ((*bufp)[pos] && ' ' != (*bufp)[pos]) + pos++; + while (' ' == (*bufp)[pos]) + pos++; + } + if ( ! roffnode_push(r, tok, ln, ppos)) return(ROFF_ERR); - if ('\0' == (*bufp)[pos]) { - ROFF_MDEBUG(r, "opening ignore block"); + if ('\0' == (*bufp)[pos]) return(ROFF_IGN); - } sv = pos; while ((*bufp)[pos] && ' ' != (*bufp)[pos] && @@ -441,6 +529,9 @@ roff_ig(ROFF_ARGS) assert(pos > sv); sz = (size_t)(pos - sv); + if (1 == sz && '.' == (*bufp)[sv]) + return(ROFF_IGN); + r->last->end = malloc(sz + 1); if (NULL == r->last->end) { @@ -451,8 +542,6 @@ roff_ig(ROFF_ARGS) memcpy(r->last->end, *bufp + sv, sz); r->last->end[(int)sz] = '\0'; - ROFF_MDEBUG(r, "opening explicit ignore block"); - if ((*bufp)[pos]) if ( ! (*r->msg)(MANDOCERR_ARGSLOST, r->data, ln, pos, NULL)) return(ROFF_ERR); @@ -463,51 +552,155 @@ roff_ig(ROFF_ARGS) /* ARGSUSED */ static enum rofferr -roff_ig_text(ROFF_ARGS) +roff_block_sub(ROFF_ARGS) { + enum rofft t; + int i, j; + /* + * First check whether a custom macro exists at this level. If + * it does, then check against it. This is some of groff's + * stranger behaviours. If we encountered a custom end-scope + * tag and that tag also happens to be a "real" macro, then we + * need to try interpreting it again as a real macro. If it's + * not, then return ignore. Else continue. + */ + + if (r->last->end) { + i = pos + 1; + while (' ' == (*bufp)[i] || '\t' == (*bufp)[i]) + i++; + + for (j = 0; r->last->end[j]; j++, i++) + if ((*bufp)[i] != r->last->end[j]) + break; + + if ('\0' == r->last->end[j] && + ('\0' == (*bufp)[i] || + ' ' == (*bufp)[i] || + '\t' == (*bufp)[i])) { + roffnode_pop(r); + roffnode_cleanscope(r); + + if (ROFF_MAX != roff_parse(*bufp, &pos)) + return(ROFF_RERUN); + return(ROFF_IGN); + } + } + + /* + * If we have no custom end-query or lookup failed, then try + * pulling it out of the hashtable. + */ + + ppos = pos; + t = roff_parse(*bufp, &pos); + + /* If we're not a comment-end, then throw it away. */ + if (ROFF_cblock != t) + return(ROFF_IGN); + + assert(roffs[t].proc); + return((*roffs[t].proc)(r, t, bufp, + szp, ln, ppos, pos, offs)); +} + + +/* ARGSUSED */ +static enum rofferr +roff_block_text(ROFF_ARGS) +{ + return(ROFF_IGN); } /* ARGSUSED */ static enum rofferr -roff_if_text(ROFF_ARGS) +roff_cond_sub(ROFF_ARGS) { + enum rofft t; + enum roffrule rr; + + ppos = pos; + rr = r->last->rule; + + roff_cond_text(r, tok, bufp, szp, ln, ppos, pos, offs); + + if (ROFF_MAX == (t = roff_parse(*bufp, &pos))) + return(ROFFRULE_DENY == rr ? ROFF_IGN : ROFF_CONT); + + /* + * A denied conditional must evaluate its children if and only + * if they're either structurally required (such as loops and + * conditionals) or a closing macro. + */ + if (ROFFRULE_DENY == rr) + if ( ! (ROFFMAC_STRUCT & roffs[t].flags)) + if (ROFF_ccond != t) + return(ROFF_IGN); + + assert(roffs[t].proc); + return((*roffs[t].proc) + (r, t, bufp, szp, ln, ppos, pos, offs)); +} + + +/* ARGSUSED */ +static enum rofferr +roff_cond_text(ROFF_ARGS) +{ char *ep, *st; + enum roffrule rr; + rr = r->last->rule; + + /* + * We