/* vi:set ts=8 sts=4 sw=4 noet: * * VIM - Vi IMproved by Bram Moolenaar * * Do ":help uganda" in Vim to read copying and usage conditions. * Do ":help credits" in Vim to see a list of people who contributed. * See README.txt for an overview of the Vim source code. */ /* * eval.c: Expression evaluation. */ #define USING_FLOAT_STUFF #include "vim.h" #if defined(FEAT_EVAL) || defined(PROTO) #ifdef VMS # include #endif static char *e_dictrange = N_("E719: Cannot use [:] with a Dictionary"); #ifdef FEAT_FLOAT static char *e_float_as_string = N_("E806: using Float as a String"); #endif #define NAMESPACE_CHAR (char_u *)"abglstvw" /* * When recursively copying lists and dicts we need to remember which ones we * have done to avoid endless recursiveness. This unique ID is used for that. * The last bit is used for previous_funccal, ignored when comparing. */ static int current_copyID = 0; static int echo_attr = 0; // attributes used for ":echo" /* * Info used by a ":for" loop. */ typedef struct { int fi_semicolon; // TRUE if ending in '; var]' int fi_varcount; // nr of variables in the list listwatch_T fi_lw; // keep an eye on the item used. list_T *fi_list; // list being used int fi_bi; // index of blob blob_T *fi_blob; // blob being used } forinfo_T; static int tv_op(typval_T *tv1, typval_T *tv2, char_u *op); static int eval2(char_u **arg, typval_T *rettv, int evaluate); static int eval3(char_u **arg, typval_T *rettv, int evaluate); static int eval4(char_u **arg, typval_T *rettv, int evaluate); static int eval5(char_u **arg, typval_T *rettv, int evaluate); static int eval6(char_u **arg, typval_T *rettv, int evaluate, int want_string); static int eval7(char_u **arg, typval_T *rettv, int evaluate, int want_string); static int eval7_leader(typval_T *rettv, char_u *start_leader, char_u **end_leaderp); static int free_unref_items(int copyID); static char_u *make_expanded_name(char_u *in_start, char_u *expr_start, char_u *expr_end, char_u *in_end); static int tv_check_lock(typval_T *tv, char_u *name, int use_gettext); /* * Return "n1" divided by "n2", taking care of dividing by zero. */ varnumber_T num_divide(varnumber_T n1, varnumber_T n2) { varnumber_T result; if (n2 == 0) // give an error message? { if (n1 == 0) result = VARNUM_MIN; // similar to NaN else if (n1 < 0) result = -VARNUM_MAX; else result = VARNUM_MAX; } else result = n1 / n2; return result; } /* * Return "n1" modulus "n2", taking care of dividing by zero. */ varnumber_T num_modulus(varnumber_T n1, varnumber_T n2) { // Give an error when n2 is 0? return (n2 == 0) ? 0 : (n1 % n2); } #if defined(EBCDIC) || defined(PROTO) /* * Compare struct fst by function name. */ static int compare_func_name(const void *s1, const void *s2) { struct fst *p1 = (struct fst *)s1; struct fst *p2 = (struct fst *)s2; return STRCMP(p1->f_name, p2->f_name); } /* * Sort the function table by function name. * The sorting of the table above is ASCII dependent. * On machines using EBCDIC we have to sort it. */ static void sortFunctions(void) { int funcCnt = (int)(sizeof(functions) / sizeof(struct fst)) - 1; qsort(functions, (size_t)funcCnt, sizeof(struct fst), compare_func_name); } #endif /* * Initialize the global and v: variables. */ void eval_init(void) { evalvars_init(); func_init(); #ifdef EBCDIC /* * Sort the function table, to enable binary search. */ sortFunctions(); #endif } #if defined(EXITFREE) || defined(PROTO) void eval_clear(void) { evalvars_clear(); free_scriptnames(); // must come after evalvars_clear(). free_locales(); // autoloaded script names free_autoload_scriptnames(); // unreferenced lists and dicts (void)garbage_collect(FALSE); // functions not garbage collected free_all_functions(); } #endif /* * Top level evaluation function, returning a boolean. * Sets "error" to TRUE if there was an error. * Return TRUE or FALSE. */ int eval_to_bool( char_u *arg, int *error, char_u **nextcmd, int skip) // only parse, don't execute { typval_T tv; varnumber_T retval = FALSE; if (skip) ++emsg_skip; if (eval0(arg, &tv, nextcmd, !skip) == FAIL) *error = TRUE; else { *error = FALSE; if (!skip) { retval = (tv_get_number_chk(&tv, error) != 0); clear_tv(&tv); } } if (skip) --emsg_skip; return (int)retval; } /* * Call eval1() and give an error message if not done at a lower level. */ static int eval1_emsg(char_u **arg, typval_T *rettv, int evaluate) { char_u *start = *arg; int ret; int did_emsg_before = did_emsg; int called_emsg_before = called_emsg; ret = eval1(arg, rettv, evaluate); if (ret == FAIL) { // Report the invalid expression unless the expression evaluation has // been cancelled due to an aborting error, an interrupt, or an // exception, or we already gave a more specific error. // Also check called_emsg for when using assert_fails(). if (!aborting() && did_emsg == did_emsg_before && called_emsg == called_emsg_before) semsg(_(e_invexpr2), start); } return ret; } /* * Evaluate an expression, which can be a function, partial or string. * Pass arguments "argv[argc]". * Return the result in "rettv" and OK or FAIL. */ int eval_expr_typval(typval_T *expr, typval_T *argv, int argc, typval_T *rettv) { char_u *s; char_u buf[NUMBUFLEN]; funcexe_T funcexe; if (expr->v_type == VAR_FUNC) { s = expr->vval.v_string; if (s == NULL || *s == NUL) return FAIL; vim_memset(&funcexe, 0, sizeof(funcexe)); funcexe.evaluate = TRUE; if (call_func(s, -1, rettv, argc, argv, &funcexe) == FAIL) return FAIL; } else if (expr->v_type == VAR_PARTIAL) { partial_T *partial = expr->vval.v_partial; if (partial->pt_func != NULL && partial->pt_func->uf_dfunc_idx >= 0) { if (call_def_function(partial->pt_func, argc, argv, rettv) == FAIL) return FAIL; } else { s = partial_name(partial); if (s == NULL || *s == NUL) return FAIL; vim_memset(&funcexe, 0, sizeof(funcexe)); funcexe.evaluate = TRUE; funcexe.partial = partial; if (call_func(s, -1, rettv, argc, argv, &funcexe) == FAIL) return FAIL; } } else { s = tv_get_string_buf_chk(expr, buf); if (s == NULL) return FAIL; s = skipwhite(s); if (eval1_emsg(&s, rettv, TRUE) == FAIL) return FAIL; if (*s != NUL) // check for trailing chars after expr { clear_tv(rettv); semsg(_(e_invexpr2), s); return FAIL; } } return OK; } /* * Like eval_to_bool() but using a typval_T instead of a string. * Works for string, funcref and partial. */ int eval_expr_to_bool(typval_T *expr, int *error) { typval_T rettv; int res; if (eval_expr_typval(expr, NULL, 0, &rettv) == FAIL) { *error = TRUE; return FALSE; } res = (tv_get_number_chk(&rettv, error) != 0); clear_tv(&rettv); return res; } /* * Top level evaluation function, returning a string. If "skip" is TRUE, * only parsing to "nextcmd" is done, without reporting errors. Return * pointer to allocated memory, or NULL for failure or when "skip" is TRUE. */ char_u * eval_to_string_skip( char_u *arg, char_u **nextcmd, int skip) // only parse, don't execute { typval_T tv; char_u *retval; if (skip) ++emsg_skip; if (eval0(arg, &tv, nextcmd, !skip) == FAIL || skip) retval = NULL; else { retval = vim_strsave(tv_get_string(&tv)); clear_tv(&tv); } if (skip) --emsg_skip; return retval; } /* * Skip over an expression at "*pp". * Return FAIL for an error, OK otherwise. */ int skip_expr(char_u **pp) { typval_T rettv; *pp = skipwhite(*pp); return eval1(pp, &rettv, FALSE); } /* * Top level evaluation function, returning a string. * When "convert" is TRUE convert a List into a sequence of lines and convert * a Float to a String. * Return pointer to allocated memory, or NULL for failure. */ char_u * eval_to_string( char_u *arg, char_u **nextcmd, int convert) { typval_T tv; char_u *retval; garray_T ga; #ifdef FEAT_FLOAT char_u numbuf[NUMBUFLEN]; #endif if (eval0(arg, &tv, nextcmd, TRUE) == FAIL) retval = NULL; else { if (convert && tv.v_type == VAR_LIST) { ga_init2(&ga, (int)sizeof(char), 80); if (tv.vval.v_list != NULL) { list_join(&ga, tv.vval.v_list, (char_u *)"\n", TRUE, FALSE, 0); if (tv.vval.v_list->lv_len > 0) ga_append(&ga, NL); } ga_append(&ga, NUL); retval = (char_u *)ga.ga_data; } #ifdef FEAT_FLOAT else if (convert && tv.v_type == VAR_FLOAT) { vim_snprintf((char *)numbuf, NUMBUFLEN, "%g", tv.vval.v_float); retval = vim_strsave(numbuf); } #endif else retval = vim_strsave(tv_get_string(&tv)); clear_tv(&tv); } return retval; } /* * Call eval_to_string() without using current local variables and using * textlock. When "use_sandbox" is TRUE use the sandbox. */ char_u * eval_to_string_safe( char_u *arg, char_u **nextcmd, int use_sandbox) { char_u *retval; funccal_entry_T funccal_entry; save_funccal(&funccal_entry); if (use_sandbox) ++sandbox; ++textlock; retval = eval_to_string(arg, nextcmd, FALSE); if (use_sandbox) --sandbox; --textlock; restore_funccal(); return retval; } /* * Top level evaluation function, returning a number. * Evaluates "expr" silently. * Returns -1 for an error. */ varnumber_T eval_to_number(char_u *expr) { typval_T rettv; varnumber_T retval; char_u *p = skipwhite(expr); ++emsg_off; if (eval1(&p, &rettv, TRUE) == FAIL) retval = -1; else { retval = tv_get_number_chk(&rettv, NULL); clear_tv(&rettv); } --emsg_off; return retval; } /* * Top level evaluation function. * Returns an allocated typval_T with the result. * Returns NULL when there is an error. */ typval_T * eval_expr(char_u *arg, char_u **nextcmd) { typval_T *tv; tv = ALLOC_ONE(typval_T); if (tv != NULL && eval0(arg, tv, nextcmd, TRUE) == FAIL) VIM_CLEAR(tv); return tv; } /* * Call some Vim script function and return the result in "*rettv". * Uses argv[0] to argv[argc - 1] for the function arguments. argv[argc] * should have type VAR_UNKNOWN. * Returns OK or FAIL. */ int call_vim_function( char_u *func, int argc, typval_T *argv, typval_T *rettv) { int ret; funcexe_T funcexe; rettv->v_type = VAR_UNKNOWN; // clear_tv() uses this vim_memset(&funcexe, 0, sizeof(funcexe)); funcexe.firstline = curwin->w_cursor.lnum; funcexe.lastline = curwin->w_cursor.lnum; funcexe.evaluate = TRUE; ret = call_func(func, -1, rettv, argc, argv, &funcexe); if (ret == FAIL) clear_tv(rettv); return ret; } /* * Call Vim script function "func" and return the result as a number. * Returns -1 when calling the function fails. * Uses argv[0] to argv[argc - 1] for the function arguments. argv[argc] should * have type VAR_UNKNOWN. */ varnumber_T call_func_retnr( char_u *func, int argc, typval_T *argv) { typval_T rettv; varnumber_T retval; if (call_vim_function(func, argc, argv, &rettv) == FAIL) return -1; retval = tv_get_number_chk(&rettv, NULL); clear_tv(&rettv); return retval; } /* * Call Vim script function "func" and return the result as a string. * Returns NULL when calling the function fails. * Uses argv[0] to argv[argc - 1] for the function arguments. argv[argc] should * have type VAR_UNKNOWN. */ void * call_func_retstr( char_u *func, int argc, typval_T *argv) { typval_T rettv; char_u *retval; if (call_vim_function(func, argc, argv, &rettv) == FAIL) return NULL; retval = vim_strsave(tv_get_string(&rettv)); clear_tv(&rettv); return retval; } /* * Call Vim script function "func" and return the result as a List. * Uses argv[0] to argv[argc - 1] for the function arguments. argv[argc] should * have type VAR_UNKNOWN. * Returns NULL when there is something wrong. */ void * call_func_retlist( char_u *func, int argc, typval_T *argv) { typval_T rettv; if (call_vim_function(func, argc, argv, &rettv) == FAIL) return NULL; if (rettv.v_type != VAR_LIST) { clear_tv(&rettv); return NULL; } return rettv.vval.v_list; } #ifdef FEAT_FOLDING /* * Evaluate 'foldexpr'. Returns the foldlevel, and any character preceding * it in "*cp". Doesn't give error messages. */ int eval_foldexpr(char_u *arg, int *cp) { typval_T tv; varnumber_T retval; char_u *s; int use_sandbox = was_set_insecurely((char_u *)"foldexpr", OPT_LOCAL); ++emsg_off; if (use_sandbox) ++sandbox; ++textlock; *cp = NUL; if (eval0(arg, &tv, NULL, TRUE) == FAIL) retval = 0; else { // If the result is a number, just return the number. if (tv.v_type == VAR_NUMBER) retval = tv.vval.v_number; else if (tv.v_type != VAR_STRING || tv.vval.v_string == NULL) retval = 0; else { // If the result is a string, check if there is a non-digit before // the number. s = tv.vval.v_string; if (!VIM_ISDIGIT(*s) && *s != '-') *cp = *s++; retval = atol((char *)s); } clear_tv(&tv); } --emsg_off; if (use_sandbox) --sandbox; --textlock; return (int)retval; } #endif /* * Get an lval: variable, Dict item or List item that can be assigned a value * to: "name", "na{me}", "name[expr]", "name[expr:expr]", "name[expr][expr]", * "name.key", "name.key[expr]" etc. * Indexing only works if "name" is an existing List or Dictionary. * "name" points to the start of the name. * If "rettv" is not NULL it points to the value to be assigned. * "unlet" is TRUE for ":unlet": slightly different behavior when something is * wrong; must end in space or cmd separator. * * flags: * GLV_QUIET: do not give error