From a46157ba61f2c65f88b42abb31db60c4a714f87b Mon Sep 17 00:00:00 2001 From: Eduardo Chappa Date: Fri, 31 May 2013 17:16:44 -0600 Subject: * somehow all.patch got here. Reversing. --- pith/osdep/color.c | 1256 +--------------------------------------------------- pith/osdep/color.h | 25 +- 2 files changed, 2 insertions(+), 1279 deletions(-) (limited to 'pith/osdep') diff --git a/pith/osdep/color.c b/pith/osdep/color.c index cad5502e..faf3c675 100644 --- a/pith/osdep/color.c +++ b/pith/osdep/color.c @@ -31,7 +31,7 @@ static char rcsid[] = "$Id: color.c 761 2007-10-23 22:35:18Z hubert@u.washington #include #include "./color.h" -#include "./collate.h" + /* @@ -91,1257 +91,3 @@ pico_set_colorp(COLOR_PAIR *col, int flags) { return(pico_set_colors(col ? col->fg : NULL, col ? col->bg : NULL, flags)); } - - - /* - * Extended Justification support also does not belong here - * but otherwise webpine will not build, so we move everything - * here. Hopefully this will be the permanent place for these - * routines. These routines used to be in pico/word.c - */ -#define NSTRING 256 -#include "../../include/general.h" - -/* Support of indentation of paragraphs */ -#define is_indent_char(c) (((c) == '.' || (c) == '}' || (c) == RPAREN || \ - (c) == '*' || (c) == '+' || is_a_digit(c) || \ - ISspace(c) || (c) == '-' || \ - (c) == ']') ? 1 : 0) -#define allowed_after_digit(c,word,k) ((((c) == '.' && \ - allowed_after_period(next((word),(k)))) ||\ - (c) == RPAREN || (c) == '}' || (c) == ']' ||\ - ISspace(c) || is_a_digit(c) || \ - ((c) == '-' ) && \ - allowed_after_dash(next((word),(k)))) \ - ? 1 : 0) -#define allowed_after_period(c) (((c) == RPAREN || (c) == '}' || (c) == ']' ||\ - ISspace(c) || (c) == '-' || \ - is_a_digit(c)) ? 1 : 0) -#define allowed_after_parenth(c) (ISspace(c) ? 1 : 0) -#define allowed_after_space(c) (ISspace(c) ? 1 : 0) -#define allowed_after_braces(c) (ISspace(c) ? 1 : 0) -#define allowed_after_star(c) ((ISspace(c) || (c) == RPAREN ||\ - (c) == ']' || (c) == '}') ? 1 : 0) -#define allowed_after_dash(c) ((ISspace(c) || is_a_digit(c)) ? 1 : 0) -#define EOLchar(c) (((c) == '.' || (c) == ':' || (c) == '?' ||\ - (c) == '!') ? 1 : 0) - - -/* Extended justification support */ -#define is_cquote(c) ((c) == '>' || (c) == '|' || (c) == ']' || (c) == ':') -#define is_cword(c) ((((c) >= 'a') && ((c) <= 'z')) || \ - (((c) >= 'A') && ((c) <= 'Z')) || \ - (((c) >= '0') && ((c) <= '9')) || \ - ((c) == ' ') || ((c) == '?') || \ - ((c) == '@') || ((c) == '.') || \ - ((c) == '!') || ((c) == '\'') || \ - ((c) == ',') || ((c) == '\"') ? 1 : 0) -#define isaquote(c) ((c) == '\"' || (c) == '\'') -#define is8bit(c) ((((int) (c)) & 0x80) ? 1 : 0) -#define iscontrol(c) (iscntrl(((int) (c)) & 0x7f) ? 1 : 0) -#define forbidden(c) (((c) == '\"') || ((c) == '\'') || ((c) == '$') ||\ - ((c) == ',') || ((c) == '.') || ((c) == '-') ||\ - ((c) == LPAREN) || ((c) == '/')|| ((c) == '`') ||\ - ((c) == '{') || ((c) == '\\') || (iscontrol((c))) ||\ - (((c) >= '0') && ((c) <= '9')) || ((c) == '?')) -#define is_cletter(c) ((((c) >= 'a') && ((c) <= 'z'))) ||\ - ((((c) >= 'A') && ((c) <= 'Z'))||\ - is8bit(c)) -#define is_cnumber(c) ((c) >= '0' && (c) <= '9') -#define allwd_after_word(c) (((c) == ' ') || ((c) == '>') || is_cletter(c)) -#define allwd_after_qsword(c) (((c) != '\\') && ((c) != RPAREN)) -#define before(word,i) (((i) > 0) ? (word)[(i) - 1] : 0) -#define next(w,i) ((((w)[(i)]) != 0) ? ((w)[(i) + 1]) : 0) -#define now(w,i) ((w)[(i)]) -#define is_qsword(c) (((c) == ':') || ((c) == RPAREN) ? 1 : 0) -#define is_colon(c) (((c) == ':') ? 1 : 0) -#define is_rarrow(c) (((c) == '>') ? 1 : 0) -#define is_tilde(c) (((c) == '~') ? 1 : 0) -#define is_dash(c) (((c) == '-') ? 1 : 0) -#define is_pound(c) (((c) == '#') ? 1 : 0) -#define is_a_digit(c) ((((c) >= '0') && ((c) <= '9')) ? 1 : 0) -#define is_allowed(c) (is_cquote(c) || is_cword(c) || is_dash(c) || \ - is_pound(c)) -#define qs_allowed(a) (((a)->qstype != qsGdb) && ((a)->qstype != qsProg)) - -/* Internal justification functions */ -QSTRING_S *is_quote(char **, char *, int); -QSTRING_S *qs_normal_part(QSTRING_S *); -QSTRING_S *qs_remove_trailing_spaces(QSTRING_S *); -QSTRING_S *trim_qs_from_cl(QSTRING_S *, QSTRING_S *, QSTRING_S *); -QSTRING_S *fix_qstring(QSTRING_S *, QSTRING_S *, QSTRING_S *); -QSTRING_S *fix_qstring_allowed(QSTRING_S *, QSTRING_S *, QSTRING_S *); -QSTRING_S *qs_add(char **, char *, QStrType, int, int, int, int); -QSTRING_S *remove_qsword(QSTRING_S *); -QSTRING_S *do_raw_quote_match(char **, char *, char *, char *, QSTRING_S **, QSTRING_S **); -void free_qs(QSTRING_S **); -int word_is_prog(char *); -int qstring_is_normal(QSTRING_S *); -int exists_good_part(QSTRING_S *); -int strcmp_qs(char *, char *); -int count_levels_qstring(QSTRING_S *); -int same_qstring(QSTRING_S *, QSTRING_S *); -int isaword(char *,int ,int); -int isamailbox(char *,int ,int); -int double_check_qstr(char *); - -int -word_is_prog(char *word) -{ - static char *list1[] = {"#include", - "#define", - "#ifdef", - "#ifndef", - "#elif", - "#if", - NULL}; - static char *list2[] = {"#else", - "#endif", - NULL}; - int i, j = strlen(word), k, rv = 0; - - for(i = 0; rv == 0 && list1[i] && (k = strlen(list1[i])) && k < j; i++) - if(!strncmp(list1[i], word, k) && ISspace(word[k])) - rv++; - - if(rv) - return rv; - - for(i = 0; rv == 0 && list2[i] && (k = strlen(list2[i])) && k <= j; i++) - if(!strncmp(list2[i], word, k) && (!word[k] || ISspace(word[k]))) - rv++; - - return rv; -} - -/* - * This function creates a qstring pointer with the information that - * is_quote handles to it. - * Parameters: - * qs - User supplied quote string - * word - The line of text that the user is trying to read/justify - * beginw - Where we need to start copying from - * endw - Where we end copying - * offset - Any offset in endw that we need to account for - * typeqs - type of the string to be created - * neednext - boolean, indicating if we need to compute the next field - * of leave it NULL - * - * It is a mistake to call this function if beginw >= endw + offset. - * Please note the equality sign in the above inequality (this is because - * we always assume that qstring->value != ""). - */ -QSTRING_S * -qs_add(char **qs, char word[NSTRING], QStrType typeqs, int beginw, int endw, - int offset, int neednext) -{ - QSTRING_S *qstring, *nextqs; - int i; - - qstring = (QSTRING_S *) malloc (sizeof(QSTRING_S)); - memset (qstring, 0, sizeof(QSTRING_S)); - qstring->qstype = qsNormal; - - if (beginw == 0){ - beginw = endw + offset; - qstring->qstype = typeqs; - } - - nextqs = neednext ? is_quote(qs, word+beginw, 1) : NULL; - - qstring->value = (char *) malloc((beginw+1)*sizeof(char)); - strncpy(qstring->value, word, beginw); - qstring->value[beginw] = '\0'; - - qstring->next = nextqs; - - return qstring; -} - -int -qstring_is_normal(QSTRING_S *cl) -{ - for (;cl && (cl->qstype == qsNormal); cl = cl->next); - return cl ? 0 : 1; -} - -/* - * Given a quote string, this function returns the part that is the leading - * normal part of it. (the normal part is the part that is tagged qsNormal, - * that is to say, the one that is not controversial at all (like qsString - * for example). - */ -QSTRING_S * -qs_normal_part(QSTRING_S *cl) -{ - - if (!cl) /* nothing in, nothing out */ - return cl; - - if (cl->qstype != qsNormal) - free_qs(&cl); - - if (cl) - cl->next = qs_normal_part(cl->next); - - return cl; -} - -/* - * this function removes trailing spaces from a quote string, but leaves the - * last one if there are trailing spaces - */ -QSTRING_S * -qs_remove_trailing_spaces(QSTRING_S *cl) -{ - QSTRING_S *rl = cl; - if (!cl) /* nothing in, nothing out */ - return cl; - - if (cl->next) - cl->next = qs_remove_trailing_spaces(cl->next); - else{ - if (value_is_space(cl->value)) - free_qs(&cl); - else{ - int i, l; - i = l = strlen(cl->value) - 1; - while (cl->value && cl->value[i] - && ISspace(cl->value[i])) - i--; - i += (i < l) ? 2 : 1; - cl->value[i] = '\0'; - } - } - return cl; -} - -/* - * This function returns if two strings are the same quote string. - * The call is not symmetric. cl must preceed the line nl. This function - * should be called for comparing the last part of cl and nl. - */ -int -strcmp_qs(char *valuecl, char *valuenl) -{ - int j; - - for (j = 0; valuecl[j] && (valuecl[j] == valuenl[j]); j++); - return !strcmp(valuecl, valuenl) - || (valuenl[j] && value_is_space(valuenl+j) - && value_is_space(valuecl+j) - && strlenis(valuecl+j) >= strlenis(valuenl+j)) - || (!valuenl[j] && value_is_space(valuecl+j)); -} - -int -count_levels_qstring(QSTRING_S *cl) -{ - int count; - for (count = 0; cl ; count++, cl = cl->next); - - return count; -} - -int -value_is_space(char *value) -{ - for (; value && *value && ISspace(*value); value++); - - return value && *value ? 0 : 1; -} - -void -free_qs(QSTRING_S **cl) -{ - if (!