display the value of the text if out current evaluation + * scope permits us to do so. + */ + st = &(*bufp)[pos]; if (NULL == (ep = strstr(st, "\\}"))) { roffnode_cleanscope(r); - return(ROFF_IGN); + return(ROFFRULE_DENY == rr ? ROFF_IGN : ROFF_CONT); } - if (ep > st && '\\' != *(ep - 1)) { - ROFF_MDEBUG(r, "closing explicit scope (in-line)"); + if (ep == st || (ep > st && '\\' != *(ep - 1))) roffnode_pop(r); - } roffnode_cleanscope(r); - return(ROFF_IGN); + return(ROFFRULE_DENY == rr ? ROFF_IGN : ROFF_CONT); } /* ARGSUSED */ static enum rofferr -roff_if(ROFF_ARGS) +roff_cond(ROFF_ARGS) { + int cpos; /* position of the condition */ int sv; - /* - * Read ahead past the conditional. - * FIXME: this does not work, as conditionals don't end on - * whitespace, but are parsed according to a formal grammar. - * It's good enough for now, however. - */ + /* Stack overflow! */ - while ((*bufp)[pos] && ' ' != (*bufp)[pos]) - pos++; + if (ROFF_ie == tok && r->rstackpos == RSTACK_MAX - 1) { + (*r->msg)(MANDOCERR_MEM, r->data, ln, ppos, NULL); + return(ROFF_ERR); + } + cpos = pos; + + if (ROFF_if == tok || ROFF_ie == tok) { + /* + * Read ahead past the conditional. FIXME: this does + * not work, as conditionals don't end on whitespace, + * but are parsed according to a formal grammar. It's + * good enough for now, however. + */ + while ((*bufp)[pos] && ' ' != (*bufp)[pos]) + pos++; + } + sv = pos; while (' ' == (*bufp)[pos]) pos++; @@ -518,7 +711,6 @@ roff_if(ROFF_ARGS) * really doing anything. Warn about this. It's probably * wrong. */ - if ('\0' == (*bufp)[pos] && sv != pos) { if ( ! (*r->msg)(MANDOCERR_NOARGS, r->data, ln, ppos, NULL)) return(ROFF_ERR); @@ -528,16 +720,41 @@ roff_if(ROFF_ARGS) if ( ! roffnode_push(r, tok, ln, ppos)) return(ROFF_ERR); - /* Don't evaluate: just assume NO. */ + /* XXX: Implement more conditionals. */ + if (ROFF_if == tok || ROFF_ie == tok) + r->last->rule = 'n' == (*bufp)[cpos] ? + ROFFRULE_ALLOW : ROFFRULE_DENY; + else if (ROFF_el == tok) { + /* + * An `.el' will get the value of the current rstack + * entry set in prior `ie' calls or defaults to DENY. + */ + if (r->rstackpos < 0) + r->last->rule = ROFFRULE_DENY; + else + r->last->rule = r->rstack[r->rstackpos]; + } + if (ROFF_ie == tok) { + /* + * An if-else will put the NEGATION of the current + * evaluated conditional into the stack. + */ + r->rstackpos++; + if (ROFFRULE_DENY == r->last->rule) + r->rstack[r->rstackpos] = ROFFRULE_ALLOW; + else + r->rstack[r->rstackpos] = ROFFRULE_DENY; + } + if (r->last->parent && ROFFRULE_DENY == r->last->parent->rule) + r->last->rule = ROFFRULE_DENY; + r->last->endspan = 1; if ('\\' == (*bufp)[pos] && '{' == (*bufp)[pos + 1]) { - ROFF_MDEBUG(r, "opening explicit scope"); r->last->endspan = -1; pos += 2; - } else - ROFF_MDEBUG(r, "opening implicit scope"); + } /* * If there are no arguments on the line, the next-line scope is @@ -551,4 +768,13 @@ roff_if(ROFF_ARGS) *offs = pos; return(ROFF_RERUN); +} + + +/* ARGSUSED */ +static enum rofferr +roff_line(ROFF_ARGS) +{ + + return(ROFF_IGN); }