messages * GLV_READ_ONLY: will not change the variable * GLV_NO_AUTOLOAD: do not use script autoloading * * Returns a pointer to just after the name, including indexes. * When an evaluation error occurs "lp->ll_name" is NULL; * Returns NULL for a parsing error. Still need to free items in "lp"! */ char_u * get_lval( char_u *name, typval_T *rettv, lval_T *lp, int unlet, int skip, int flags, // GLV_ values int fne_flags) // flags for find_name_end() { char_u *p; char_u *expr_start, *expr_end; int cc; dictitem_T *v; typval_T var1; typval_T var2; int empty1 = FALSE; listitem_T *ni; char_u *key = NULL; int len; hashtab_T *ht; int quiet = flags & GLV_QUIET; // Clear everything in "lp". vim_memset(lp, 0, sizeof(lval_T)); if (skip) { // When skipping just find the end of the name. lp->ll_name = name; return find_name_end(name, NULL, NULL, FNE_INCL_BR | fne_flags); } // Find the end of the name. p = find_name_end(name, &expr_start, &expr_end, fne_flags); lp->ll_name_end = p; if (expr_start != NULL) { // Don't expand the name when we already know there is an error. if (unlet && !VIM_ISWHITE(*p) && !ends_excmd(*p) && *p != '[' && *p != '.') { emsg(_(e_trailing)); return NULL; } lp->ll_exp_name = make_expanded_name(name, expr_start, expr_end, p); if (lp->ll_exp_name == NULL) { // Report an invalid expression in braces, unless the // expression evaluation has been cancelled due to an // aborting error, an interrupt, or an exception. if (!aborting() && !quiet) { emsg_severe = TRUE; semsg(_(e_invarg2), name); return NULL; } } lp->ll_name = lp->ll_exp_name; } else { lp->ll_name = name; if (current_sctx.sc_version == SCRIPT_VERSION_VIM9 && *p == ':') { scriptitem_T *si = SCRIPT_ITEM(current_sctx.sc_sid); char_u *tp = skipwhite(p + 1); // parse the type after the name lp->ll_type = parse_type(&tp, &si->sn_type_list); lp->ll_name_end = tp; } } // Without [idx] or .key we are done. if ((*p != '[' && *p != '.') || lp->ll_name == NULL) return p; cc = *p; *p = NUL; // Only pass &ht when we would write to the variable, it prevents autoload // as well. v = find_var(lp->ll_name, (flags & GLV_READ_ONLY) ? NULL : &ht, flags & GLV_NO_AUTOLOAD); if (v == NULL && !quiet) semsg(_(e_undefvar), lp->ll_name); *p = cc; if (v == NULL) return NULL; /* * Loop until no more [idx] or .key is following. */ lp->ll_tv = &v->di_tv; var1.v_type = VAR_UNKNOWN; var2.v_type = VAR_UNKNOWN; while (*p == '[' || (*p == '.' && lp->ll_tv->v_type == VAR_DICT)) { if (!(lp->ll_tv->v_type == VAR_LIST && lp->ll_tv->vval.v_list != NULL) && !(lp->ll_tv->v_type == VAR_DICT && lp->ll_tv->vval.v_dict != NULL) && !(lp->ll_tv->v_type == VAR_BLOB && lp->ll_tv->vval.v_blob != NULL)) { if (!quiet) emsg(_("E689: Can only index a List, Dictionary or Blob")); return NULL; } if (lp->ll_range) { if (!quiet) emsg(_("E708: [:] must come last")); return NULL; } len = -1; if (*p == '.') { key = p + 1; for (len = 0; ASCII_ISALNUM(key[len]) || key[len] == '_'; ++len) ; if (len == 0) { if (!quiet) emsg(_(e_emptykey)); return NULL; } p = key + len; } else { // Get the index [expr] or the first index [expr: ]. p = skipwhite(p + 1); if (*p == ':') empty1 = TRUE; else { empty1 = FALSE; if (eval1(&p, &var1, TRUE) == FAIL) // recursive! return NULL; if (tv_get_string_chk(&var1) == NULL) { // not a number or string clear_tv(&var1); return NULL; } } // Optionally get the second index [ :expr]. if (*p == ':') { if (lp->ll_tv->v_type == VAR_DICT) { if (!quiet) emsg(_(e_dictrange)); clear_tv(&var1); return NULL; } if (rettv != NULL && !(rettv->v_type == VAR_LIST && rettv->vval.v_list != NULL) && !(rettv->v_type == VAR_BLOB && rettv->vval.v_blob != NULL)) { if (!quiet) emsg(_("E709: [:] requires a List or Blob value")); clear_tv(&var1); return NULL; } p = skipwhite(p + 1); if (*p == ']') lp->ll_empty2 = TRUE; else { lp->ll_empty2 = FALSE; if (eval1(&p, &var2, TRUE) == FAIL) // recursive! { clear_tv(&var1); return NULL; } if (tv_get_string_chk(&var2) == NULL) { // not a number or string clear_tv(&var1); clear_tv(&var2); return NULL; } } lp->ll_range = TRUE; } else lp->ll_range = FALSE; if (*p != ']') { if (!quiet) emsg(_(e_missbrac)); clear_tv(&var1); clear_tv(&var2); return NULL; } // Skip to past ']'. ++p; } if (lp->ll_tv->v_type == VAR_DICT) { if (len == -1) { // "[key]": get key from "var1" key = tv_get_string_chk(&var1); // is number or string if (key == NULL) { clear_tv(&var1); return NULL; } } lp->ll_list = NULL; lp->ll_dict = lp->ll_tv->vval.v_dict; lp->ll_di = dict_find(lp->ll_dict, key, len); // When assigning to a scope dictionary check that a function and // variable name is valid (only variable name unless it is l: or // g: dictionary). Disallow overwriting a builtin function. if (rettv != NULL && lp->ll_dict->dv_scope != 0) { int prevval; int wrong; if (len != -1) { prevval = key[len]; key[len] = NUL; } else prevval = 0; // avoid compiler warning wrong = (lp->ll_dict->dv_scope == VAR_DEF_SCOPE && rettv->v_type == VAR_FUNC && var_check_func_name(key, lp->ll_di == NULL)) || !valid_varname(key); if (len != -1) key[len] = prevval; if (wrong) return NULL; } if (lp->ll_di == NULL) { // Can't add "v:" or "a:" variable. if (lp->ll_dict == get_vimvar_dict() || &lp->ll_dict->dv_hashtab == get_funccal_args_ht()) { semsg(_(e_illvar), name); clear_tv(&var1); return NULL; } // Key does not exist in dict: may need to add it. if (*p == '[' || *p == '.' || unlet) { if (!quiet) semsg(_(e_dictkey), key); clear_tv(&var1); return NULL; } if (len == -1) lp->ll_newkey = vim_strsave(key); else lp->ll_newkey = vim_strnsave(key, len); clear_tv(&var1); if (lp->ll_newkey == NULL) p = NULL; break; } // existing variable, need to check if it can be changed else if ((flags & GLV_READ_ONLY) == 0 && var_check_ro(lp->ll_di->di_flags, name, FALSE)) { clear_tv(&var1); return NULL; } clear_tv(&var1); lp->ll_tv = &lp->ll_di->di_tv; } else if (lp->ll_tv->v_type == VAR_BLOB) { long bloblen = blob_len(lp->ll_tv->vval.v_blob); /* * Get the number and item for the only or first index of the List. */ if (empty1) lp->ll_n1 = 0; else // is number or string lp->ll_n1 = (long)tv_get_number(&var1); clear_tv(&var1); if (lp->ll_n1 < 0 || lp->ll_n1 > bloblen || (lp->ll_range && lp->ll_n1 == bloblen)) { if (!quiet) semsg(_(e_blobidx), lp->ll_n1); clear_tv(&var2); return NULL; } if (lp->ll_range && !lp->ll_empty2) { lp->ll_n2 = (long)tv_get_number(&var2); clear_tv(&var2); if (lp->ll_n2 < 0 || lp->ll_n2 >= bloblen || lp->ll_n2 < lp->ll_n1) { if (!quiet) semsg(_(e_blobidx), lp->ll_n2); return NULL; } } lp->ll_blob = lp->ll_tv->vval.v_blob; lp->ll_tv = NULL; break; } else { /* * Get the number and item for the only or first index of the List. */ if (empty1) lp->ll_n1 = 0; else // is number or string lp->ll_n1 = (long)tv_get_number(&var1); clear_tv(&var1); lp->ll_dict = NULL; lp->ll_list = lp->ll_tv->vval.v_list; lp->ll_li = list_find(lp->ll_list, lp->ll_n1); if (lp->ll_li == NULL) { if (lp->ll_n1 < 0) { lp->ll_n1 = 0; lp->ll_li = list_find(lp->ll_list, lp->ll_n1); } } if (lp->ll_li == NULL) { clear_tv(&var2); if (!quiet) semsg(_(e_listidx), lp->ll_n1); return NULL; } /* * May need to find the item or absolute index for the second * index of a range. * When no index given: "lp->ll_empty2" is TRUE. * Otherwise "lp->ll_n2" is set to the second index. */ if (lp->ll_range && !lp->ll_empty2) { lp->ll_n2 = (long)tv_get_number(&var2); // is number or string clear_tv(&var2); if (lp->ll_n2 < 0) { ni = list_find(lp->ll_list, lp->ll_n2); if (ni == NULL) { if (!quiet) semsg(_(e_listidx), lp->ll_n2); return NULL; } lp->ll_n2 = list_idx_of_item(lp->ll_list, ni); } // Check that lp->ll_n2 isn't before lp->ll_n1. if (lp->ll_n1 < 0) lp->ll_n1 = list_idx_of_item(lp->ll_list, lp->ll_li); if (lp->ll_n2 < lp->ll_n1) { if (!quiet) semsg(_(e_listidx), lp->ll_n2); return NULL; } } lp->ll_tv = &lp->ll_li->li_tv; } } clear_tv(&var1); lp->ll_name_end = p; return p; } /* * Clear lval "lp" that was filled by get_lval(). */ void clear_lval(lval_T *lp) { vim_free(lp->ll_exp_name); vim_free(lp->ll_newkey); } /* * Set a variable that was parsed by get_lval() to "rettv". * "endp" points to just after the parsed name. * "op" is NULL, "+" for "+=", "-" for "-=", "*" for "*=", "/" for "/=", * "%" for "%=", "." for ".=" or "=" for "=". */ void set_var_lval( lval_T *lp, char_u *endp, typval_T *rettv, int copy, int flags, // LET_IS_CONST and/or LET_NO_COMMAND char_u *op) { int cc; listitem_T *ri; dictitem_T *di; if (lp->ll_tv == NULL) { cc = *endp; *endp = NUL; if (lp->ll_blob != NULL) { int error = FALSE, val; if (op != NULL && *op != '=') { semsg(_(e_letwrong), op); return; } if (lp->ll_range && rettv->v_type == VAR_BLOB) { int il, ir; if (lp->ll_empty2) lp->ll_n2 = blob_len(lp->ll_blob) - 1; if (lp->ll_n2 - lp->ll_n1 + 1 != blob_len(rettv->vval.v_blob)) { emsg(_("E972: Blob value does not have the right number of bytes")); return; } if (lp->ll_empty2) lp->ll_n2 = blob_len(lp->ll_blob); ir = 0; for (il = lp->ll_n1; il <= lp->ll_n2; il++) blob_set(lp->ll_blob, il, blob_get(rettv->vval.v_blob, ir++)); } else { val = (int)tv_get_number_chk(rettv, &error); if (!error) { garray_T *gap = &lp->ll_blob->bv_ga; // Allow for appending a byte. Setting a byte beyond // the end is an error otherwise. if (lp->ll_n1 < gap->ga_len || (lp->ll_n1 == gap->ga_len && ga_grow(&lp->ll_blob->bv_ga, 1) == OK)) { blob_set(lp->ll_blob, lp->ll_n1, val); if (lp->ll_n1 == gap->ga_len) ++gap->ga_len; } // error for invalid range was already given in get_lval() } } } else if (op != NULL && *op != '=') { typval_T tv; if (flags & LET_IS_CONST) { emsg(_(e_cannot_mod)); *endp = cc; return; } // handle +=, -=, *=, /=, %= and .= di = NULL; if (get_var_tv(lp->ll_name, (int)STRLEN(lp->ll_name), &tv, &di, TRUE, FALSE) == OK) { if ((di == NULL || (!var_check_ro(di->di_flags, lp->ll_name, FALSE) && !tv_check_lock(&di->di_tv, lp->ll_name, FALSE))) && tv_op(&tv, rettv, op) == OK) set_var(lp->ll_name, &tv, FALSE); clear_tv(&tv); } } else set_var_const(lp->ll_name, lp->ll_type, rettv, copy, flags); *endp = cc; } else if (var_check_lock(lp->ll_newkey == NULL ? lp->ll_tv->v_lock : lp->ll_tv->vval.v_dict->dv_lock, lp->ll_name, FALSE)) ; else if (lp->ll_range) { listitem_T *ll_li = lp->ll_li; int ll_n1 = lp->ll_n1; if (flags & LET_IS_CONST) { emsg(_("E996: Cannot lock a range")); return; } /* * Check whether any of the list items is locked */ for (ri = rettv->vval.v_list->lv_first; ri != NULL && ll_li != NULL; ) { if (var_check_lock(ll_li->li_tv.v_lock, lp->ll_name, FALSE)) return; ri = ri->li_next; if (ri == NULL || (!lp->ll_empty2 && lp->ll_n2 == ll_n1)) break; ll_li = ll_li->li_next; ++ll_n1; } /* * Assign the List values to the list items. */ for (ri = rettv->vval.v_list->lv_first; ri != NULL; ) { if (op != NULL && *op != '=') tv_op(&lp->ll_li->li_tv, &ri->li_tv, op); else { clear_tv(&lp->ll_li->li_tv); copy_tv(&ri->li_tv, &lp->ll_li->li_tv); } ri = ri->li_next; if (ri == NULL || (!lp->ll_empty2 && lp->ll_n2 == lp->ll_n1)) break; if (lp->ll_li->li_next == NULL) { // Need to add an empty item. if (list_append_number(lp->ll_list, 0) == FAIL) { ri = NULL; break; } } lp->ll_li = lp->ll_li->li_next; ++lp->ll_n1; } if (ri != NULL) emsg(_("E710: List value has more items than target")); else if (lp->ll_empty2 ? (lp->ll_li != NULL && lp->ll_li->li_next != NULL) : lp->ll_n1 != lp->ll_n2) emsg(_("E711: List value has not enough items")); } else { /* * Assign to a List or Dictionary item. */ if (flags & LET_IS_CONST) { emsg(_("E996: Cannot lock a list or dict")); return; } if (lp->ll_newkey != NULL) { if (op != NULL && *op != '=') { semsg(_(e_letwrong), op); return; } // Need to add an item to the Dictionary. di = dictitem_alloc(lp->ll_newkey); if (di == NULL) return; if (dict_add(lp->ll_tv->vval.v_dict, di) == FAIL) { vim_free(di); return; } lp->ll_tv = &di->di_tv; } else if (op != NULL && *op != '=') { tv_op(lp->ll_tv, rettv, op); return; } else clear_tv(lp->ll_tv); /* * Assign the value to the variable or list item. */ if (copy) copy_tv(rettv, lp->ll_tv); else { *lp->ll_tv = *rettv; lp->ll_tv->v_lock = 0; init_tv(rettv); } } } /* * Handle "tv1 += tv2", "tv1 -= tv2", "tv1 *= tv2", "tv1 /= tv2", "tv1 %= tv2" * and "tv1 .