(*cl)) - return; - - if ((*cl)->next) - free_qs(&((*cl)->next)); - - (*cl)->next = (QSTRING_S *) NULL; - - if ((*cl)->value) - free((void *)(*cl)->value); - (*cl)->value = (char *) NULL; - free((void *)(*cl)); - *cl = (QSTRING_S *) NULL; -} - -/* - * This function returns the number of agreements between - * cl and nl. The call is not symmetric. cl must be the line - * preceding nl. - */ -int -same_qstring(QSTRING_S *cl, QSTRING_S *nl) -{ - int same = 0, done = 0; - - for (;cl && nl && !done; cl = cl->next, nl = nl->next) - if (cl->qstype == nl->qstype - && (!strcmp(cl->value, nl->value) - || (!cl->next && strcmp_qs(cl->value, nl->value)))) - same++; - else - done++; - return same; -} - -QSTRING_S * -trim_qs_from_cl(QSTRING_S *cl, QSTRING_S *nl, QSTRING_S *pl) -{ - QSTRING_S *cqstring = pl ? pl : nl; - QSTRING_S *tl = pl ? pl : nl; - int p, c; - - if (qstring_is_normal(tl)) - return tl; - - p = same_qstring(pl ? pl : cl, pl ? cl : nl); - - for (c = 1; c < p; c++, cl = cl->next, tl = tl->next); - - /* - * cl->next and tl->next differ, it may be because cl->next does not - * exist or tl->next does not exist or simply both exist but are - * different. In this last case, it may be that cl->next->value is made - * of spaces. If this is the case, tl advances once more. - */ - - if (tl->next){ - if (cl && cl->next && value_is_space(cl->next->value)) - tl = tl->next; - if (tl->next) - free_qs(&(tl->next)); - } - - if (!p) - free_qs(&cqstring); - - return cqstring; -} - -/* This function trims cl so that it returns a real quote string based - * on information gathered from the previous and next lines. pl and cl are - * also trimmed, but that is done in another function, not here. - */ -QSTRING_S * -fix_qstring(QSTRING_S *cl, QSTRING_S *nl, QSTRING_S *pl) -{ - QSTRING_S *cqstring = cl, *nqstring = nl, *pqstring = pl; - int c, n; - - if (qstring_is_normal(cl)) - return cl; - - c = count_levels_qstring(cl); - n = same_qstring(cl,nl); - - if (!n){ /* no next line or no agreement with next line */ - int p = same_qstring(pl, cl); /* number of agreements between pl and cl */ - QSTRING_S *tl; /* test line */ - - /* - * Here p <= c, so either p < c or p == c. If p == c, we are done, - * and return cl. If not, there are two cases, either p == 0 or - * 0 < p < c. In the first case, we do not have enough evidence - * to return anything other than the normal part of cl, in the second - * case we can only return p levels of cl. - */ - - if (p == c) - tl = cqstring; - else{ - if (p){ - for (c = 1; c < p; c++) - cl = cl->next; - free_qs(&(cl->next)); - tl = cqstring; - } - else{ - int done = 0; - QSTRING_S *al = cl; /* another line */ - /* - * Ok, we really don't have enough evidence to return anything, - * different from the normal part of cl, but it could be possible - * that we may want to accept the not-normal part, so we better - * make an extra test to determine what needs to be freed - */ - while (pl && cl && cl->qstype == pl->qstype - && !strucmp(cl->value, pl->value)){ - cl = cl->next; - pl = pl->next; - } - if (pl && cl && cl->qstype == pl->qstype - && strcmp_qs(pl->value, cl->value)) - cl = cl->next; /* next level differs only in spaces */ - while (!done){ - while (cl && cl->qstype == qsNormal) - cl = cl->next; - if (cl){ - if ((cl->qstype == qsString) - && (cl->value[strlen(cl->value) - 1] == '>')) - cl = cl->next; - else done++; - } - else done++; - } - if (al == cl){ - free_qs(&(cl)); - tl = cl; - } - else { - while (al && (al->next != cl)) - al = al->next; - cl = al; - if (cl && cl->next) - free_qs(&(cl->next)); - tl = cqstring; - } - } - } - return tl; - } - if (n + 1 < c){ /* if there are not enough agreements */ - int p = same_qstring(pl, cl); /* number of agreement between pl and cl */ - QSTRING_S *tl; /* test line */ - /* - * There's no way we can use cl in this case, but we can use - * part of cl, this is if pl does not have more agreements - * with cl. - */ - if (p == c) - tl = cqstring; - else{ - int m = p < n ? n : p; - for (c = 1; c < m; c++){ - pl = pl ? pl->next : (QSTRING_S *) NULL; - nl = nl ? nl->next : (QSTRING_S *) NULL; - cl = cl->next; - } - if (p == n && pl && pl->next && nl && nl->next - && ((cl->next->qstype == pl->next->qstype) - || (cl->next->qstype == nl->next->qstype)) - && (strcmp_qs(cl->next->value, pl->next->value) - || strcmp_qs(pl->next->value, cl->next->value) - || strcmp_qs(cl->next->value, nl->next->value) - || strcmp_qs(nl->next->value, cl->next->value))) - cl = cl->next; /* next level differs only in spaces */ - if (cl->next) - free_qs(&(cl->next)); - tl = cqstring; - } - return tl; - } - if (n + 1 == c){ - int p = same_qstring(pl, cl); - QSTRING_S *tl; /* test line */ - - /* - * p <= c, so p <= n+1, which means p < n + 1 or p == n + 1. - * If p < n + 1, then p <= n. - * so we have three possibilities: - * p == n + 1 or p == n or p < n. - * In the first case we copy p == n + 1 == c levels, in the second - * and third case we copy n levels, and check if we can copy the - * n + 1 == c level. - */ - if (p == n + 1) /* p == c, in the above sense of c */ - tl = cl; /* use cl, this is enough evidence */ - else{ - for (c = 1; c < n; c++) - cl = cl->next; - /* - * Here c == n, we only have one more level of cl, and at least one - * more level of nl - */ - if (cl->next->qstype == qsNormal) - cl = cl->next; - if (cl->next) - free_qs(&(cl->next)); - tl = cqstring; - } - return tl; - } - if (n == c) /* Yeah!!! */ - return cqstring; -} - -QSTRING_S * -fix_qstring_allowed(QSTRING_S *cl, QSTRING_S *nl, QSTRING_S *pl) -{ - if(!cl) - return (QSTRING_S *) NULL; - - if (qs_allowed(cl)) - cl->next = fix_qstring_allowed(cl->next, (nl ? nl->next : NULL), - (pl ? pl->next : NULL)); - else - if((nl && cl->qstype == nl->qstype) || (pl && cl->qstype == pl->qstype) - || (!nl && !pl)) - free_qs(&cl); - return cl; -} - -/* - * This function flattens the quote string returned to us by is_quote. A - * crash in this function implies a bug elsewhere. - */ -void -flatten_qstring(QSTRING_S *qs, char *buff, int bufflen) -{ - int i, j; - if(!buff || bufflen <= 0) - return; - - for (i = 0; qs; qs = qs->next) - for (j = 0; i < bufflen - 1 - && (qs->value[j]) && (buff[i++] = qs->value[j]); j++); - buff[i] = '\0'; -} - -extern int list_len; - - -int -double_check_qstr(char *q) -{ - if(!q || !*q) - return 0; - - return (*q == '#') ? 1 : 0; -} - -/* - * Given a string, we return the position where the function thinks that - * the quote string is over, if you are ever thinking of fixing something, - * you got to the right place. Memory freed by caller. Experience shows - * that it only makes sense to initialize memory when we need it, not at - * the start of this function. - */ -QSTRING_S * -is_quote (char **qs,char *word, int been_here) -{ - int i = 0, j, nxt, prev, finished = 0, offset; - unsigned char c; - QSTRING_S *qstring = (QSTRING_S *) NULL; - - if (word == NULL || word[0] == '\0') - return (QSTRING_S *) NULL; - - while (!finished){ - /* - * Before we apply our rules, let's advance past the quote string - * given by the user, this will avoid not recognition of the - * user's indent string and application of the arbitrary rules - * below. Notice that this step may bring bugs into this - * procedure, but these bugs will only appear if the indent string - * is really really strange and the text to be justified - * cooperates a lot too, so in general this will not be a problem. - * If you are concerned about this bug, simply remove the - * following lines after this comment and before the "switch" - * command below and use a more normal quote string!. - */ - for(j = 0; j < list_len; j++){ - if(!double_check_qstr(qs[j])){ - i += advance_quote_string(qs[j], word, i); - if (!word[i]) /* went too far? */ - return qs_add(qs, word, qsNormal, 0, i, 0, 0); - } - else - break; - } - - switch (c = (unsigned char) now(word,i)){ - case NBSP: - case TAB : - case ' ' : { QSTRING_S *nextqs, *d; - - for (; ISspace(word[i]); i++); /* FIX ME */ - nextqs = is_quote(qs,word+i, 1); - /* - * Merge qstring and nextqs, since this is an artificial - * separation, unless nextqs is of different type. - * What this means in practice is that if - * qs->qstype == qsNormal and qs->next != NULL, then - * qs->next->qstype != qsNormal. - * - * Can't use qs_add to merge because it could lead - * to an infinite loop (e.g a line "^ ^"). - */ - i += nextqs && nextqs->qstype == qsNormal - ? strlen(nextqs->value) : 0; - qstring = (QSTRING_S *) malloc (sizeof(QSTRING_S)); - memset (qstring, 0, sizeof(QSTRING_S)); - qstring->value = (char *) malloc((i+1)*sizeof(char)); - strncpy(qstring->value, word, i); - qstring->value[i] = '\0'; - qstring->qstype = qsNormal; - if(nextqs && nextqs->qstype == qsNormal){ - d = nextqs->next; - nextqs->next = NULL; - qstring->next = d; - free_qs(&nextqs); - } - else - qstring->next = nextqs; - - return qstring; - } - break; - case RPAREN: /* parenthesis ')' */ - if ((i != 0) || ((i == 0) && been_here)) - i++; - else - if (i == 0) - return qs_add(qs, word, qsChar, i, i, 1, 1); - else - finished++; - break; - - case ':': /* colon */ - case '~': nxt = next(word,i); - if ((is_tilde(c) && (nxt == '/')) - || (is_colon(c) && !is_cquote(nxt) - && !is_cword(nxt) && nxt != RPAREN)) - finished++; - else if (is_cquote(c) - || is_cquote(nxt) - || (c != '~' && nxt == RPAREN) - || (i != 0 && ISspace(nxt)) - || is_cquote(prev = before(word,i)) - || (ISspace(prev) && !is_tilde(c)) - || (is_tilde(c) && nxt != '/')) - i++; - else if (i == 0 && been_here) - return qs_add(qs, word, qsChar, i, i, 1, 1); - else - finished++; - break; - - case '<' : - case '=' : - case '-' : offset = is_cquote(nxt = next(word,i)) ? 