=================================================================== RCS file: /cvs/mandoc/roff.c,v retrieving revision 1.75 retrieving revision 1.82 diff -u -p -r1.75 -r1.82 --- mandoc/roff.c 2010/05/16 13:49:23 1.75 +++ mandoc/roff.c 2010/05/17 02:01:05 1.82 @@ -1,4 +1,4 @@ -/* $Id: roff.c,v 1.75 2010/05/16 13:49:23 kristaps Exp $ */ +/* $Id: roff.c,v 1.82 2010/05/17 02:01:05 kristaps Exp $ */ /* * Copyright (c) 2010 Kristaps Dzonsons * @@ -26,44 +26,48 @@ #include "mandoc.h" #include "roff.h" +#define RSTACK_MAX 128 + #define ROFF_CTL(c) \ ('.' == (c) || '\'' == (c)) -#ifdef ROFF_DEBUG -#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_ccond, -#if 0 ROFF_am, ROFF_ami, + ROFF_am1, ROFF_de, ROFF_dei, + ROFF_de1, + ROFF_el, + ROFF_ie, + ROFF_if, ROFF_ig, - ROFF_close, -#endif + 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 */ \ @@ -79,32 +83,40 @@ typedef enum rofferr (*roffproc)(ROFF_ARGS); struct roffmac { const char *name; /* macro name */ - roffproc proc; + 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 */ }; -static enum rofferr roff_if(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); -#if 0 -static enum rofferr roff_new_close(ROFF_ARGS); -static enum rofferr roff_new_ig(ROFF_ARGS); -static enum rofferr roff_sub_ig(ROFF_ARGS); -#endif +static enum rofferr roff_cond(ROFF_ARGS); +static enum rofferr roff_cond_text(ROFF_ARGS); +static enum rofferr roff_cond_sub(ROFF_ARGS); const struct roffmac roffs[ROFF_MAX] = { - { "if", roff_if }, - { "\\}", roff_ccond }, -#if 0 - { "am", roff_sub_ig, roff_new_ig }, - { "ami", roff_sub_ig, roff_new_ig }, - { "de", roff_sub_ig, roff_new_ig }, - { "dei", roff_sub_ig, roff_new_ig }, - { "ig", roff_sub_ig, roff_new_ig }, - { ".", NULL, roff_new_close }, -#endif + { "am", roff_block, roff_block_text, roff_block_sub, 0 }, + { "ami", roff_block, roff_block_text, roff_block_sub, 0 }, + { "am1", roff_block, roff_block_text, roff_block_sub, 0 }, + { "de", roff_block, roff_block_text, roff_block_sub, 0 }, + { "dei", roff_block, roff_block_text, roff_block_sub, 0 }, + { "de1", roff_block, roff_block_text, roff_block_sub, 0 }, + { "el", roff_cond, roff_cond_text, roff_cond_sub, ROFFMAC_STRUCT }, + { "ie", roff_cond, roff_cond_text, roff_cond_sub, ROFFMAC_STRUCT }, + { "if", roff_cond, roff_cond_text, roff_cond_sub, ROFFMAC_STRUCT }, + { "ig", roff_block, roff_block_text, roff_block_sub, 0 }, + { ".", roff_cblock, NULL, NULL, 0 }, + { "\\}", roff_ccond, NULL, NULL, 0 }, }; static void roff_free1(struct roff *); static enum rofft roff_hash_find(const char *); +static void roffnode_cleanscope(struct roff *); static int roffnode_push(struct roff *, enum rofft, int, int); static void roffnode_pop(struct roff *); @@ -141,6 +153,11 @@ 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); @@ -166,6 +183,7 @@ 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); @@ -210,6 +228,7 @@ roff_alloc(const mandocmsg msg, void *data) r->msg = msg; r->data = data; + r->rstackpos = -1; return(r); } @@ -221,28 +240,45 @@ roff_parseln(struct roff *r, int ln, enum rofft t; int ppos; - /* Return when in free text without a context. */ + /* + * 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])) { - /* XXX: this assumes we're just discarding. */ - while (r->last) { - if (r->last->endspan-- < 0) - break; - ROFF_MDEBUG(r, "closing implicit scope"); - roffnode_pop(r); - } - return(ROFF_IGN); + 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])) return(ROFF_CONT); - /* There's nothing on the stack: make us anew. */ + /* + * If a scope is open, go to the child handler for that macro, + * as it may want to preprocess before doing anything with it. + */ + 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))) return(ROFF_CONT); assert(roffs[t].proc); - return((*roffs[t].proc)(r, t, bufp, szp, ln, ppos, pos, offs)); + return((*roffs[t].proc) + (r, t, bufp, szp, ln, ppos, pos, offs)); } @@ -280,7 +316,7 @@ 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) @@ -298,181 +334,372 @@ roff_parse(const char *buf, int *pos) } -#if 0 /* ARGSUSED */ static enum rofferr -roff_sub_ig(ROFF_ARGS) +roff_cblock(ROFF_ARGS) { - int i, j; - /* Ignore free-text lines. */ + /* + * A block-close `..' should only be invoked as a child of an + * ignore macro, otherwise raise a warning and just ignore it. + */ - if ('.' != (*bufp)[ppos] && '\'' != (*bufp)[ppos]) + if (NULL == r->last) { + if ( ! (*r->msg)(MANDOCERR_NOSCOPE, r->data, ln, ppos, NULL)) + return(ROFF_ERR); return(ROFF_IGN); + } - if (r->last->end) { - i = ppos + 1; + 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); + } - while ((*bufp)[i] && ' ' == (*bufp)[i]) - i++; + if ((*bufp)[pos]) + if ( ! (*r->msg)(MANDOCERR_ARGSLOST, r->data, ln, pos, NULL)) + return(ROFF_ERR); - for (j = 0; r->last->end[j]; i++, j++) - if ((*bufp)[i] != r->last->end[j]) - return(ROFF_IGN); + roffnode_pop(r); + roffnode_cleanscope(r); + return(ROFF_IGN); - if (r->last->end[j]) - return(ROFF_IGN); - if ((*bufp)[i] && ' ' != (*bufp)[i]) - return(ROFF_IGN); +} - while (' ' == (*bufp)[i]) - i++; - } else if (ROFF_close != roff_parse(*bufp, &i)) - return(ROFF_IGN); +static void +roffnode_cleanscope(struct roff *r) +{ - roffnode_pop(r); - - if ('\0' == (*bufp)[i]) - return(ROFF_IGN); - if ( ! (*r->msg)(MANDOCERR_ARGSLOST, r->data, ln, i, NULL)) - return(ROFF_ERR); - - return(ROFF_IGN); + while (r->last) { + if (--r->last->endspan < 0) + break; + roffnode_pop(r); + } } /* ARGSUSED */ static enum rofferr -roff_new_close(ROFF_ARGS) +roff_ccond(ROFF_ARGS) { - if ( ! (*r->msg)(MANDOCERR_NOSCOPE, r->data, ln, ppos, NULL)) - return(ROFF_ERR); + if (NULL == r->last) { + if ( ! (*r->msg)(MANDOCERR_NOSCOPE, r->data, ln, ppos, NULL)) + return(ROFF_ERR); + return(ROFF_IGN); + } + 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); + } + + if (r->last->endspan > -1) { + if ( ! (*r->msg)(MANDOCERR_NOSCOPE, r->data, ln, ppos, NULL)) + return(ROFF_ERR); + return(ROFF_IGN); + } + + if ((*bufp)[pos]) + if ( ! (*r->msg)(MANDOCERR_ARGSLOST, r->data, ln, pos, NULL)) + return(ROFF_ERR); + + roffnode_pop(r); + roffnode_cleanscope(r); return(ROFF_IGN); } -#endif /* ARGSUSED */ static enum rofferr -roff_ccond(ROFF_ARGS) +roff_block(ROFF_ARGS) { + int sv; + size_t sz; - if (NULL == r->last || ROFF_if != r->last->tok || r->last->endspan > -1) { - if ( ! (*r->msg)(MANDOCERR_NOSCOPE, r->data, ln, ppos, NULL)) + 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++; } - ROFF_MDEBUG(r, "closing explicit scope"); - roffnode_pop(r); + if ( ! roffnode_push(r, tok, ln, ppos)) + return(ROFF_ERR); - while (r->last) { - if (--r->last->endspan < 0) - break; + if ('\0' == (*bufp)[pos]) + return(ROFF_IGN); - ROFF_MDEBUG(r, "closing implicit scope"); - roffnode_pop(r); + sv = pos; + while ((*bufp)[pos] && ' ' != (*bufp)[pos] && + '\t' != (*bufp)[pos]) + pos++; + + /* + * Note: groff does NOT like escape characters in the input. + * Instead of detecting this, we're just going to let it fly and + * to hell with it. + */ + + 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) { + (*r->msg)(MANDOCERR_MEM, r->data, ln, pos, NULL); + return(ROFF_ERR); } + memcpy(r->last->end, *bufp + sv, sz); + r->last->end[(int)sz] = '\0'; + + if ((*bufp)[pos]) + if ( ! (*r->msg)(MANDOCERR_ARGSLOST, r->data, ln, pos, NULL)) + return(ROFF_ERR); + return(ROFF_IGN); } /* ARGSUSED */ static enum rofferr -roff_if(ROFF_ARGS) +roff_block_sub(ROFF_ARGS) { + enum rofft t; + int i, j; /* - * 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. + * 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 ( ! roffnode_push(r, tok, ln, ppos)) - return(ROFF_ERR); + if (r->last->end) { + i = pos + 1; + while (' ' == (*bufp)[i] || '\t' == (*bufp)[i]) + i++; - while ((*bufp)[pos] && ' ' != (*bufp)[pos]) - pos++; - while (' ' == (*bufp)[pos]) - pos++; + for (j = 0; r->last->end[j]; j++, i++) + if ((*bufp)[i] != r->last->end[j]) + break; - /* Don't evaluate: just assume NO. */ + if ('\0' == r->last->end[j] && + ('\0' == (*bufp)[i] || + ' ' == (*bufp)[i] || + '\t' == (*bufp)[i])) { + roffnode_pop(r); + roffnode_cleanscope(r); - r->last->endspan = 1; + if (ROFF_MAX != roff_parse(*bufp, &pos)) + return(ROFF_RERUN); + return(ROFF_IGN); + } + } - 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 we have no custom end-query or lookup failed, then try + * pulling it out of the hashtable. + */ - if ('\0' == (*bufp)[pos]) + 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); - *offs = pos; - return(ROFF_RERUN); + assert(roffs[t].proc); + return((*roffs[t].proc)(r, t, bufp, + szp, ln, ppos, pos, offs)); } -#if 0 +/* ARGSUSED */ static enum rofferr -roff_new_ig(ROFF_ARGS) +roff_block_text(ROFF_ARGS) { - int i; - if ( ! roffnode_push(r, tok, ln, ppos)) - return(ROFF_ERR); - + return(ROFF_IGN); +} + + +/* ARGSUSED */ +static enum rofferr +roff_cond_sub(ROFF_ARGS) +{ + enum rofft t; + enum roffrule rr; + + ppos = pos; + rr = r->last->rule; + + roffnode_cleanscope(r); + + if (ROFF_MAX == (t = roff_parse(*bufp, &pos))) + return(ROFFRULE_DENY == rr ? ROFF_IGN : ROFF_CONT); + /* - * Other macros (not `ig') using this routine have additional - * crap here that we discard. + * 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); - if (ROFF_ig != tok) { - while ((*bufp)[ppos] && ' ' != (*bufp)[ppos]) - ppos++; - while (' ' == (*bufp)[ppos]) - ppos++; + 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(ROFFRULE_DENY == rr ? ROFF_IGN : ROFF_CONT); } - i = (int)ppos; + if (ep > st && '\\' != *(ep - 1)) + roffnode_pop(r); - while ((*bufp)[i] && ' ' != (*bufp)[i]) - i++; + roffnode_cleanscope(r); + return(ROFFRULE_DENY == rr ? ROFF_IGN : ROFF_CONT); +} - if (i == (int)ppos) - return(ROFF_IGN); - if ((*bufp)[i]) - if ( ! (*r->msg)(MANDOCERR_ARGSLOST, r->data, ln, i, NULL)) - return(ROFF_ERR); +/* ARGSUSED */ +static enum rofferr +roff_cond(ROFF_ARGS) +{ + int sv; + /* Stack overflow! */ + + if (ROFF_ie == tok && r->rstackpos == RSTACK_MAX - 1) { + (*r->msg)(MANDOCERR_MEM, r->data, ln, ppos, NULL); + return(ROFF_ERR); + } + + 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++; + /* - * If the macro has arguments, the first argument (up to the - * next whitespace) is interpreted as an argument marking the - * macro close. Thus, `.ig foo' will close at `.foo'. - * - * NOTE: the closing macro `.foo' in the above case is not - * allowed to have leading spaces with old groff! Thus `.foo' - * != `. foo'. Oh yeah, everything after the `.foo' is lost. - * Merry fucking Christmas. + * Roff is weird. If we have just white-space after the + * conditional, it's considered the BODY and we exit without + * 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); + return(ROFF_IGN); + } - r->last->end = malloc((size_t)(i - ppos) + 1); - if (NULL == r->last->end) { - (*r->msg)(MANDOCERR_MEM, r->data, ln, ppos, NULL); + if ( ! roffnode_push(r, tok, ln, ppos)) return(ROFF_ERR); + + /* TODO: here we would evaluate the conditional. */ + + 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]; + } else 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; } - memcpy(r->last->end, &(*bufp)[ppos], (size_t)(i - ppos)); - r->last->end[i - ppos] = '\0'; + r->last->endspan = 1; - return(ROFF_IGN); + if ('\\' == (*bufp)[pos] && '{' == (*bufp)[pos + 1]) { + r->last->endspan = -1; + pos += 2; + } + + /* + * If there are no arguments on the line, the next-line scope is + * assumed. + */ + + if ('\0' == (*bufp)[pos]) + return(ROFF_IGN); + + /* Otherwise re-run the roff parser after recalculating. */ + + *offs = pos; + return(ROFF_RERUN); } -#endif