= tv2" * Returns OK or FAIL. */ static int tv_op(typval_T *tv1, typval_T *tv2, char_u *op) { varnumber_T n; char_u numbuf[NUMBUFLEN]; char_u *s; // Can't do anything with a Funcref, Dict, v:true on the right. if (tv2->v_type != VAR_FUNC && tv2->v_type != VAR_DICT && tv2->v_type != VAR_BOOL && tv2->v_type != VAR_SPECIAL) { switch (tv1->v_type) { case VAR_UNKNOWN: case VAR_VOID: case VAR_DICT: case VAR_FUNC: case VAR_PARTIAL: case VAR_BOOL: case VAR_SPECIAL: case VAR_JOB: case VAR_CHANNEL: break; case VAR_BLOB: if (*op != '+' || tv2->v_type != VAR_BLOB) break; // BLOB += BLOB if (tv1->vval.v_blob != NULL && tv2->vval.v_blob != NULL) { blob_T *b1 = tv1->vval.v_blob; blob_T *b2 = tv2->vval.v_blob; int i, len = blob_len(b2); for (i = 0; i < len; i++) ga_append(&b1->bv_ga, blob_get(b2, i)); } return OK; case VAR_LIST: if (*op != '+' || tv2->v_type != VAR_LIST) break; // List += List if (tv1->vval.v_list != NULL && tv2->vval.v_list != NULL) list_extend(tv1->vval.v_list, tv2->vval.v_list, NULL); return OK; case VAR_NUMBER: case VAR_STRING: if (tv2->v_type == VAR_LIST) break; if (vim_strchr((char_u *)"+-*/%", *op) != NULL) { // nr += nr , nr -= nr , nr *=nr , nr /= nr , nr %= nr n = tv_get_number(tv1); #ifdef FEAT_FLOAT if (tv2->v_type == VAR_FLOAT) { float_T f = n; if (*op == '%') break; switch (*op) { case '+': f += tv2->vval.v_float; break; case '-': f -= tv2->vval.v_float; break; case '*': f *= tv2->vval.v_float; break; case '/': f /= tv2->vval.v_float; break; } clear_tv(tv1); tv1->v_type = VAR_FLOAT; tv1->vval.v_float = f; } else #endif { switch (*op) { case '+': n += tv_get_number(tv2); break; case '-': n -= tv_get_number(tv2); break; case '*': n *= tv_get_number(tv2); break; case '/': n = num_divide(n, tv_get_number(tv2)); break; case '%': n = num_modulus(n, tv_get_number(tv2)); break; } clear_tv(tv1); tv1->v_type = VAR_NUMBER; tv1->vval.v_number = n; } } else { if (tv2->v_type == VAR_FLOAT) break; // str .= str s = tv_get_string(tv1); s = concat_str(s, tv_get_string_buf(tv2, numbuf)); clear_tv(tv1); tv1->v_type = VAR_STRING; tv1->vval.v_string = s; } return OK; case VAR_FLOAT: #ifdef FEAT_FLOAT { float_T f; if (*op == '%' || *op == '.' || (tv2->v_type != VAR_FLOAT && tv2->v_type != VAR_NUMBER && tv2->v_type != VAR_STRING)) break; if (tv2->v_type == VAR_FLOAT) f = tv2->vval.v_float; else f = tv_get_number(tv2); switch (*op) { case '+': tv1->vval.v_float += f; break; case '-': tv1->vval.v_float -= f; break; case '*': tv1->vval.v_float *= f; break; case '/': tv1->vval.v_float /= f; break; } } #endif return OK; } } semsg(_(e_letwrong), op); return FAIL; } /* * Evaluate the expression used in a ":for var in expr" command. * "arg" points to "var". * Set "*errp" to TRUE for an error, FALSE otherwise; * Return a pointer that holds the info. Null when there is an error. */ void * eval_for_line( char_u *arg, int *errp, char_u **nextcmdp, int skip) { forinfo_T *fi; char_u *expr; typval_T tv; list_T *l; *errp = TRUE; // default: there is an error fi = ALLOC_CLEAR_ONE(forinfo_T); if (fi == NULL) return NULL; expr = skip_var_list(arg, TRUE, &fi->fi_varcount, &fi->fi_semicolon); if (expr == NULL) return fi; expr = skipwhite(expr); if (expr[0] != 'i' || expr[1] != 'n' || !VIM_ISWHITE(expr[2])) { emsg(_(e_missing_in)); return fi; } if (skip) ++emsg_skip; if (eval0(skipwhite(expr + 2), &tv, nextcmdp, !skip) == OK) { *errp = FALSE; if (!skip) { if (tv.v_type == VAR_LIST) { l = tv.vval.v_list; if (l == NULL) { // a null list is like an empty list: do nothing clear_tv(&tv); } else { // Need a real list here. range_list_materialize(l); // No need to increment the refcount, it's already set for // the list being used in "tv". fi->fi_list = l; list_add_watch(l, &fi->fi_lw); fi->fi_lw.lw_item = l->lv_first; } } else if (tv.v_type == VAR_BLOB) { fi->fi_bi = 0; if (tv.vval.v_blob != NULL) { typval_T btv; // Make a copy, so that the iteration still works when the // blob is changed. blob_copy(tv.vval.v_blob, &btv); fi->fi_blob = btv.vval.v_blob; } clear_tv(&tv); } else { emsg(_(e_listreq)); clear_tv(&tv); } } } if (skip) --emsg_skip; return fi; } /* * Use the first item in a ":for" list. Advance to the next. * Assign the values to the variable (list). "arg" points to the first one. * Return TRUE when a valid item was found, FALSE when at end of list or * something wrong. */ int next_for_item(void *fi_void, char_u *arg) { forinfo_T *fi = (forinfo_T *)fi_void; int result; int flag = current_sctx.sc_version == SCRIPT_VERSION_VIM9 ? LET_NO_COMMAND : 0; listitem_T *item; if (fi->fi_blob != NULL) { typval_T tv; if (fi->fi_bi >= blob_len(fi->fi_blob)) return FALSE; tv.v_type = VAR_NUMBER; tv.v_lock = VAR_FIXED; tv.vval.v_number = blob_get(fi->fi_blob, fi->fi_bi); ++fi->fi_bi; return ex_let_vars(arg, &tv, TRUE, fi->fi_semicolon, fi->fi_varcount, flag, NULL) == OK; } item = fi->fi_lw.lw_item; if (item == NULL) result = FALSE; else { fi->fi_lw.lw_item = item->li_next; result = (ex_let_vars(arg, &item->li_tv, TRUE, fi->fi_semicolon, fi->fi_varcount, flag, NULL) == OK); } return result; } /* * Free the structure used to store info used by ":for". */ void free_for_info(void *fi_void) { forinfo_T *fi = (forinfo_T *)fi_void; if (fi != NULL && fi->fi_list != NULL) { list_rem_watch(fi->fi_list, &fi->fi_lw); list_unref(fi->fi_list); } if (fi != NULL && fi->fi_blob != NULL) blob_unref(fi->fi_blob); vim_free(fi); } void set_context_for_expression( expand_T *xp, char_u *arg, cmdidx_T cmdidx) { int got_eq = FALSE; int c; char_u *p; if (cmdidx == CMD_let || cmdidx == CMD_const) { xp->xp_context = EXPAND_USER_VARS; if (vim_strpbrk(arg, (char_u *)"\"'+-*/%.=!?~|&$([<>,#") == NULL) { // ":let var1 var2 ...": find last space. for (p = arg + STRLEN(arg); p >= arg; ) { xp->xp_pattern = p; MB_PTR_BACK(arg, p); if (VIM_ISWHITE(*p)) break; } return; } } else xp->xp_context = cmdidx == CMD_call ? EXPAND_FUNCTIONS : EXPAND_EXPRESSION; while ((xp->xp_pattern = vim_strpbrk(arg, (char_u *)"\"'+-*/%.=!?~|&$([<>,#")) != NULL) { c = *xp->xp_pattern; if (c == '&') { c = xp->xp_pattern[1]; if (c == '&') { ++xp->xp_pattern; xp->xp_context = cmdidx != CMD_let || got_eq ? EXPAND_EXPRESSION : EXPAND_NOTHING; } else if (c != ' ') { xp->xp_context = EXPAND_SETTINGS; if ((c == 'l' || c == 'g') && xp->xp_pattern[2] == ':') xp->xp_pattern += 2; } } else if (c == '$') { // environment variable xp->xp_context = EXPAND_ENV_VARS; } else if (c == '=') { got_eq = TRUE; xp->xp_context = EXPAND_EXPRESSION; } else if (c == '#' && xp->xp_context == EXPAND_EXPRESSION) { // Autoload function/variable contains '#'. break; } else if ((c == '<' || c == '#') && xp->xp_context == EXPAND_FUNCTIONS && vim_strchr(xp->xp_pattern, '(') == NULL) { // Function name can start with "" and contain '#'. break; } else if (cmdidx != CMD_let || got_eq) { if (c == '"') // string { while ((c = *++xp->xp_pattern) != NUL && c != '"') if (c == '\\' && xp->xp_pattern[1] != NUL) ++xp->xp_pattern; xp->xp_context = EXPAND_NOTHING; } else if (c == '\'') // literal string { // Trick: '' is like stopping and starting a literal string. while ((c = *++xp->xp_pattern) != NUL && c != '\'') /* skip */ ; xp->xp_context = EXPAND_NOTHING; } else if (c == '|') { if (xp->xp_pattern[1] == '|') { ++xp->xp_pattern; xp->xp_context = EXPAND_EXPRESSION; } else xp->xp_context = EXPAND_COMMANDS; } else xp->xp_context = EXPAND_EXPRESSION; } else // Doesn't look like something valid, expand as an expression // anyway. xp->xp_context = EXPAND_EXPRESSION; arg = xp->xp_pattern; if (*arg != NUL) while ((c = *++arg) != NUL && (c == ' ' || c == '\t')) /* skip */ ; } xp->xp_pattern = arg; } /* * Return TRUE if "pat" matches "text". * Does not use 'cpo' and always uses 'magic'. */ int pattern_match(char_u *pat, char_u *text, int ic) { int matches = FALSE; char_u *save_cpo; regmatch_T regmatch; // avoid 'l' flag in 'cpoptions' save_cpo = p_cpo; p_cpo = (char_u *)""; regmatch.regprog = vim_regcomp(pat, RE_MAGIC + RE_STRING); if (regmatch.regprog != NULL) { regmatch.rm_ic = ic; matches = vim_regexec_nl(®match, text, (colnr_T)0); vim_regfree(regmatch.regprog); } p_cpo = save_cpo; return matches; } /* * Handle a name followed by "(". Both for just "name(arg)" and for * "expr->name(arg)". * Returns OK or FAIL. */ static int eval_func( char_u **arg, // points to "(", will be advanced char_u *name, int name_len, typval_T *rettv, int evaluate, typval_T *basetv) // "expr" for "expr->name(arg)" { char_u *s = name; int len = name_len; partial_T *partial; int ret = OK; if (!evaluate) check_vars(s, len); // If "s" is the name of a variable of type VAR_FUNC // use its contents. s = deref_func_name(s, &len, &partial, !evaluate); // Need to make a copy, in case evaluating the arguments makes // the name invalid. s = vim_strsave(s); if (s == NULL) ret = FAIL; else { funcexe_T funcexe; // Invoke the function. vim_memset(&funcexe, 0, sizeof(funcexe)); funcexe.firstline = curwin->w_cursor.lnum; funcexe.lastline = curwin->w_cursor.lnum; funcexe.evaluate = evaluate; funcexe.partial = partial; funcexe.basetv = basetv; ret = get_func_tv(s, len, rettv, arg, &funcexe); } vim_free(s); // If evaluate is FALSE rettv->v_type was not set in // get_func_tv, but it's needed in handle_subscript() to parse // what follows. So set it here. if (rettv->v_type == VAR_UNKNOWN && !evaluate && **arg == '(') { rettv->vval.v_string = NULL; rettv->v_type = VAR_FUNC; } // Stop the expression evaluation when immediately // aborting on error, or when an interrupt occurred or // an exception was thrown but not caught. if (evaluate && aborting()) { if (ret == OK) clear_tv(rettv); ret = FAIL; } return ret; } /* * The "evaluate" argument: When FALSE, the argument is only parsed but not * executed. The function may return OK, but the rettv will be of type * VAR_UNKNOWN. The function still returns FAIL for a syntax error. */ /* * Handle zero level expression. * This calls eval1() and handles error message and nextcmd. * Put the result in "rettv" when returning OK and "evaluate" is TRUE. * Note: "rettv.v_lock" is not set. * Return OK or FAIL. */ int eval0( char_u *arg, typval_T *rettv, char_u **nextcmd, int evaluate) { int ret; char_u *p; int did_emsg_before = did_emsg; int called_emsg_before = called_emsg; p = skipwhite(arg); ret = eval1(&p, rettv, evaluate); if (ret == FAIL || !ends_excmd(*p)) { if (ret != FAIL) clear_tv(rettv); /* * Report the invalid expression unless the expression evaluation has * been cancelled due to an aborting error, an interrupt, or an * exception, or we already gave a more specific error. * Also check called_emsg for when using assert_fails(). */ if (!aborting() && did_emsg == did_emsg_before && called_emsg == called_emsg_before) semsg(_(e_invexpr2), arg); ret = FAIL; } if (nextcmd != NULL) *nextcmd = check_nextcmd(p); return ret; } /* * Handle top level expression: * expr2 ? expr1 : expr1 * * "arg" must point to the first non-white of the expression. * "arg" is advanced to the next non-white after the recognized expression. * * Note: "rettv.v_lock" is not set. * * Return OK or FAIL. */ int eval1(char_u **arg, typval_T *rettv, int evaluate) { int result; typval_T var2; /* * Get the first variable. */ if (eval2(arg, rettv, evaluate) == FAIL) return FAIL; if ((*arg)[0] == '?') { result = FALSE; if (evaluate) { int error = FALSE; if (tv_get_number_chk(rettv, &error) != 0) result = TRUE; clear_tv(rettv); if (error) return FAIL; } /* * Get the second variable. */ *arg = skipwhite(*arg + 1); if (eval1(arg, rettv, evaluate && result) == FAIL) // recursive! return FAIL; /* * Check for the ":". */ if ((*arg)[0] != ':') { emsg(_(e_missing_colon)); if (evaluate && result) clear_tv(rettv); return FAIL; } /* * Get the third variable. */ *arg = skipwhite(*arg + 1); if (eval1(arg, &var2, evaluate && !result) == FAIL) // recursive! { if (evaluate && result) clear_tv(rettv); return FAIL; } if (evaluate && !result) *rettv = var2; } return OK; } /* * Handle first level expression: * expr2 || expr2 || expr2 logical OR * * "arg" must point to the first non-white of the expression. * "arg" is advanced to the next non-white after the recognized expression. * * Return OK or FAIL. */ static int eval2(char_u **arg, typval_T *rettv, int evaluate) { typval_T var2; long result; int first; int error = FALSE; /* * Get the first variable. */ if (eval3(arg, rettv, evaluate) == FAIL) return FAIL; /* * Repeat until there is no following "||". */ first = TRUE; result = FALSE; while ((*arg)[0] == '|' && (*arg)[1] == '|') { if (evaluate && first) { if (tv_get_number_chk(rettv, &error) != 0) result = TRUE; clear_tv(rettv); if (error) return FAIL; first = FALSE; } /* * Get the second variable. */ *arg = skipwhite(*arg + 2); if (eval3(arg, &var2, evaluate && !result) == FAIL) return FAIL; /* * Compute the result. */ if (evaluate && !result) { if (tv_get_number_chk(&var2, &error) != 0) result = TRUE; clear_tv(&var2); if (error) return FAIL; } if (evaluate) { rettv->v_type = VAR_NUMBER; rettv->vval.v_number = result; } } return OK; } /* * Handle second level expression: * expr3 && expr3 && expr3 logical AND * * "arg" must point to the first non-white of the expression. * "arg" is advanced to the next non-white after the recognized expression. * * Return OK or FAIL. */ static int eval3(char_u **arg, typval_T *rettv, int evaluate) { typval_T var2; long result; int first; int error = FALSE; /* * Get the first variable. */ if (eval4(arg, rettv, evaluate) == FAIL) return FAIL; /* * Repeat until there is no following "&&". */ first = TRUE; result = TRUE; while ((*arg)[0] == '&' && (*arg)[1] == '&') { if (evaluate && first) { if (tv_get_number_chk(rettv, &error) == 0) result = FALSE; clear_tv(rettv); if (error) return FAIL; first = FALSE; } /* * Get the second variable. */ *arg = skipwhite(*arg + 2); if (eval4(arg, &var2, evaluate && result) == FAIL) return FAIL; /* * Compute the result. */ if (evaluate && result) { if (tv_get_number_chk(&var2, &error) == 0) result = FALSE; clear_tv(&var2); if (error) return FAIL; } if (evaluate) { rettv->v_type = VAR_NUMBER; rettv->vval.v_number = result; } } return OK; } /* * Handle third level expression: * var1 == var2 * var1 =~ var2 * var1 != var2 * var1 !