2 - : (nxt == c && is_cquote(next(word,i+1))) ? 3 : -1; - - if (offset > 0) - return qs_add(qs, word, qsString, i, i, offset, 1); - else - finished++; - break; - - case '[' : - case '+' : /* accept +>, *> */ - case '*' : if (is_rarrow(nxt = next(word, i)) || /* stars */ - (ISspace(nxt) && is_rarrow(next(word,i+1)))) - i++; - else - finished++; - break; - - case '^' : - case '!' : - case '%' : if (next(word,i) != c) - return qs_add(qs, word, qsChar, i, i+1, 0, 1); - else - finished++; - break; - - case '_' : if(ISspace(next(word, i))) - return qs_add(qs, word, qsChar, i, i+1, 0, 1); - else - finished++; - break; - - case '#' : { QStrType qstype = qsChar; - if((nxt = next(word, i)) != c){ - if(isdigit((int) nxt)) - qstype = qsGdb; - else - if(word_is_prog(word)) - qstype = qsProg; - return qs_add(qs, word, qstype, i, i+1, 0, 1); - } - else - finished++; - break; - } - - default: - if (is_cquote(c)) - i++; - else if (is_cletter(c)){ - for (j = i; (is_cletter(nxt = next(word,j)) || is_cnumber(nxt)) - && !(ISspace(nxt));j++); - /* - * The whole reason why we are splitting the quote - * string is so that we will be able to accept quote - * strings that are strange in some way. Here we got to - * a point in which a quote string might exist, but it - * could be strange, so we need to create a "next" field - * for the quote string to warn us that something - * strange is coming. We need to confirm if this is a - * good choice later. For now we will let it pass. - */ - if (isaword(word,i,j) || isamailbox(word,i,j)){ - int offset; - QStrType qstype; - - offset = (is_cquote(c = next(word,j)) - || (c == RPAREN)) ? 2 - : ((ISspace(c) - && is_cquote(next(word,j+1))) ? 3 : -1); - - qstype = (is_cquote(c) || (c == RPAREN)) - ? (is_qsword(c) ? qsWord : qsString) - : ((ISspace(c) && is_cquote(next(word,j+1))) - ? (is_qsword(next(word,j+1)) - ? qsWord : qsString) - : qsString); - - /* - * qsWords are valid quote strings only when - * they are followed by text. - */ - if (offset > 0 && qstype == qsWord && - !allwd_after_qsword(now(word,j + offset))) - offset = -1; - - if (offset > 0) - return qs_add(qs, word, qstype, i, j, offset, 1); - } - finished++; - } - else{ - if(i > 0) - return qs_add(qs, word, qsNormal, 0, i, 0, 1); - else if(!forbidden(c)) - return qs_add(qs, word, qsChar, 0, 1, 0, 1); - else /* chao pescao */ - finished++; - } - break; - } /* End Switch */ - } /* End while */ - if (i > 0) - qstring = qs_add(qs, word, qsNormal, 0, i, 0, 0); - return qstring; -} - -int -isaword(char word[NSTRING], int i, int j) -{ - return i <= j && is_cletter(word[i]) ? - (i < j ? isaword(word,i+1,j) : 1) : 0; -} - -int -isamailbox(char word[NSTRING], int i, int j) -{ - return i <= j && (is_cletter(word[i]) || is_a_digit(word[i]) - || word[i] == '.') - ? (i < j ? isamailbox(word,i+1,j) : 1) : 0; -} - -/* - This routine removes the last part that is qsword or qschar that is not - followed by a normal part. This means that if a qsword or qschar is - followed by a qsnormal (or qsstring), we accept the qsword (or qschar) - as part of a quote string. - */ -QSTRING_S * -remove_qsword(QSTRING_S *cl) -{ - QSTRING_S *np = cl; - QSTRING_S *cp = np; /* this variable trails cl */ - - while(1){ - while (cl && cl->qstype == qsNormal) - cl = cl->next; - - if (cl){ - if (((cl->qstype == qsWord) || (cl->qstype == qsChar)) - && !exists_good_part(cl)){ - if (np == cl) /* qsword or qschar at the beginning */ - free_qs(&cp); - else{ - while (np->next != cl) - np = np->next; - free_qs(&(np->next)); - } - break; - } - else - cl = cl->next; - } - else - break; - } - return cp; -} - -int -exists_good_part (QSTRING_S *cl) -{ - return (cl ? (((cl->qstype != qsWord) && (cl->qstype != qsChar) - && qs_allowed(cl) && !value_is_space(cl->value)) - ? 