~ var2 * var1 > var2 * var1 >= var2 * var1 < var2 * var1 <= var2 * var1 is var2 * var1 isnot var2 * * "arg" must point to the first non-white of the expression. * "arg" is advanced to the next non-white after the recognized expression. * * Return OK or FAIL. */ static int eval4(char_u **arg, typval_T *rettv, int evaluate) { typval_T var2; char_u *p; int i; exptype_T type = EXPR_UNKNOWN; int len = 2; int ic; /* * Get the first variable. */ if (eval5(arg, rettv, evaluate) == FAIL) return FAIL; p = *arg; switch (p[0]) { case '=': if (p[1] == '=') type = EXPR_EQUAL; else if (p[1] == '~') type = EXPR_MATCH; break; case '!': if (p[1] == '=') type = EXPR_NEQUAL; else if (p[1] == '~') type = EXPR_NOMATCH; break; case '>': if (p[1] != '=') { type = EXPR_GREATER; len = 1; } else type = EXPR_GEQUAL; break; case '<': if (p[1] != '=') { type = EXPR_SMALLER; len = 1; } else type = EXPR_SEQUAL; break; case 'i': if (p[1] == 's') { if (p[2] == 'n' && p[3] == 'o' && p[4] == 't') len = 5; i = p[len]; if (!isalnum(i) && i != '_') type = len == 2 ? EXPR_IS : EXPR_ISNOT; } break; } /* * If there is a comparative operator, use it. */ if (type != EXPR_UNKNOWN) { // extra question mark appended: ignore case if (p[len] == '?') { ic = TRUE; ++len; } // extra '#' appended: match case else if (p[len] == '#') { ic = FALSE; ++len; } // nothing appended: use 'ignorecase' else ic = p_ic; /* * Get the second variable. */ *arg = skipwhite(p + len); if (eval5(arg, &var2, evaluate) == FAIL) { clear_tv(rettv); return FAIL; } if (evaluate) { int ret = typval_compare(rettv, &var2, type, ic); clear_tv(&var2); return ret; } } return OK; } void eval_addblob(typval_T *tv1, typval_T *tv2) { blob_T *b1 = tv1->vval.v_blob; blob_T *b2 = tv2->vval.v_blob; blob_T *b = blob_alloc(); int i; if (b != NULL) { for (i = 0; i < blob_len(b1); i++) ga_append(&b->bv_ga, blob_get(b1, i)); for (i = 0; i < blob_len(b2); i++) ga_append(&b->bv_ga, blob_get(b2, i)); clear_tv(tv1); rettv_blob_set(tv1, b); } } int eval_addlist(typval_T *tv1, typval_T *tv2) { typval_T var3; // concatenate Lists if (list_concat(tv1->vval.v_list, tv2->vval.v_list, &var3) == FAIL) { clear_tv(tv1); clear_tv(tv2); return FAIL; } clear_tv(tv1); *tv1 = var3; return OK; } /* * Handle fourth level expression: * + number addition * - number subtraction * . string concatenation (if script version is 1) * .. string concatenation * * "arg" must point to the first non-white of the expression. * "arg" is advanced to the next non-white after the recognized expression. * * Return OK or FAIL. */ static int eval5(char_u **arg, typval_T *rettv, int evaluate) { typval_T var2; int op; varnumber_T n1, n2; #ifdef FEAT_FLOAT float_T f1 = 0, f2 = 0; #endif char_u *s1, *s2; char_u buf1[NUMBUFLEN], buf2[NUMBUFLEN]; char_u *p; int concat; /* * Get the first variable. */ if (eval6(arg, rettv, evaluate, FALSE) == FAIL) return FAIL; /* * Repeat computing, until no '+', '-' or '.' is following. */ for (;;) { // "." is only string concatenation when scriptversion is 1 op = **arg; concat = op == '.' && (*(*arg + 1) == '.' || current_sctx.sc_version < 2); if (op != '+' && op != '-' && !concat) break; if ((op != '+' || (rettv->v_type != VAR_LIST && rettv->v_type != VAR_BLOB)) #ifdef FEAT_FLOAT && (op == '.' || rettv->v_type != VAR_FLOAT) #endif ) { // For "list + ...", an illegal use of the first operand as // a number cannot be determined before evaluating the 2nd // operand: if this is also a list, all is ok. // For "something . ...", "something - ..." or "non-list + ...", // we know that the first operand needs to be a string or number // without evaluating the 2nd operand. So check before to avoid // side effects after an error. if (evaluate && tv_get_string_chk(rettv) == NULL) { clear_tv(rettv); return FAIL; } } /* * Get the second variable. */ if (op == '.' && *(*arg + 1) == '.') // .. string concatenation ++*arg; *arg = skipwhite(*arg + 1); if (eval6(arg, &var2, evaluate, op == '.') == FAIL) { clear_tv(rettv); return FAIL; } if (evaluate) { /* * Compute the result. */ if (op == '.') { s1 = tv_get_string_buf(rettv, buf1); // already checked s2 = tv_get_string_buf_chk(&var2, buf2); if (s2 == NULL) // type error ? { clear_tv(rettv); clear_tv(&var2); return FAIL; } p = concat_str(s1, s2); clear_tv(rettv); rettv->v_type = VAR_STRING; rettv->vval.v_string = p; } else if (op == '+' && rettv->v_type == VAR_BLOB && var2.v_type == VAR_BLOB) eval_addblob(rettv, &var2); else if (op == '+' && rettv->v_type == VAR_LIST && var2.v_type == VAR_LIST) { if (eval_addlist(rettv, &var2) == FAIL) return FAIL; } else { int error = FALSE; #ifdef FEAT_FLOAT if (rettv->v_type == VAR_FLOAT) { f1 = rettv->vval.v_float; n1 = 0; } else #endif { n1 = tv_get_number_chk(rettv, &error); if (error) { // This can only happen for "list + non-list". For // "non-list + ..." or "something - ...", we returned // before evaluating the 2nd operand. clear_tv(rettv); return FAIL; } #ifdef FEAT_FLOAT if (var2.v_type == VAR_FLOAT) f1 = n1; #endif } #ifdef FEAT_FLOAT if (var2.v_type == VAR_FLOAT) { f2 = var2.vval.v_float; n2 = 0; } else #endif { n2 = tv_get_number_chk(&var2, &error); if (error) { clear_tv(rettv); clear_tv(&var2); return FAIL; } #ifdef FEAT_FLOAT if (rettv->v_type == VAR_FLOAT) f2 = n2; #endif } clear_tv(rettv); #ifdef FEAT_FLOAT // If there is a float on either side the result is a float. if (rettv->v_type == VAR_FLOAT || var2.v_type == VAR_FLOAT) { if (op == '+') f1 = f1 + f2; else f1 = f1 - f2; rettv->v_type = VAR_FLOAT; rettv->vval.v_float = f1; } else #endif { if (op == '+') n1 = n1 + n2; else n1 = n1 - n2; rettv->v_type = VAR_NUMBER; rettv->vval.v_number = n1; } } clear_tv(&var2); } } return OK; } /* * Handle fifth level expression: * * number multiplication * / number division * % number modulo * * "arg" must point to the first non-white of the expression. * "arg" is advanced to the next non-white after the recognized expression. * * Return OK or FAIL. */ static int eval6( char_u **arg, typval_T *rettv, int evaluate, int want_string) // after "." operator { typval_T var2; int op; varnumber_T n1, n2; #ifdef FEAT_FLOAT int use_float = FALSE; float_T f1 = 0, f2 = 0; #endif int error = FALSE; /* * Get the first variable. */ if (eval7(arg, rettv, evaluate, want_string) == FAIL) return FAIL; /* * Repeat computing, until no '*', '/' or '%' is following. */ for (;;) { op = **arg; if (op != '*' && op != '/' && op != '%') break; if (evaluate) { #ifdef FEAT_FLOAT if (rettv->v_type == VAR_FLOAT) { f1 = rettv->vval.v_float; use_float = TRUE; n1 = 0; } else #endif n1 = tv_get_number_chk(rettv, &error); clear_tv(rettv); if (error) return FAIL; } else n1 = 0; /* * Get the second variable. */ *arg = skipwhite(*arg + 1); if (eval7(arg, &var2, evaluate, FALSE) == FAIL) return FAIL; if (evaluate) { #ifdef FEAT_FLOAT if (var2.v_type == VAR_FLOAT) { if (!use_float) { f1 = n1; use_float = TRUE; } f2 = var2.vval.v_float; n2 = 0; } else #endif { n2 = tv_get_number_chk(&var2, &error); clear_tv(&var2); if (error) return FAIL; #ifdef FEAT_FLOAT if (use_float) f2 = n2; #endif } /* * Compute the result. * When either side is a float the result is a float. */ #ifdef FEAT_FLOAT if (use_float) { if (op == '*') f1 = f1 * f2; else if (op == '/') { # ifdef VMS // VMS crashes on divide by zero, work around it if (f2 == 0.0) { if (f1 == 0) f1 = -1 * __F_FLT_MAX - 1L; // similar to NaN else if (f1 < 0) f1 = -1 * __F_FLT_MAX; else f1 = __F_FLT_MAX; } else f1 = f1 / f2; # else // We rely on the floating point library to handle divide // by zero to result in "inf" and not a crash. f1 = f1 / f2; # endif } else { emsg(_(e_modulus)); return FAIL; } rettv->v_type = VAR_FLOAT; rettv->vval.v_float = f1; } else #endif { if (op == '*') n1 = n1 * n2; else if (op == '/') n1 = num_divide(n1, n2); else n1 = num_modulus(n1, n2); rettv->v_type = VAR_NUMBER; rettv->vval.v_number = n1; } } } return OK; } /* * Handle sixth level expression: * number number constant * 0zFFFFFFFF Blob constant * "string" string constant * 'string' literal string constant * &option-name option value * @r register contents * identifier variable value * function() function call * $VAR environment variable * (expression) nested expression * [expr, expr] List * {arg, arg -> expr} Lambda * {key: val, key: val} Dictionary * #{key: val, key: val} Dictionary with literal keys * * Also handle: * ! in front logical NOT * - in front unary minus * + in front unary plus (ignored) * trailing [] subscript in String or List * trailing .name entry in Dictionary * trailing ->name() method call * * "arg" must point to the first non-white of the expression. * "arg" is advanced to the next non-white after the recognized expression. * * Return OK or FAIL. */ static int eval7( char_u **arg, typval_T *rettv, int evaluate, int want_string) // after "." operator { int len; char_u *s; char_u *start_leader, *end_leader; int ret = OK; char_u *alias; /* * Initialise variable so that clear_tv() can't mistake this for a * string and free a string that isn't there. */ rettv->v_type = VAR_UNKNOWN; /* * Skip '!', '-' and '+' characters. They are handled later. */ start_leader = *arg; while (**arg == '!' || **arg == '-' || **arg == '+') *arg = skipwhite(*arg + 1); end_leader = *arg; if (**arg == '.' && (!isdigit(*(*arg + 1)) #ifdef FEAT_FLOAT || current_sctx.sc_version < 2 #endif )) { semsg(_(e_invexpr2), *arg); ++*arg; return FAIL; } switch (**arg) { /* * Number constant. */ case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9': case '.': ret = get_number_tv(arg, rettv, evaluate, want_string); break; /* * String constant: "string". */ case '"': ret = get_string_tv(arg, rettv, evaluate); break; /* * Literal string constant: 'str''ing'. */ case '\'': ret = get_lit_string_tv(arg, rettv, evaluate); break; /* * List: [expr, expr] */ case '[': ret = get_list_tv(arg, rettv, evaluate, TRUE); break; /* * Dictionary: #{key: val, key: val} */ case '#': if ((*arg)[1] == '{') { ++*arg; ret = eval_dict(arg, rettv, evaluate, TRUE); } else ret = NOTDONE; break; /* * Lambda: {arg, arg -> expr} * Dictionary: {'key': val, 'key': val} */ case '{': ret = get_lambda_tv(arg, rettv, evaluate); if (ret == NOTDONE) ret = eval_dict(arg, rettv, evaluate, FALSE); break; /* * Option value: &name */ case '&': ret = get_option_tv(arg, rettv, evaluate); break; /* * Environment variable: $VAR. */ case '$': ret = get_env_tv(arg, rettv, evaluate); break; /* * Register contents: @r. */ case '@': ++*arg; if (evaluate) { rettv->v_type = VAR_STRING; rettv->vval.v_string = get_reg_contents(**arg, GREG_EXPR_SRC); } if (**arg != NUL) ++*arg; break; /* * nested expression: (expression). */ case '(': *arg = skipwhite(*arg + 1); ret = eval1(arg, rettv, evaluate); // recursive! if (**arg == ')') ++*arg; else if (ret == OK) { emsg(_(e_missing_close)); clear_tv(rettv); ret = FAIL; } break; default: ret = NOTDONE; break; } if (ret == NOTDONE) { /* * Must be a variable or function name. * Can also be a curly-braces kind of name: {expr}. */ s = *arg; len = get_name_len(arg, &alias, evaluate, TRUE); if (alias != NULL) s = alias; if (len <= 0) ret = FAIL; else { if (**arg == '(') // recursive! ret = eval_func(arg, s, len, rettv, evaluate, NULL); else if (evaluate) ret = get_var_tv(s, len, rettv, NULL, TRUE, FALSE); else { check_vars(s, len); ret = OK; } } vim_free(alias); } *arg = skipwhite(*arg); // Handle following '[', '(' and '.' for expr[expr], expr.name, // expr(expr), expr->name(expr) if (ret == OK) ret = handle_subscript(arg, rettv, evaluate, TRUE, start_leader, &end_leader); /* * Apply logical NOT and unary '-', from right to left, ignore '+'. */ if (ret == OK && evaluate && end_leader > start_leader) ret = eval7_leader(rettv, start_leader, &end_leader); return ret; } /* * Apply the leading "!" and "-" before an eval7 expression to "rettv". * Adjusts "end_leaderp" until it is at "start_leader". */ static int eval7_leader(typval_T *rettv, char_u *start_leader, char_u **end_leaderp) { char_u *end_leader = *end_leaderp; int ret = OK; int error = FALSE; varnumber_T val = 0; #ifdef FEAT_FLOAT float_T f = 0.0; if (rettv->v_type == VAR_FLOAT) f = rettv->vval.v_float; else #endif val = tv_get_number_chk(rettv, &error); if (error) { clear_tv(rettv); ret = FAIL; } else { while (end_leader > start_leader) { --end_leader; if (*end_leader == '!') { #ifdef FEAT_FLOAT if (rettv->v_type == VAR_FLOAT) f = !f; else #endif val = !val; } else if (*end_leader == '-') { #ifdef FEAT_FLOAT if (rettv->v_type == VAR_FLOAT) f = -f; else #endif val = -val; } } #ifdef FEAT_FLOAT if (rettv->v_type == VAR_FLOAT) { clear_tv(rettv); rettv->vval.v_float = f; } else #endif { clear_tv(rettv); rettv->v_type = VAR_NUMBER; rettv->vval.v_number = val; } } *end_leaderp = end_leader; return ret; } /* * Call the function referred to in "rettv". */ static int call_func_rettv( char_u **arg, typval_T *rettv, int evaluate, dict_T *selfdict, typval_T *basetv) { partial_T *pt = NULL; funcexe_T funcexe; typval_T functv; char_u *s; int ret; // need to copy the funcref so that we can clear rettv if (evaluate) { functv = *rettv; rettv->v_type = VAR_UNKNOWN; // Invoke the function. Recursive! if (functv.v_type == VAR_PARTIAL) { pt = functv.vval.v_partial; s = partial_name(pt); } else s = functv.vval.v_string; } else s = (char_u *)""; vim_memset(&funcexe, 0, sizeof(funcexe)); funcexe.firstline = curwin->w_cursor.lnum; funcexe.lastline = curwin->w_cursor.lnum; funcexe.evaluate = evaluate; funcexe.partial = pt; funcexe.selfdict = selfdict; funcexe.basetv = basetv; ret = get_func_tv(s, -1, rettv, arg, &funcexe); // Clear the funcref afterwards, so that deleting it while // evaluating the arguments is possible (see test55). if (evaluate) clear_tv(&functv); return ret; } /* * Evaluate "->method()". * "*arg" points to the '-'. * Returns FAIL or OK. "*arg" is advanced to after the ')'. */ static int eval_lambda( char_u **arg, typval_T *rettv, int evaluate, int verbose) // give error messages { typval_T base = *rettv; int ret; // Skip over the ->. *arg += 2; rettv->v_type = VAR_UNKNOWN; ret = get_lambda_tv(arg, rettv, evaluate); if (ret != OK) return FAIL; else if (**arg != '(') { if (verbose) { if (*skipwhite(*arg) == '(') emsg(_(e_nowhitespace)); else semsg(_(e_missing_paren), "lambda"); } clear_tv(rettv); ret = FAIL; } else ret = call_func_rettv(arg, rettv, evaluate, NULL, &base); // Clear the funcref afterwards, so that deleting it while // evaluating the arguments is possible (see test55). if (evaluate) clear_tv(&base); return ret; } /* * Evaluate "->method()". * "*arg" points to the '-'. * Returns FAIL or OK. "*arg" is advanced to after the ')'. */ static int eval_method( char_u **arg, typval_T *rettv, int evaluate, int verbose) // give error messages { char_u *name; long len; char_u *alias; typval_T base = *rettv; int ret; // Skip over the ->. *arg += 2; rettv->v_type = VAR_UNKNOWN; name = *arg; len = get_name_len(arg, &alias, evaluate, TRUE); if (alias != NULL) name = alias; if (len <= 0) { if (verbose) emsg(_("E260: Missing name after ->")); ret = FAIL; } else { if (**arg != '(') { if (verbose) semsg(_(e_missing_paren), name); ret = FAIL; } else if (VIM_ISWHITE((*arg)[-1])) { if (verbose) emsg(_(e_nowhitespace)); ret = FAIL; } else ret = eval_func(arg, name, len, rettv, evaluate, &base); } // Clear the funcref afterwards, so that deleting it while // evaluating the arguments is possible (see test55). if (evaluate) clear_tv(&base); return ret; } /* * Evaluate an "[expr]" or "[expr:expr]" index. Also "dict.key". * "*arg" points to the '[' or '.'. * Returns FAIL or OK. "*arg" is advanced to after the ']'. */ static int eval_index( char_u **arg, typval_T *rettv, int evaluate, int verbose) // give error messages { int empty1 = FALSE, empty2 = FALSE; typval_T var1, var2; long i; long n1, n2 = 0; long len = -1; int range = FALSE; char_u *s; char_u *key = NULL; switch (rettv->v_type) { case VAR_FUNC: case VAR_PARTIAL: if (verbose) emsg(_("E695: Cannot index a Funcref")); return FAIL; case VAR_FLOAT: #ifdef FEAT_FLOAT if (verbose) emsg(_(e_float_as_string)); return FAIL; #endif case VAR_BOOL: case VAR_SPECIAL: case VAR_JOB: case VAR_CHANNEL: if (verbose) emsg(_("E909: Cannot index a special variable")); return FAIL; case VAR_UNKNOWN: case VAR_VOID: if (evaluate) return FAIL; // FALLTHROUGH case VAR_STRING: case VAR_NUMBER: case VAR_LIST: case VAR_DICT: case VAR_BLOB: break; } init_tv(&var1); init_tv(&var2); if (**arg == '.') { /* * dict.name */ key = *arg + 1; for (len = 0; ASCII_ISALNUM(key[len]) || key[len] == '_'; ++len) ; if (len == 0) return FAIL; *arg = skipwhite(key + len); } else { /* * something[idx] * * Get the (first) variable from inside the []. */ *arg = skipwhite(*arg + 1); if (**arg == ':') empty1 = TRUE; else if (eval1(arg, &var1, evaluate) == FAIL) // recursive! return FAIL; else if (evaluate && tv_get_string_chk(&var1) == NULL) { // not a number or string clear_tv(&var1); return FAIL; } /* * Get the second variable from inside the [:]. */ if (**arg == ':') { range = TRUE; *arg = skipwhite(*arg + 1); if (**arg == ']') empty2 = TRUE; else if (eval1(arg, &var2, evaluate) == FAIL) // recursive! { if (!empty1) clear_tv(&var1); return FAIL; } else if (evaluate && tv_get_string_chk(&var2) == NULL) { // not a number or string if (!empty1) clear_tv(&var1); clear_tv(&var2); return FAIL; } } // Check for the ']'. if (**arg != ']') { if (verbose) emsg(_(e_missbrac)); clear_tv(&var1); if (range) clear_tv(&var2); return FAIL; } *arg = skipwhite(*arg + 1); // skip the ']' } if (evaluate) { n1 = 0; if (!empty1 && rettv->v_type != VAR_DICT) { n1 = tv_get_number(&var1); clear_tv(&var1); } if (range) { if (empty2) n2 = -1; else { n2 = tv_get_number(&var2); clear_tv(&var2); } } switch (rettv->v_type) { case VAR_UNKNOWN: case VAR_VOID: case VAR_FUNC: case VAR_PARTIAL: case VAR_FLOAT: case VAR_BOOL: case VAR_SPECIAL: case VAR_JOB: case VAR_CHANNEL: break; // not evaluating, skipping over subscript case VAR_NUMBER: case VAR_STRING: s = tv_get_string(rettv); len = (long)STRLEN(s); if (range) { // The resulting variable is a substring. If the indexes // are out of range the result is empty. if (n1 < 0) { n1 = len + n1; if (n1 < 0) n1 = 0; } if (n2 < 0) n2 = len + n2; else if (n2 >= len) n2 = len; if (n1 >= len || n2 < 0 || n1 > n2) s = NULL; else s = vim_strnsave(s + n1, (int)(n2 - n1 + 1)); } else { // The resulting variable is a string of a single // character. If the index is too big or negative the // result is empty. if (n1 >= len || n1 < 0) s = NULL; else s = vim_strnsave(s + n1, 1); } clear_tv(rettv); rettv->v_type = VAR_STRING; rettv->vval.v_string = s; break; case VAR_BLOB: len = blob_len(rettv->vval.v_blob); if (range) { // The resulting variable is a sub-blob. If the indexes // are out of range the result is empty. if (n1 < 0) { n1 = len + n1; if (n1 < 0) n1 = 0; } if (n2 < 0) n2 = len + n2; else if (n2 >= len) n2 = len - 1; if (n1 >= len || n2 < 0 || n1 > n2) { clear_tv(rettv); rettv->v_type = VAR_BLOB; rettv->vval.v_blob = NULL; } else { blob_T *blob = blob_alloc(); if (blob != NULL) { if (ga_grow(&blob->bv_ga, n2 - n1 + 1) == FAIL) { blob_free(blob); return FAIL; } blob->bv_ga.ga_len = n2 - n1 + 1; for (i = n1; i <= n2; i++) blob_set(blob, i - n1, blob_get(rettv->vval.v_blob, i)); clear_tv(rettv); rettv_blob_set(rettv, blob); } } } else { // The resulting variable is a byte value. // If the index is too big or negative that is an error. if (n1 < 0) n1 = len + n1; if (n1 < len && n1 >= 0) { int v = blob_get(rettv->vval.v_blob, n1); clear_tv(rettv); rettv->v_type = VAR_NUMBER; rettv->vval.v_number = v; } else semsg(_(e_blobidx), n1); } break; case VAR_LIST: len = list_len(rettv->vval.v_list); if (n1 < 0) n1 = len + n1; if (!empty1 && (n1 < 0 || n1 >= len)) { // For a range we allow invalid values and return an empty // list. A list index out of range is an error. if (!range) { if (verbose) semsg(_(e_listidx), n1); return FAIL; } n1 = len; } if (range) { list_T *l; listitem_T *item; if (n2 < 0) n2 = len + n2; else if (n2 >= len) n2 = len - 1; if (!empty2 && (n2 < 0 || n2 + 1 < n1)) n2 = -1; l = list_alloc(); if (l == NULL) return FAIL; for (item = list_find(rettv->vval.v_list, n1); n1 <= n2; ++n1) { if (list_append_tv(l, &item->li_tv) == FAIL) { list_free(l); return FAIL; } item = item->li_next; } clear_tv(rettv); rettv_list_set(rettv, l); } else { copy_tv(&list_find(rettv->vval.v_list, n1)->li_tv, &var1); clear_tv(rettv); *rettv = var1; } break; case VAR_DICT: if (range) { if (verbose) emsg(_(e_dictrange)); if (len == -1) clear_tv(&var1); return FAIL; } { dictitem_T *item; if (len == -1) { key = tv_get_string_chk(&var1); if (key == NULL) { clear_tv(&var1); return FAIL; } } item = dict_find(rettv->vval.v_dict, key, (int)len); if (item == NULL && verbose) semsg(_(e_dictkey), key); if (len == -1) clear_tv(&var1); if (item == NULL) return FAIL; copy_tv(&item->di_tv, &var1); clear_tv(rettv); *rettv = var1; } break; } } return OK; } /* * Get an option value. * "arg" points to the '&' or '+' before the option name. * "arg" is advanced to character after the option name. * Return OK or FAIL. */ int get_option_tv( char_u **arg, typval_T *rettv, // when NULL, only check if option exists int evaluate) { char_u *option_end; long numval; char_u *stringval; int opt_type; int c; int working = (**arg == '+'); // has("+option") int ret = OK; int opt_flags; /* * Isolate the option name and find its value. */ option_end = find_option_end(arg, &opt_flags); if (option_end == NULL) { if (rettv != NULL) semsg(_("E112: Option name missing: %s"), *arg); return FAIL; } if (!evaluate) { *arg = option_end; return OK; } c = *option_end; *option_end = NUL; opt_type = get_option_value(*arg, &numval, rettv == NULL ? NULL : &stringval, opt_flags); if (opt_type == -3) // invalid name { if (rettv != NULL) semsg(_(e_unknown_option), *arg); ret = FAIL; } else if (rettv != NULL) { if (opt_type == -2) // hidden string option { rettv->v_type = VAR_STRING; rettv->vval.v_string = NULL; } else if (opt_type == -1) // hidden number option { rettv->v_type = VAR_NUMBER; rettv->vval.v_number = 0; } else if (opt_type == 1) // number option { rettv->v_type = VAR_NUMBER; rettv->vval.v_number = numval; } else // string option { rettv->v_type = VAR_STRING; rettv->vval.v_string = stringval; } } else if (working && (opt_type == -2 || opt_type == -1)) ret = FAIL; *option_end = c; // put back for error messages *arg = option_end; return ret; } /* * Allocate a variable for a number constant. Also deals with "0z" for blob. * Return OK or FAIL. */ int get_number_tv( char_u **arg, typval_T *rettv, int evaluate, int want_string UNUSED) { int len; #ifdef FEAT_FLOAT char_u *p; int get_float = FALSE; // We accept a float when the format matches // "[0-9]\+\.[0-9]\+\([eE][+-]\?[0-9]\+\)\?". This is very // strict to avoid backwards compatibility problems. // With script version 2 and later the leading digit can be // omitted. // Don't look for a float after the "." operator, so that // ":let vers = 1.2.3" doesn't fail. if (**arg == '.') p = *arg; else p = skipdigits(*arg + 1); if (!want_string && p[0] == '.' && vim_isdigit(p[1])) { get_float = TRUE; p = skipdigits(p + 2); if (*p == 'e' || *p == 'E') { ++p; if (*p == '-' || *p == '+') ++p; if (!vim_isdigit(*p)) get_float = FALSE; else p = skipdigits(p + 1); } if (ASCII_ISALPHA(*p) || *p == '.') get_float = FALSE; } if (get_float) { float_T f; *arg += string2float(*arg, &f); if (evaluate) { rettv->v_type = VAR_FLOAT; rettv->vval.v_float = f; } } else #endif if (**arg == '0' && ((*arg)[1] == 'z' || (*arg)[1] == 'Z')) { char_u *bp; blob_T *blob = NULL; // init for gcc // Blob constant: 0z0123456789abcdef if (evaluate) blob = blob_alloc(); for (bp = *arg + 2; vim_isxdigit(bp[0]); bp += 2) { if (!vim_isxdigit(bp[1])) { if (blob != NULL) { emsg(_("E973: Blob literal should have an even number of hex characters")); ga_clear(&blob->bv_ga); VIM_CLEAR(blob); } return FAIL; } if (blob != NULL) ga_append(&blob->bv_ga, (hex2nr(*bp) << 4) + hex2nr(*(bp+1))); if (bp[2] == '.' && vim_isxdigit(bp[3])) ++bp; } if (blob != NULL) rettv_blob_set(rettv, blob); *arg = bp; } else { varnumber_T n; // decimal, hex or octal number vim_str2nr(*arg, NULL, &len, current_sctx.sc_version >= 4 ? STR2NR_NO_OCT + STR2NR_QUOTE : STR2NR_ALL, &n, NULL, 0, TRUE); if (len == 0) { semsg(_(e_invexpr2), *arg); return FAIL; } *arg += len; if (evaluate) { rettv->v_type = VAR_NUMBER; rettv->vval.v_number = n; } } return OK; } /* * Allocate a variable for a string constant. * Return OK or FAIL. */ int get_string_tv(char_u **arg, typval_T *rettv, int evaluate) { char_u *p; char_u *name; int extra = 0; /* * Find the end of the string, skipping backslashed characters. */ for (p = *arg + 1; *p != NUL && *p != '"'; MB_PTR_ADV(p)) { if (*p == '\\' && p[1] != NUL) { ++p; // A "\" form occupies at least 4 characters, and produces up // to 6 characters: reserve space for 2 extra if (*p == '<') extra += 2; } } if (*p != '"') { semsg(_("E114: Missing quote: %s"), *arg); return FAIL; } // If only parsing, set *arg and return here if (!evaluate) { *arg = p + 1; return OK; } /* * Copy the string into allocated memory, handling backslashed * characters. */ name = alloc(p - *arg + extra); if (name == NULL) return FAIL; rettv->v_type = VAR_STRING; rettv->vval.v_string = name; for (p = *arg + 1; *p != NUL && *p != '"'; ) { if (*p == '\\') { switch (*++p) { case 'b': *name++ = BS; ++p; break; case 'e': *name++ = ESC; ++p; break; case 'f': *name++ = FF; ++p; break; case 'n': *name++ = NL; ++p; break; case 'r': *name++ = CAR; ++p; break; case 't': *name++ = TAB; ++p; break; case 'X': // hex: "\x1", "\x12" case 'x': case 'u': // Unicode: "\u0023" case 'U': if (vim_isxdigit(p[1])) { int n, nr; int c = toupper(*p); if (c == 'X') n = 2; else if (*p == 'u') n = 4; else n = 8; nr = 0; while (--n >= 0 && vim_isxdigit(p[1])) { ++p; nr = (nr << 4) + hex2nr(*p); } ++p; // For "\u" store the number according to // 'encoding'. if (c != 'X') name += (*mb_char2bytes)(nr, name); else *name++ = nr; } break; // octal: "\1", "\12", "\123" case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': *name = *p++ - '0'; if (*p >= '0' && *p <= '7') { *name = (*name << 3) + *p++ - '0'; if (*p >= '0' && *p <= '7') *name = (*name << 3) + *p++ - '0'; } ++name; break; // Special key, e.g.: "\" case '<': extra = trans_special(&p, name, TRUE, TRUE, TRUE, NULL); if (extra != 0) { name += extra; break; } // FALLTHROUGH default: MB_COPY_CHAR(p, name); break; } } else MB_COPY_CHAR(p, name); } *name = NUL; if (*p != NUL) // just in case ++p; *arg = p; return OK; } /* * Allocate a variable for a 'str''ing' constant. * Return OK or FAIL. */ int get_lit_string_tv(char_u **arg, typval_T *rettv, int evaluate) { char_u *p; char_u *str; int reduce = 0; /* * Find the end of the string, skipping ''. */ for (p = *arg + 1; *p != NUL; MB_PTR_ADV(p)) { if (*p == '\'') { if (p[1] != '\'') break; ++reduce; ++p; } } if (*p != '\'') { semsg(_("E115: Missing quote: %s"), *arg); return FAIL; } // If only parsing return after setting "*arg" if (!evaluate) { *arg = p + 1; return OK; } /* * Copy the string into allocated memory, handling '' to ' reduction. */ str = alloc((p - *arg) - reduce); if (str == NULL) return FAIL; rettv->v_type = VAR_STRING; rettv->vval.v_string = str; for (p = *arg + 1; *p != NUL; ) { if (*p == '\'') { if (p[1] != '\'') break; ++p; } MB_COPY_CHAR(p, str); } *str = NUL; *arg = p + 1; return OK; } /* * Return the function name of the partial. */ char_u * partial_name(partial_T *pt) { if (pt->pt_name != NULL) return pt->pt_name; return pt->pt_func->uf_name; } static void partial_free(partial_T *pt) { int i; for (i = 0; i < pt->pt_argc; ++i) clear_tv(&pt->pt_argv[i]); vim_free(pt->pt_argv); dict_unref(pt->pt_dict); if (pt->pt_name != NULL) { func_unref(pt->pt_name); vim_free(pt->pt_name); } else func_ptr_unref(pt->pt_func); vim_free(pt); } /* * Unreference a closure: decrement the reference count and free it when it * becomes zero. */ void partial_unref(partial_T *pt) { if (pt != NULL && --pt->pt_refcount <= 0) partial_free(pt); } static int tv_equal_recurse_limit; static int func_equal( typval_T *tv1, typval_T *tv2, int ic) // ignore case { char_u *s1, *s2; dict_T *d1, *d2; int a1, a2; int i; // empty and NULL function name considered the same s1 = tv1->v_type == VAR_FUNC ? tv1->vval.v_string : partial_name(tv1->vval.v_partial); if (s1 != NULL && *s1 == NUL) s1 = NULL; s2 = tv2->v_type == VAR_FUNC ? tv2->vval.v_string : partial_name(tv2->vval.v_partial); if (s2 != NULL && *s2 == NUL) s2 = NULL; if (s1 == NULL || s2 == NULL) { if (s1 != s2) return FALSE; } else if (STRCMP(s1, s2) != 0) return FALSE; // empty dict and NULL dict is different d1 = tv1->v_type == VAR_FUNC ? NULL : tv1->vval.v_partial->pt_dict; d2 = tv2->v_type == VAR_FUNC ? NULL : tv2->vval.v_partial->pt_dict; if (d1 == NULL || d2 == NULL) { if (d1 != d2) return FALSE; } else if (!dict_equal(d1, d2, ic, TRUE)) return FALSE; // empty list and no list considered the same a1 = tv1->v_type == VAR_FUNC ? 0 : tv1->vval.v_partial->pt_argc; a2 = tv2->v_type == VAR_FUNC ? 0 : tv2->vval.v_partial->pt_argc; if (a1 != a2) return FALSE; for (i = 0; i < a1; ++i) if (!tv_equal(tv1->vval.v_partial->pt_argv + i, tv2->vval.v_partial->pt_argv + i, ic, TRUE)) return FALSE; return TRUE; } /* * Return TRUE if "tv1" and "tv2" have the same value. * Compares the items just like "==" would compare them, but strings and * numbers are different. Floats and numbers are also different. */ int tv_equal( typval_T *tv1, typval_T *tv2, int ic, // ignore case int recursive) // TRUE when used recursively { char_u buf1[NUMBUFLEN], buf2[NUMBUFLEN]; char_u *s1, *s2; static int recursive_cnt = 0; // catch recursive loops int r; // Catch lists and dicts that have an endless loop by limiting // recursiveness to a limit. We guess they are equal then. // A fixed limit has the problem of still taking an awful long time. // Reduce the limit every time running into it. That should work fine for // deeply linked structures that are not recursively linked and catch // recursiveness quickly. if (!recursive) tv_equal_recurse_limit = 1000; if (recursive_cnt >= tv_equal_recurse_limit) { --tv_equal_recurse_limit; return TRUE; } // For VAR_FUNC and VAR_PARTIAL compare the function name, bound dict and // arguments. if ((tv1->v_type == VAR_FUNC || (tv1->v_type == VAR_PARTIAL && tv1->vval.v_partial != NULL)) && (tv2->v_type == VAR_FUNC || (tv2->v_type == VAR_PARTIAL && tv2->vval.v_partial != NULL))) { ++recursive_cnt; r = func_equal(tv1, tv2, ic); --recursive_cnt; return r; } if (tv1->v_type != tv2->v_type) return FALSE; switch (tv1->v_type) { case VAR_LIST: ++recursive_cnt; r = list_equal(tv1->vval.v_list, tv2->vval.v_list, ic, TRUE); --recursive_cnt; return r; case VAR_DICT: ++recursive_cnt; r = dict_equal(tv1->vval.v_dict, tv2->vval.v_dict, ic, TRUE); --recursive_cnt; return r; case VAR_BLOB: return blob_equal(tv1->vval.v_blob, tv2->vval.v_blob); case VAR_NUMBER: case VAR_BOOL: case VAR_SPECIAL: return tv1->vval.v_number == tv2->vval.v_number; case VAR_STRING: s1 = tv_get_string_buf(tv1, buf1); s2 = tv_get_string_buf(tv2, buf2); return ((ic ? MB_STRICMP(s1, s2) : STRCMP(s1, s2)) == 0); case VAR_FLOAT: #ifdef FEAT_FLOAT return tv1->vval.v_float == tv2->vval.v_float; #endif case VAR_JOB: #ifdef FEAT_JOB_CHANNEL return tv1->vval.v_job == tv2->vval.v_job; #endif case VAR_CHANNEL: #ifdef FEAT_JOB_CHANNEL return tv1->vval.v_channel == tv2->vval.v_channel; #endif case VAR_PARTIAL: return tv1->vval.v_partial == tv2->vval.v_partial; case VAR_FUNC: return tv1->vval.v_string == tv2->vval.v_string; case VAR_UNKNOWN: case VAR_VOID: break; } // VAR_UNKNOWN can be the result of a invalid expression, let's say it // does not equal anything, not even itself. return FALSE; } /* * Return the next (unique) copy ID. * Used for serializing nested structures. */ int get_copyID(void) { current_copyID += COPYID_INC; return current_copyID; } /* * Garbage collection for lists and dictionaries. * * We use reference counts to be able to free most items right away when they * are no longer used. But for composite items it's possible that it becomes * unused while the reference count is > 0: When there is a recursive * reference. Example: * :let l = [1, 2, 3] * :let d = {9: l} * :let l[1] = d * * Since this is quite unusual we handle this with garbage collection: every * once in a while find out which lists and dicts are not referenced from any * variable. * * Here is a good reference text about garbage collection (refers to Python * but it applies to all reference-counting mechanisms): * http://python.ca/nas/python/gc/ */ /* * Do garbage collection for lists and dicts. * When "testing" is TRUE this is called from test_garbagecollect_now(). * Return TRUE if some memory was freed. */ int garbage_collect(int testing) { int copyID; int abort = FALSE; buf_T *buf; win_T *wp; int did_free = FALSE; tabpage_T *tp; if (!testing) { // Only do this once. want_garbage_collect = FALSE; may_garbage_collect = FALSE; garbage_collect_at_exit = FALSE; } // The execution stack can grow big, limit the size. if (exestack.ga_maxlen - exestack.ga_len > 500) { size_t new_len; char_u *pp; int n; // Keep 150% of the current size, with a minimum of the growth size. n = exestack.ga_len / 2; if (n < exestack.ga_growsize) n = exestack.ga_growsize; // Don't make it bigger though. if (exestack.ga_len + n < exestack.ga_maxlen) { new_len = exestack.ga_itemsize * (exestack.ga_len + n); pp = vim_realloc(exestack.ga_data, new_len); if (pp == NULL) return FAIL; exestack.ga_maxlen = exestack.ga_len + n; exestack.ga_data = pp; } } // We advance by two because we add one for items referenced through // previous_funccal. copyID = get_copyID(); /* * 1. Go through all accessible variables and mark all lists and dicts * with copyID. */ // Don't free variables in the previous_funccal list unless they are only // referenced through previous_funccal. This must be first, because if // the item is referenced elsewhere the funccal must not be freed. abort = abort || set_ref_in_previous_funccal(copyID); // script-local variables abort = abort || garbage_collect_scriptvars(copyID); // buffer-local variables FOR_ALL_BUFFERS(buf) abort = abort || set_ref_in_item(&buf->b_bufvar.di_tv, copyID, NULL, NULL); // window-local variables FOR_ALL_TAB_WINDOWS(tp, wp) abort = abort || set_ref_in_item(&wp->w_winvar.di_tv, copyID, NULL, NULL); if (aucmd_win != NULL) abort = abort || set_ref_in_item(&aucmd_win->w_winvar.di_tv, copyID, NULL, NULL); #ifdef FEAT_PROP_POPUP for (wp = first_popupwin; wp != NULL; wp = wp->w_next) abort = abort || set_ref_in_item(&wp->w_winvar.di_tv, copyID, NULL, NULL); FOR_ALL_TABPAGES(tp) for (wp = tp->tp_first_popupwin; wp != NULL; wp = wp->w_next) abort = abort || set_ref_in_item(&wp->w_winvar.di_tv, copyID, NULL, NULL); #endif // tabpage-local variables FOR_ALL_TABPAGES(tp) abort = abort || set_ref_in_item(&tp->tp_winvar.di_tv, copyID, NULL, NULL); // global variables abort = abort || garbage_collect_globvars(copyID); // function-local variables abort = abort || set_ref_in_call_stack(copyID); // named functions (matters for closures) abort = abort || set_ref_in_functions(copyID); // function call arguments, if v:testing is set. abort = abort || set_ref_in_func_args(copyID); // v: vars abort = abort || garbage_collect_vimvars(copyID); // callbacks in buffers abort = abort || set_ref_in_buffers(copyID); #ifdef FEAT_LUA abort = abort || set_ref_in_lua(copyID); #endif #ifdef FEAT_PYTHON abort = abort || set_ref_in_python(copyID); #endif #ifdef FEAT_PYTHON3 abort = abort || set_ref_in_python3(copyID); #endif #ifdef FEAT_JOB_CHANNEL abort = abort || set_ref_in_channel(copyID); abort = abort || set_ref_in_job(copyID); #endif #ifdef FEAT_NETBEANS_INTG abort = abort || set_ref_in_nb_channel(copyID); #endif #ifdef FEAT_TIMERS abort = abort || set_ref_in_timer(copyID); #endif #ifdef FEAT_QUICKFIX abort = abort || set_ref_in_quickfix(copyID); #endif #ifdef FEAT_TERMINAL abort = abort || set_ref_in_term(copyID); #endif #ifdef FEAT_PROP_POPUP abort = abort || set_ref_in_popups(copyID); #endif if (!abort) { /* * 2. Free lists and dictionaries that are not referenced. */ did_free = free_unref_items(copyID); /* * 3. Check if any funccal can be freed now. * This may call us back recursively. */ free_unref_funccal(copyID, testing); } else if (p_verbose > 0) { verb_msg(_("Not enough memory to set references, garbage collection aborted!")); } return did_free; } /* * Free lists, dictionaries, channels and jobs that are no longer referenced. */ static int free_unref_items(int copyID) { int did_free = FALSE; // Let all "free" functions know that we are here. This means no // dictionaries, lists, channels or jobs are to be freed, because we will // do that here. in_free_unref_items = TRUE; /* * PASS 1: free the contents of the items. We don't free the items * themselves yet, so that it is possible to decrement refcount counters */ // Go through the list of dicts and free items without the copyID. did_free |= dict_free_nonref(copyID); // Go through the list of lists and free items without the copyID. did_free |= list_free_nonref(copyID); #ifdef FEAT_JOB_CHANNEL // Go through the list of jobs and free items without the copyID. This // must happen before doing channels, because jobs refer to channels, but // the reference from the channel to the job isn't tracked. did_free |= free_unused_jobs_contents(copyID, COPYID_MASK); // Go through the list of channels and free items without the copyID. did_free |= free_unused_channels_contents(copyID, COPYID_MASK); #endif /* * PASS 2: free the items themselves. */ dict_free_items(copyID); list_free_items(copyID); #ifdef FEAT_JOB_CHANNEL // Go through the list of jobs and free items without the copyID. This // must happen before doing channels, because jobs refer to channels, but // the reference from the channel to the job isn't tracked. free_unused_jobs(copyID, COPYID_MASK); // Go through the list of channels and free items