1 : exists_good_part(cl->next)) - : 0); -} - -int -line_isblank(char **q, char *GLine, char *NLine, char *PLine, int buflen) -{ - int n = 0; - QSTRING_S *cl; - char qstr[NSTRING]; - - cl = do_raw_quote_match(q, GLine, NLine, PLine, NULL, NULL); - - flatten_qstring(cl, qstr, NSTRING); - - free_qs(&cl); - - for(n = strlen(qstr); n < buflen && GLine[n]; n++) - if(!ISspace((unsigned char) GLine[n])) - return(FALSE); - - return(TRUE); -} - -QSTRING_S * -do_raw_quote_match(char **q, char *GLine, char *NLine, char *PLine, QSTRING_S **nlp, QSTRING_S **plp) -{ - QSTRING_S *cl, *nl = NULL, *pl = NULL; - char nbuf[NSTRING], pbuf[NSTRING], buf[NSTRING]; - int emptypl = 0, emptynl = 0; - - if (!(cl = is_quote(q, GLine, 0))) /* if nothing in, nothing out */ - return cl; - - nl = is_quote(q, NLine, 0); /* Next Line */ - if (nlp) *nlp = nl; - pl = is_quote(q, PLine, 0); /* Previous Line */ - if (plp) *plp = pl; - /* - * If there's nothing in the preceeding or following line - * there is not enough information to accept it or discard it. In this - * case it's likely to be an isolated line, so we better accept it - * if it does not look like a word. - */ - flatten_qstring(pl, pbuf, NSTRING); - emptypl = (!PLine || !PLine[0] || - (pl && value_is_space(pbuf)) && !PLine[strlen(pbuf)]) ? 1 : 0; - if (emptypl){ - flatten_qstring(nl, nbuf, NSTRING); - emptynl = (!NLine || !NLine[0] || - (nl && value_is_space(nbuf) && !NLine[strlen(nbuf)])) ? 1 : 0; - if (emptynl){ - cl = remove_qsword(cl); - if((cl = fix_qstring_allowed(cl, NULL, NULL)) != NULL) - cl = qs_remove_trailing_spaces(cl); - free_qs(&nl); - free_qs(&pl); - if(nlp) *nlp = NULL; - if(plp) *plp = NULL; - - return cl; - } - } - - /* - * If either cl, nl or pl contain suspicious characters that may make - * them (or not) be quote strings, we need to fix them, so that the - * next pass will be done correctly. - */ - - cl = fix_qstring(cl, nl, pl); - nl = trim_qs_from_cl(cl, nl, NULL); - pl = trim_qs_from_cl(cl, NULL, pl); - if((cl = fix_qstring_allowed(cl, nl, pl)) != NULL){ - nl = trim_qs_from_cl(cl, nl, NULL); - pl = trim_qs_from_cl(cl, NULL, pl); - } - else{ - free_qs(&nl); - free_qs(&pl); - } - if(nlp) - *nlp = nl; - else - free_qs(&nl); - if(plp) - *plp = pl; - else - free_qs(&pl); - return cl; -} - -QSTRING_S * -do_quote_match(char **q, char *GLine, char *NLine, char *PLine, char *rqstr, -int rqstrlen, int plb) -{ - QSTRING_S *cl, *nl = NULL, *pl = NULL; - int c, n, p,i, j, NewP, NewC, NewN, clength, same = 0; - char nbuf[NSTRING], pbuf[NSTRING], buf[NSTRING]; - - if(rqstr) - *rqstr = '\0'; - - /* if nothing in, nothing out */ - cl = do_raw_quote_match(q, GLine, NLine, PLine, &nl, &pl); - if(cl == NULL){ - free_qs(&nl); - free_qs(&pl); - return cl; - } - - flatten_qstring(cl, rqstr, rqstrlen); - flatten_qstring(cl, buf, NSTRING); - flatten_qstring(nl, nbuf, NSTRING); - flatten_qstring(pl, pbuf, NSTRING); - - /* - * Once upon a time, is_quote used to return the length of the quote - * string that it had found. One day, not long ago, black hand came - * and changed all that, and made is_quote return a quote string - * divided in several fields, making the algorithm much more - * complicated. Fortunately black hand left a few comments in the - * source code to make it more understandable. Because of this change - * we need to compute the lengths of the quote strings separately - */ - c = buf && buf[0] ? strlen(buf) : 0; - n = nbuf && nbuf[0] ? strlen(nbuf) : 0; - p = pbuf && pbuf[0] ? strlen(pbuf) : 0; - /* - * When quote strings contain only blank spaces (ascii code 32) the - * above count is equal to the length of the quote string, but if - * there are TABS, the length of the quote string as seen by the user - * is different than the number that was just computed. Because of - * this we demand a recount (hmm.. unless you are in Florida, where - * recounts are forbidden) - */ - NewP = strlenis(pbuf); - NewC = strlenis(buf); - NewN = strlenis(nbuf); - - /* - * For paragraphs with spaces in the first line, but no space in the - * quote string of the second line, we make sure we choose the quote - * string without a space at the end of it. - */ - if ((NLine && !NLine[0]) - && ((PLine && !PLine[0]) - || (((same = same_qstring(pl, cl)) != 0) - && (same != count_levels_qstring(cl))))) - cl = qs_remove_trailing_spaces(cl); - else - if (NewC > NewN){ - int agree = 0; - for (j = 0; (j < n) && (GLine[j] == NLine[j]); j++); - clength = j; - /* clength is the common length in which Gline and Nline agree */ - /* j < n means that they do not agree fully */ - /* GLine = " \tText" - NLine = " Text" */ - if(j == n) - agree++; - if (clength < n){ /* see if buf and nbuf are padded with spaces and tabs */ - for (i = clength; i < n && ISspace(NLine[i]); i++); - if (i == n){/* padded NLine until the end of spaces? */ - for (i = clength; i < c && ISspace(GLine[i]); i++); - if (i == c) /* Padded CLine until the end of spaces? */ - agree++; - } - } - if (agree){ - for (j = clength; j < c && ISspace(GLine[j]); j++); - if (j == c){ - /* - * If we get here, it means that the current line has the same - * quote string (visually) than the next line, but both of them - * are padded with different amount of TABS or spaces at the end. - * The current line (GLine) has more spaces/TABs than the next - * line. This is the typical situation that is found at the - * begining of a paragraph. We need to check this, however, by - * checking the previous line. This avoids that we confuse - * ourselves with being in the last line of a paragraph. - * Example when it should not free_qs(cl) - * " Text in Paragraph 1" (PLine) - * " Text in Paragraph 1" (GLine) - * " Other Paragraph Number 2" (NLine) - * - * Example when it should free_qs(cl): - * ":) " (PLine) p = 3, j = 3 - * ":) Text" (GLine) c = 5 - * ":) More text" (NLine) n = 3 - * - * Example when it should free_qs(cl): - * ":) " (PLine) p = 3, j = 3 - * ":) > > > Text" (GLine) c = 11 - * ":) > > > More text" (NLine) n = 9 - * - * Example when it should free_qs(cl): - * ":) :) " (PLine) p = 6, j = 3 - * ":) > > > Text" (GLine) c = 11 - * ":) > > > More text" (NLine) n = 9 - * - * Example when it should free_qs(cl): - * ":) > > > " (PLine) p = 13, j = 11 - * ":) > > > Text" (GLine) c = 11 - * ":) > > > More text" (NLine) n = 9 - * - * The following example is very interesting. The "Other Text" - * line below should free the quote string an make it equal to the - * quote string of the line below it, but any algorithm trying - * to advance past that line should make it stop there, so - * we need one more check, to check the raw quote string and the - * processed quote string at the same time. - * FREE qs in this example. - * " Some Text" (PLine) p = 3, j = 0 - * "\tOther Text" (GLine) c = 1 - * " More Text" (NLine) n = 3 - * - */ - for (j = 0; (j < p) && (GLine[j] == PLine[j]); j++); - if ((p != c || j != p) && NLine[n]) - if(!get_indent_raw_line(q, PLine, nbuf, NSTRING, p, plb) - || NewP + strlenis(nbuf) != NewC){ - free_qs(&cl); - free_qs(&pl); - return nl; - } - } - } - } - - free_qs(&nl); - free_qs(&pl); - - return cl; -} - -/* - * Given a line, an initial position, and a quote string, we advance the - * current line past the quote string, including arbitraty spaces - * contained in the line, except that it removes trailing spaces. We do - * not handle TABs, if any, contained in the quote string. At least not - * yet. - * - * Arguments: q - quote string - * l - a line to process - * i - position in the line to start processing. i = 0 is the - * begining of that line. - */ -int -advance_quote_string(char *q, char l[NSTRING], int i) -{ - int n = 0, j = 0, is = 0, es = 0; - int k, m, p, adv; - char qs[NSTRING] = {'\0'}; - if(!q || !*q) - return(0); - for (p = strlen(q); (p > 0) && (q[p - 1] == ' '); p--, es++); - if (!p){ /* string contains only spaces */ - for (k = 0; ISspace(l[i + k]); k++); - k -= k % es; - return k; - } - for (is = 0; ISspace(q[is]); is++); /* count initial spaces */ - for (m = 0 ; is + m < p ; m++) - qs[m] = q[is + m]; /* qs = quote string without any space at the end */ - /* advance as many spaces as there are at the begining */ - for (k = 0; ISspace(l[i + j]); k++, j++); - /* now find the visible string in the line */ - for (m = 0; qs[m] && l[i + j] == qs[m]; m++, j++); - if (!qs[m]){ /* no match */ - /* - * So far we have advanced at least "is" spaces, plus the visible - * string "qs". Now we need to advance the trailing number of - * spaces "es". If we can do that, we have found the quote string. - */ - for (p = 0; ISspace(l[i + j + p]); p++); - adv = advance_quote_string(q, l, i + j + ((p < es) ? p : es)); - n = ((p < es) ? 