without the copyID. free_unused_channels(copyID, COPYID_MASK); #endif in_free_unref_items = FALSE; return did_free; } /* * Mark all lists and dicts referenced through hashtab "ht" with "copyID". * "list_stack" is used to add lists to be marked. Can be NULL. * * Returns TRUE if setting references failed somehow. */ int set_ref_in_ht(hashtab_T *ht, int copyID, list_stack_T **list_stack) { int todo; int abort = FALSE; hashitem_T *hi; hashtab_T *cur_ht; ht_stack_T *ht_stack = NULL; ht_stack_T *tempitem; cur_ht = ht; for (;;) { if (!abort) { // Mark each item in the hashtab. If the item contains a hashtab // it is added to ht_stack, if it contains a list it is added to // list_stack. todo = (int)cur_ht->ht_used; for (hi = cur_ht->ht_array; todo > 0; ++hi) if (!HASHITEM_EMPTY(hi)) { --todo; abort = abort || set_ref_in_item(&HI2DI(hi)->di_tv, copyID, &ht_stack, list_stack); } } if (ht_stack == NULL) break; // take an item from the stack cur_ht = ht_stack->ht; tempitem = ht_stack; ht_stack = ht_stack->prev; free(tempitem); } return abort; } /* * Mark a dict and its items with "copyID". * Returns TRUE if setting references failed somehow. */ int set_ref_in_dict(dict_T *d, int copyID) { if (d != NULL && d->dv_copyID != copyID) { d->dv_copyID = copyID; return set_ref_in_ht(&d->dv_hashtab, copyID, NULL); } return FALSE; } /* * Mark a list and its items with "copyID". * Returns TRUE if setting references failed somehow. */ int set_ref_in_list(list_T *ll, int copyID) { if (ll != NULL && ll->lv_copyID != copyID) { ll->lv_copyID = copyID; return set_ref_in_list_items(ll, copyID, NULL); } return FALSE; } /* * Mark all lists and dicts referenced through list "l" with "copyID". * "ht_stack" is used to add hashtabs to be marked. Can be NULL. * * Returns TRUE if setting references failed somehow. */ int set_ref_in_list_items(list_T *l, int copyID, ht_stack_T **ht_stack) { listitem_T *li; int abort = FALSE; list_T *cur_l; list_stack_T *list_stack = NULL; list_stack_T *tempitem; cur_l = l; for (;;) { if (!abort && cur_l->lv_first != &range_list_item) // Mark each item in the list. If the item contains a hashtab // it is added to ht_stack, if it contains a list it is added to // list_stack. for (li = cur_l->lv_first; !abort && li != NULL; li = li->li_next) abort = abort || set_ref_in_item(&li->li_tv, copyID, ht_stack, &list_stack); if (list_stack == NULL) break; // take an item from the stack cur_l = list_stack->list; tempitem = list_stack; list_stack = list_stack->prev; free(tempitem); } return abort; } /* * Mark all lists and dicts referenced through typval "tv" with "copyID". * "list_stack" is used to add lists to be marked. Can be NULL. * "ht_stack" is used to add hashtabs to be marked. Can be NULL. * * Returns TRUE if setting references failed somehow. */ int set_ref_in_item( typval_T *tv, int copyID, ht_stack_T **ht_stack, list_stack_T **list_stack) { int abort = FALSE; if (tv->v_type == VAR_DICT) { dict_T *dd = tv->vval.v_dict; if (dd != NULL && dd->dv_copyID != copyID) { // Didn't see this dict yet. dd->dv_copyID = copyID; if (ht_stack == NULL) { abort = set_ref_in_ht(&dd->dv_hashtab, copyID, list_stack); } else { ht_stack_T *newitem = (ht_stack_T*)malloc(sizeof(ht_stack_T)); if (newitem == NULL) abort = TRUE; else { newitem->ht = &dd->dv_hashtab; newitem->prev = *ht_stack; *ht_stack = newitem; } } } } else if (tv->v_type == VAR_LIST) { list_T *ll = tv->vval.v_list; if (ll != NULL && ll->lv_copyID != copyID) { // Didn't see this list yet. ll->lv_copyID = copyID; if (list_stack == NULL) { abort = set_ref_in_list_items(ll, copyID, ht_stack); } else { list_stack_T *newitem = (list_stack_T*)malloc( sizeof(list_stack_T)); if (newitem == NULL) abort = TRUE; else { newitem->list = ll; newitem->prev = *list_stack; *list_stack = newitem; } } } } else if (tv->v_type == VAR_FUNC) { abort = set_ref_in_func(tv->vval.v_string, NULL, copyID); } else if (tv->v_type == VAR_PARTIAL) { partial_T *pt = tv->vval.v_partial; int i; // A partial does not have a copyID, because it cannot contain itself. if (pt != NULL) { abort = set_ref_in_func(pt->pt_name, pt->pt_func, copyID); if (pt->pt_dict != NULL) { typval_T dtv; dtv.v_type = VAR_DICT; dtv.vval.v_dict = pt->pt_dict; set_ref_in_item(&dtv, copyID, ht_stack, list_stack); } for (i = 0; i < pt->pt_argc; ++i) abort = abort || set_ref_in_item(&pt->pt_argv[i], copyID, ht_stack, list_stack); } } #ifdef FEAT_JOB_CHANNEL else if (tv->v_type == VAR_JOB) { job_T *job = tv->vval.v_job; typval_T dtv; if (job != NULL && job->jv_copyID != copyID) { job->jv_copyID = copyID; if (job->jv_channel != NULL) { dtv.v_type = VAR_CHANNEL; dtv.vval.v_channel = job->jv_channel; set_ref_in_item(&dtv, copyID, ht_stack, list_stack); } if (job->jv_exit_cb.cb_partial != NULL) { dtv.v_type = VAR_PARTIAL; dtv.vval.v_partial = job->jv_exit_cb.cb_partial; set_ref_in_item(&dtv, copyID, ht_stack, list_stack); } } } else if (tv->v_type == VAR_CHANNEL) { channel_T *ch =tv->vval.v_channel; ch_part_T part; typval_T dtv; jsonq_T *jq; cbq_T *cq; if (ch != NULL && ch->ch_copyID != copyID) { ch->ch_copyID = copyID; for (part = PART_SOCK; part < PART_COUNT; ++part) { for (jq = ch->ch_part[part].ch_json_head.jq_next; jq != NULL; jq = jq->jq_next) set_ref_in_item(jq->jq_value, copyID, ht_stack, list_stack); for (cq = ch->ch_part[part].ch_cb_head.cq_next; cq != NULL; cq = cq->cq_next) if (cq->cq_callback.cb_partial != NULL) { dtv.v_type = VAR_PARTIAL; dtv.vval.v_partial = cq->cq_callback.cb_partial; set_ref_in_item(&dtv, copyID, ht_stack, list_stack); } if (ch->ch_part[part].ch_callback.cb_partial != NULL) { dtv.v_type = VAR_PARTIAL; dtv.vval.v_partial = ch->ch_part[part].ch_callback.cb_partial; set_ref_in_item(&dtv, copyID, ht_stack, list_stack); } } if (ch->ch_callback.cb_partial != NULL) { dtv.v_type = VAR_PARTIAL; dtv.vval.v_partial = ch->ch_callback.cb_partial; set_ref_in_item(&dtv, copyID, ht_stack, list_stack); } if (ch->ch_close_cb.cb_partial != NULL) { dtv.v_type = VAR_PARTIAL; dtv.vval.v_partial = ch->ch_close_cb.cb_partial; set_ref_in_item(&dtv, copyID, ht_stack, list_stack); } } } #endif return abort; } /* * Return a string with the string representation of a variable. * If the memory is allocated "tofree" is set to it, otherwise NULL. * "numbuf" is used for a number. * When "copyID" is not NULL replace recursive lists and dicts with "...". * When both "echo_style" and "composite_val" are FALSE, put quotes around * stings as "string()", otherwise does not put quotes around strings, as * ":echo" displays values. * When "restore_copyID" is FALSE, repeated items in dictionaries and lists * are replaced with "...". * May return NULL. */ char_u * echo_string_core( typval_T *tv, char_u **tofree, char_u *numbuf, int copyID, int echo_style, int restore_copyID, int composite_val) { static int recurse = 0; char_u *r = NULL; if (recurse >= DICT_MAXNEST) { if (!did_echo_string_emsg) { // Only give this message once for a recursive call to avoid // flooding the user with errors. And stop iterating over lists // and dicts. did_echo_string_emsg = TRUE; emsg(_("E724: variable nested too deep for displaying")); } *tofree = NULL; return (char_u *)"{E724}"; } ++recurse; switch (tv->v_type) { case VAR_STRING: if (echo_style && !composite_val) { *tofree = NULL; r = tv->vval.v_string; if (r == NULL) r = (char_u *)""; } else { *tofree = string_quote(tv->vval.v_string, FALSE); r = *tofree; } break; case VAR_FUNC: if (echo_style) { *tofree = NULL; r = tv->vval.v_string; } else { *tofree = string_quote(tv->vval.v_string, TRUE); r = *tofree; } break; case VAR_PARTIAL: { partial_T *pt = tv->vval.v_partial; char_u *fname = string_quote(pt == NULL ? NULL : partial_name(pt), FALSE); garray_T ga; int i; char_u *tf; ga_init2(&ga, 1, 100); ga_concat(&ga, (char_u *)"function("); if (fname != NULL) { ga_concat(&ga, fname); vim_free(fname); } if (pt != NULL && pt->pt_argc > 0) { ga_concat(&ga, (char_u *)", ["); for (i = 0; i < pt->pt_argc; ++i) { if (i > 0) ga_concat(&ga, (char_u *)", "); ga_concat(&ga, tv2string(&pt->pt_argv[i], &tf, numbuf, copyID)); vim_free(tf); } ga_concat(&ga, (char_u *)"]"); } if (pt != NULL && pt->pt_dict != NULL) { typval_T dtv; ga_concat(&ga, (char_u *)", "); dtv.v_type = VAR_DICT; dtv.vval.v_dict = pt->pt_dict; ga_concat(&ga, tv2string(&dtv, &tf, numbuf, copyID)); vim_free(tf); } ga_concat(&ga, (char_u *)")"); *tofree = ga.ga_data; r = *tofree; break; } case VAR_BLOB: r = blob2string(tv->vval.v_blob, tofree, numbuf); break; case VAR_LIST: if (tv->vval.v_list == NULL) { *tofree = NULL; r = NULL; } else if (copyID != 0 && tv->vval.v_list->lv_copyID == copyID && tv->vval.v_list->lv_len > 0) { *tofree = NULL; r = (char_u *)"[...]"; } else { int old_copyID = tv->vval.v_list->lv_copyID; tv->vval.v_list->lv_copyID = copyID; *tofree = list2string(tv, copyID, restore_copyID); if (restore_copyID) tv->vval.v_list->lv_copyID = old_copyID; r = *tofree; } break; case VAR_DICT: if (tv->vval.v_dict == NULL) { *tofree = NULL; r = NULL; } else if (copyID != 0 && tv->vval.v_dict->dv_copyID == copyID && tv->vval.v_dict->dv_hashtab.ht_used != 0) { *tofree = NULL; r = (char_u *)"{...}"; } else { int old_copyID = tv->vval.v_dict->dv_copyID; tv->vval.v_dict->dv_copyID = copyID; *tofree = dict2string(tv, copyID, restore_copyID); if (restore_copyID) tv->vval.v_dict->dv_copyID = old_copyID; r = *tofree; } break; case VAR_NUMBER: case VAR_UNKNOWN: case VAR_VOID: *tofree = NULL; r = tv_get_string_buf(tv, numbuf); break; case VAR_JOB: case VAR_CHANNEL: *tofree = NULL; r = tv_get_string_buf(tv, numbuf); if (composite_val) { *tofree = string_quote(r, FALSE); r = *tofree; } break; case VAR_FLOAT: #ifdef FEAT_FLOAT *tofree = NULL; vim_snprintf((char *)numbuf, NUMBUFLEN, "%g", tv->vval.v_float); r = numbuf; break; #endif case VAR_BOOL: case VAR_SPECIAL: *tofree = NULL; r = (char_u *)get_var_special_name(tv->vval.v_number); break; } if (--recurse == 0) did_echo_string_emsg = FALSE; return r; } /* * Return a string with the string representation of a variable. * If the memory is allocated "tofree" is set to it, otherwise NULL. * "numbuf" is used for a number. * Does not put quotes around strings, as ":echo" displays values. * When "copyID" is not NULL replace recursive lists and dicts with "...". * May return NULL. */ char_u * echo_string( typval_T *tv, char_u **tofree, char_u *numbuf, int copyID) { return echo_string_core(tv, tofree, numbuf, copyID, TRUE, FALSE, FALSE); } /* * Return a string with the string representation of a variable. * If the memory is allocated "tofree" is set to it, otherwise NULL. * "numbuf" is used for a number. * Puts quotes around strings, so that they can be parsed back by eval(). * May return NULL. */ char_u * tv2string( typval_T *tv, char_u **tofree, char_u *numbuf, int copyID) { return echo_string_core(tv, tofree, numbuf, copyID, FALSE, TRUE, FALSE); } /* * Return string "str" in ' quotes, doubling ' characters. * If "str" is NULL an empty string is assumed. * If "function" is TRUE make it function('string'). */ char_u * string_quote(char_u *str, int function) { unsigned len; char_u *p, *r, *s; len = (function ? 