0 : es) + k + m + adv; - } - return n; -} - -/* - * This function returns the effective length in screen of the quote - * string. If the string contains a TAB character, it is added here, if - * not, the length returned is the length of the string - */ -int strlenis(char *qstr) -{ - int i, rv = 0; - for (i = 0; qstr && qstr[i]; i++) - rv += ((qstr[i] == TAB) ? (~rv & 0x07) + 1 : 1); - return rv; -} - -int -is_indent (char word[NSTRING], int plb) -{ - int i = 0, finished = 0, c, nxt, j, k, digit = 0, bdigits = -1, alpha = 0; - - if (!word || !word[0]) - return i; - - for (i = 0, j = 0; ISspace(word[i]); i++, j++); - while ((i < NSTRING - 2) && !finished){ - switch (c = now(word,i)){ - case NBSP: - case TAB : - case ' ' : for (; ISspace(word[i]); i++); - if (!is_indent_char(now(word,i))) - finished++; - break; - - case '+' : - case '.' : - case ']' : - case '*' : - case '}' : - case '-' : - case RPAREN: - nxt = next(word,i); - if ((c == '.' && allowed_after_period(nxt) && alpha) - || (c == '*' && allowed_after_star(nxt)) - || (c == '}' && allowed_after_braces(nxt)) - || (c == '-' && allowed_after_dash(nxt)) - || (c == '+' && allowed_after_dash(nxt)) - || (c == RPAREN && allowed_after_parenth(nxt)) - || (c == ']' && allowed_after_parenth(nxt))) - i++; - else - finished++; - break; - - default : if (is_a_digit(c) && plb){ - if (bdigits < 0) - bdigits = i; /* first digit */ - for (k = i; is_a_digit(now(word,k)); k++); - if (k - bdigits > 2){ /* more than 2 digits? */ - i = bdigits; /* too many! */ - finished++; - } - else{ - if(allowed_after_digit(now(word,k),word,k)){ - alpha++; - i = k; - } - else{ - i = bdigits; - finished++; - } - } - } - else - finished++; - break; - - } - } - if (i == j) - i = 0; /* there must be something more than spaces in an indent string */ - return i; -} - -int -get_indent_raw_line(char **q, char *GLine, char *buf, int buflen, int k, int plb) -{ - int i, j; - char testline[1024]; - - if(k > 0){ - for(j = 0; GLine[j] != '\0'; j++){ - testline[j] = GLine[j]; - testline[j+1] = '\0'; - if(strlenis(testline) >= strlenis(buf)) - break; - } - k = ++j; /* reset k */ - } - i = is_indent(GLine+k, plb); - - for (j = 0; j < i && j < buflen && (buf[j] = GLine[j + k]); j++); - buf[j] = '\0'; - - return i; -} - -/* support for remembering quote strings across messages */ -char **allowed_qstr = NULL; -int list_len = 0; - -void -free_allowed_qstr(void) -{ - int i; - char **q = allowed_qstr; - - if(q == NULL) - return; - - for(i = 0; i < list_len; i++) - fs_give((void **)&q[i]); - - fs_give((void **)q); - list_len = 0; -} - -void -add_allowed_qstr(void *q, int type) -{ - int i; - - if(allowed_qstr == NULL){ - allowed_qstr = malloc(sizeof(char *)); - list_len = 0; - } - - if(type == 0){ - allowed_qstr[list_len] = malloc((1+strlen((char *)q))*sizeof(char)); - strcpy(allowed_qstr[list_len], (char *)q); - } - else - allowed_qstr[list_len] = (char *) ucs4_to_utf8_cpystr((UCS *)q); - - fs_resize((void **)&allowed_qstr, (++list_len + 1)*sizeof(char *)); - allowed_qstr[list_len] = NULL; -} - -void -record_quote_string (QSTRING_S *qs) -{ - int i, j, k; - - for(; qs && qs->value; qs = qs->next){ - j = 0; - for (; ;){ - k = j; - for(i = 0; i < list_len; i++){ - j += advance_quote_string(allowed_qstr[i], qs->value, j); - for(; ISspace(qs->value[j]); j++); - } - if(k == j) - break; - } - if(qs->value[j] != '\0') - add_allowed_qstr((void *)(qs->value + j), 0); - } -} - -/* type utf8: code 0; ucs4: code 1. */ -char ** -default_qstr(void *q, int type) -{ - if(allowed_qstr == NULL) - add_allowed_qstr(q, type); - - return allowed_qstr; -} - diff --git a/pith/osdep/color.h b/pith/osdep/color.h index 929389a2..32c86242 100644 --- a/pith/osdep/color.h +++ b/pith/osdep/color.h @@ -17,24 +17,6 @@ #ifndef PITH_OSDEP_COLOR_INCLUDED #define PITH_OSDEP_COLOR_INCLUDED -/* - * struct that will help us determine what the quote string of a line - * is. The "next" field indicates the presence of a possible continuation. - * The idea is that if a continuation fails, we free it and check for the - * remaining structure left - */ - -typedef enum {qsNormal, qsString, qsWord, qsChar, qsGdb, qsProg, qsText} QStrType; - -typedef struct QSTRING { - char *value; /* possible quote string */ - QStrType qstype; /* type of quote string */ - struct QSTRING *next; /* possible continuation */ -} QSTRING_S; - -#define UCH(c) ((unsigned char) (c)) -#define NBSP UCH('\240') -#define ISspace(c) (UCH(c) == ' ' || UCH(c) == TAB || UCH(c) == NBSP) #define RGBLEN 11 #define MAXCOLORLEN 11 /* longest string a color can be */ @@ -111,11 +93,6 @@ char *pico_get_last_fg_color(void); char *pico_get_last_bg_color(void); char *color_to_canonical_name(char *); int pico_count_in_color_table(void); -int is_indent(char *, int); -int get_indent_raw_line (char **, char *, char *, int, int, int); -int line_isblank(char **, char *, char *, char *, int); -int strlenis(char *); -int value_is_space(char *); -int advance_quote_string(char *, char *, int); + #endif /* PITH_OSDEP_COLOR_INCLUDED */ -- cgit v1.2.3-70-g09d2