13 : 3); if (str != NULL) { len += (unsigned)STRLEN(str); for (p = str; *p != NUL; MB_PTR_ADV(p)) if (*p == '\'') ++len; } s = r = alloc(len); if (r != NULL) { if (function) { STRCPY(r, "function('"); r += 10; } else *r++ = '\''; if (str != NULL) for (p = str; *p != NUL; ) { if (*p == '\'') *r++ = '\''; MB_COPY_CHAR(p, r); } *r++ = '\''; if (function) *r++ = ')'; *r++ = NUL; } return s; } #if defined(FEAT_FLOAT) || defined(PROTO) /* * Convert the string "text" to a floating point number. * This uses strtod(). setlocale(LC_NUMERIC, "C") has been used to make sure * this always uses a decimal point. * Returns the length of the text that was consumed. */ int string2float( char_u *text, float_T *value) // result stored here { char *s = (char *)text; float_T f; // MS-Windows does not deal with "inf" and "nan" properly. if (STRNICMP(text, "inf", 3) == 0) { *value = INFINITY; return 3; } if (STRNICMP(text, "-inf", 3) == 0) { *value = -INFINITY; return 4; } if (STRNICMP(text, "nan", 3) == 0) { *value = NAN; return 3; } f = strtod(s, &s); *value = f; return (int)((char_u *)s - text); } #endif /* * Get the value of an environment variable. * "arg" is pointing to the '$'. It is advanced to after the name. * If the environment variable was not set, silently assume it is empty. * Return FAIL if the name is invalid. */ int get_env_tv(char_u **arg, typval_T *rettv, int evaluate) { char_u *string = NULL; int len; int cc; char_u *name; int mustfree = FALSE; ++*arg; name = *arg; len = get_env_len(arg); if (evaluate) { if (len == 0) return FAIL; // invalid empty name cc = name[len]; name[len] = NUL; // first try vim_getenv(), fast for normal environment vars string = vim_getenv(name, &mustfree); if (string != NULL && *string != NUL) { if (!mustfree) string = vim_strsave(string); } else { if (mustfree) vim_free(string); // next try expanding things like $VIM and ${HOME} string = expand_env_save(name - 1); if (string != NULL && *string == '$') VIM_CLEAR(string); } name[len] = cc; rettv->v_type = VAR_STRING; rettv->vval.v_string = string; } return OK; } /* * Translate a String variable into a position. * Returns NULL when there is an error. */ pos_T * var2fpos( typval_T *varp, int dollar_lnum, // TRUE when $ is last line int *fnum) // set to fnum for '0, 'A, etc. { char_u *name; static pos_T pos; pos_T *pp; // Argument can be [lnum, col, coladd]. if (varp->v_type == VAR_LIST) { list_T *l; int len; int error = FALSE; listitem_T *li; l = varp->vval.v_list; if (l == NULL) return NULL; // Get the line number pos.lnum = list_find_nr(l, 0L, &error); if (error || pos.lnum <= 0 || pos.lnum > curbuf->b_ml.ml_line_count) return NULL; // invalid line number // Get the column number pos.col = list_find_nr(l, 1L, &error); if (error) return NULL; len = (long)STRLEN(ml_get(pos.lnum)); // We accept "$" for the column number: last column. li = list_find(l, 1L); if (li != NULL && li->li_tv.v_type == VAR_STRING && li->li_tv.vval.v_string != NULL && STRCMP(li->li_tv.vval.v_string, "$") == 0) pos.col = len + 1; // Accept a position up to the NUL after the line. if (pos.col == 0 || (int)pos.col > len + 1) return NULL; // invalid column number --pos.col; // Get the virtual offset. Defaults to zero. pos.coladd = list_find_nr(l, 2L, &error); if (error) pos.coladd = 0; return &pos; } name = tv_get_string_chk(varp); if (name == NULL) return NULL; if (name[0] == '.') // cursor return &curwin->w_cursor; if (name[0] == 'v' && name[1] == NUL) // Visual start { if (VIsual_active) return &VIsual; return &curwin->w_cursor; } if (name[0] == '\'') // mark { pp = getmark_buf_fnum(curbuf, name[1], FALSE, fnum); if (pp == NULL || pp == (pos_T *)-1 || pp->lnum <= 0) return NULL; return pp; } pos.coladd = 0; if (name[0] == 'w' && dollar_lnum) { pos.col = 0; if (name[1] == '0') // "w0": first visible line { update_topline(); // In silent Ex mode topline is zero, but that's not a valid line // number; use one instead. pos.lnum = curwin->w_topline > 0 ? curwin->w_topline : 1; return &pos; } else if (name[1] == '$') // "w$": last visible line { validate_botline(); // In silent Ex mode botline is zero, return zero then. pos.lnum = curwin->w_botline > 0 ? curwin->w_botline - 1 : 0; return &pos; } } else if (name[0] == '$') // last column or line { if (dollar_lnum) { pos.lnum = curbuf->b_ml.ml_line_count; pos.col = 0; } else { pos.lnum = curwin->w_cursor.lnum; pos.col = (colnr_T)STRLEN(ml_get_curline()); } return &pos; } return NULL; } /* * Convert list in "arg" into a position and optional file number. * When "fnump" is NULL there is no file number, only 3 items. * Note that the column is passed on as-is, the caller may want to decrement * it to use 1 for the first column. * Return FAIL when conversion is not possible, doesn't check the position for * validity. */ int list2fpos( typval_T *arg, pos_T *posp, int *fnump, colnr_T *curswantp) { list_T *l = arg->vval.v_list; long i = 0; long n; // List must be: [fnum, lnum, col, coladd, curswant], where "fnum" is only // there when "fnump" isn't NULL; "coladd" and "curswant" are optional. if (arg->v_type != VAR_LIST || l == NULL || l->lv_len < (fnump == NULL ? 2 : 3) || l->lv_len > (fnump == NULL ? 4 : 5)) return FAIL; if (fnump != NULL) { n = list_find_nr(l, i++, NULL); // fnum if (n < 0) return FAIL; if (n == 0) n = curbuf->b_fnum; // current buffer *fnump = n; } n = list_find_nr(l, i++, NULL); // lnum if (n < 0) return FAIL; posp->lnum = n; n = list_find_nr(l, i++, NULL); // col if (n < 0) return FAIL; posp->col = n; n = list_find_nr(l, i, NULL); // off if (n < 0) posp->coladd = 0; else posp->coladd = n; if (curswantp != NULL) *curswantp = list_find_nr(l, i + 1, NULL); // curswant return OK; } /* * Get the length of an environment variable name. * Advance "arg" to the first character after the name. * Return 0 for error. */ int get_env_len(char_u **arg) { char_u *p; int len; for (p = *arg; vim_isIDc(*p); ++p) ; if (p == *arg) // no name found return 0; len = (int)(p - *arg); *arg = p; return len; } /* * Get the length of the name of a function or internal variable. * "arg" is advanced to the first non-white character after the name. * Return 0 if something is wrong. */ int get_id_len(char_u **arg) { char_u *p; int len; // Find the end of the name. for (p = *arg; eval_isnamec(*p); ++p) { if (*p == ':') { // "s:" is start of "s:var", but "n:" is not and can be used in // slice "[n:]". Also "xx:" is not a namespace. len = (int)(p - *arg); if ((len == 1 && vim_strchr(NAMESPACE_CHAR, **arg) == NULL) || len > 1) break; } } if (p == *arg) // no name found return 0; len = (int)(p - *arg); *arg = skipwhite(p); return len; } /* * Get the length of the name of a variable or function. * Only the name is recognized, does not handle ".key" or "[idx]". * "arg" is advanced to the first non-white character after the name. * Return -1 if curly braces expansion failed. * Return 0 if something else is wrong. * If the name contains 'magic' {}'s, expand them and return the * expanded name in an allocated string via 'alias' - caller must free. */ int get_name_len( char_u **arg, char_u **alias, int evaluate, int verbose) { int len; char_u *p; char_u *expr_start; char_u *expr_end; *alias = NULL; // default to no alias if ((*arg)[0] == K_SPECIAL && (*arg)[1] == KS_EXTRA && (*arg)[2] == (int)KE_SNR) { // hard coded , already translated *arg += 3; return get_id_len(arg) + 3; } len = eval_fname_script(*arg); if (len > 0) { // literal "", "s:" or "" *arg += len; } /* * Find the end of the name; check for {} construction. */ p = find_name_end(*arg, &expr_start, &expr_end, len > 0 ? 0 : FNE_CHECK_START); if (expr_start != NULL) { char_u *temp_string; if (!evaluate) { len += (int)(p - *arg); *arg = skipwhite(p); return len; } /* * Include any etc in the expanded string: * Thus the -len here. */ temp_string = make_expanded_name(*arg - len, expr_start, expr_end, p); if (temp_string == NULL) return -1; *alias = temp_string; *arg = skipwhite(p); return (int)STRLEN(temp_string); } len += get_id_len(arg); // Only give an error when there is something, otherwise it will be // reported at a higher level. if (len == 0 && verbose && **arg != NUL) semsg(_(e_invexpr2), *arg); return len; } /* * Find the end of a variable or function name, taking care of magic braces. * If "expr_start" is not NULL then "expr_start" and "expr_end" are set to the * start and end of the first magic braces item. * "flags" can have FNE_INCL_BR and FNE_CHECK_START. * Return a pointer to just after the name. Equal to "arg" if there is no * valid name. */ char_u * find_name_end( char_u *arg, char_u **expr_start, char_u **expr_end, int flags) { int mb_nest = 0; int br_nest = 0; char_u *p; int len; if (expr_start != NULL) { *expr_start = NULL; *expr_end = NULL; } // Quick check for valid starting character. if ((flags & FNE_CHECK_START) && !eval_isnamec1(*arg) && *arg != '{') return arg; for (p = arg; *p != NUL && (eval_isnamec(*p) || *p == '{' || ((flags & FNE_INCL_BR) && (*p == '[' || *p == '.')) || mb_nest != 0 || br_nest != 0); MB_PTR_ADV(p)) { if (*p == '\'') { // skip over 'string' to avoid counting [ and ] inside it. for (p = p + 1; *p != NUL && *p != '\''; MB_PTR_ADV(p)) ; if (*p == NUL) break; } else if (*p == '"') { // skip over "str\"ing" to avoid counting [ and ] inside it. for (p = p + 1; *p != NUL && *p != '"'; MB_PTR_ADV(p)) if (*p == '\\' && p[1] != NUL) ++p; if (*p == NUL) break; } else if (br_nest == 0 && mb_nest == 0 && *p == ':') { // "s:" is start of "s:var", but "n:" is not and can be used in // slice "[n:]". Also "xx:" is not a namespace. But {ns}: is. len = (int)(p - arg); if ((len == 1 && vim_strchr(NAMESPACE_CHAR, *arg) == NULL) || (len > 1 && p[-1] != '}')) break; } if (mb_nest == 0) { if (*p == '[') ++br_nest; else if (*p == ']') --br_nest; } if (br_nest == 0) { if (*p == '{') { mb_nest++; if (expr_start != NULL && *expr_start == NULL) *expr_start = p; } else if (*p == '}') { mb_nest--; if (expr_start != NULL && mb_nest == 0 && *expr_end == NULL) *expr_end = p; } } } return p; } /* * Expands out the 'magic' {}'s in a variable/function name. * Note that this can call itself recursively, to deal with * constructs like foo{bar}{baz}{bam} * The four pointer arguments point to "foo{expre}ss{ion}bar" * "in_start" ^ * "expr_start" ^ * "expr_end" ^ * "in_end" ^ * * Returns a new allocated string, which the caller must free. * Returns NULL for failure. */ static char_u * make_expanded_name( char_u *in_start, char_u *expr_start, char_u *expr_end, char_u *in_end) { char_u c1; char_u *retval = NULL; char_u *temp_result; char_u *nextcmd = NULL; if (expr_end == NULL || in_end == NULL) return NULL; *expr_start = NUL; *expr_end = NUL; c1 = *in_end; *in_end = NUL; temp_result = eval_to_string(expr_start + 1, &nextcmd, FALSE); if (temp_result != NULL && nextcmd == NULL) { retval = alloc(STRLEN(temp_result) + (expr_start - in_start) + (in_end - expr_end) + 1); if (retval != NULL) { STRCPY(retval, in_start); STRCAT(retval, temp_result); STRCAT(retval, expr_end + 1); } } vim_free(temp_result); *in_end = c1; // put char back for error messages *expr_start = '{'; *expr_end = '}'; if (retval != NULL) { temp_result = find_name_end(retval, &expr_start, &expr_end, 0); if (expr_start != NULL) { // Further expansion! temp_result = make_expanded_name(retval, expr_start, expr_end, temp_result); vim_free(retval); retval = temp_result; } } return retval; } /* * Return TRUE if character "c" can be used in a variable or function name. * Does not include '{' or '}' for magic braces. */ int eval_isnamec(int c) { return (ASCII_ISALNUM(c) || c == '_' || c == ':' || c == AUTOLOAD_CHAR); } /* * Return TRUE if character "c" can be used as the first character in a * variable or function name (excluding '{' and '}'). */ int eval_isnamec1(int c) { return (ASCII_ISALPHA(c) || c == '_'); } /* * Handle: * - expr[expr], expr[expr:expr] subscript * - ".name" lookup * - function call with Funcref variable: func(expr) * - method call: var->method() * * Can all be combined in any order: dict.func(expr)[idx]['func'](expr)->len() */ int handle_subscript( char_u **arg, typval_T *rettv, int evaluate, // do more than finding the end int verbose, // give error messages char_u *start_leader, // start of '!' and '-' prefixes char_u **end_leaderp) // end of '!' and '-' prefixes { int ret = OK; dict_T *selfdict = NULL; // "." is ".name" lookup when we found a dict or when evaluating and // scriptversion is at least 2, where string concatenation is "..". while (ret == OK && (((**arg == '[' || (**arg == '.' && (rettv->v_type == VAR_DICT || (!evaluate && (*arg)[1] != '.' && current_sctx.sc_version >= 2))) || (**arg == '(' && (!evaluate || rettv->v_type == VAR_FUNC || rettv->v_type == VAR_PARTIAL))) && !VIM_ISWHITE(*(*arg - 1))) || (**arg == '-' && (*arg)[1] == '>'))) { if (**arg == '(') { ret = call_func_rettv(arg, rettv, evaluate, selfdict, NULL); // Stop the expression evaluation when immediately aborting on // error, or when an interrupt occurred or an exception was thrown // but not caught. if (aborting()) { if (ret == OK) clear_tv(rettv); ret = FAIL; } dict_unref(selfdict); selfdict = NULL; } else if (**arg == '-') { // Expression "-1.0->method()" applies the leader "-" before // applying ->. if (evaluate && *end_leaderp > start_leader) ret = eval7_leader(rettv, start_leader, end_leaderp); if (ret == OK) { if ((*arg)[2] == '{') // expr->{lambda}() ret = eval_lambda(arg, rettv, evaluate, verbose); else // expr->name() ret = eval_method(arg, rettv, evaluate, verbose); } } else // **arg == '[' || **arg == '.' { dict_unref(selfdict); if (rettv->v_type == VAR_DICT) { selfdict = rettv->vval.v_dict; if (selfdict != NULL) ++selfdict->dv_refcount; } else selfdict = NULL; if (eval_index(arg, rettv, evaluate, verbose) == FAIL) { clear_tv(rettv); ret = FAIL; } } } // Turn "dict.Func" into a partial for "Func" bound to "dict". // Don't do this when "Func" is already a partial that was bound // explicitly (pt_auto is FALSE). if (selfdict != NULL && (rettv->v_type == VAR_FUNC || (rettv->v_type == VAR_PARTIAL && (rettv->vval.v_partial->pt_auto || rettv->vval.v_partial->pt_dict == NULL)))) selfdict = make_partial(selfdict, rettv); dict_unref(selfdict); return ret; } /* * Allocate memory for a variable type-value, and make it empty (0 or NULL * value). */ typval_T * alloc_tv(void) { return ALLOC_CLEAR_ONE(typval_T); } /* * Allocate memory for a variable type-value, and assign a string to it. * The string "s" must have been allocated, it is consumed. * Return NULL for out of memory, the variable otherwise. */ typval_T * alloc_string_tv(char_u *s) { typval_T *rettv; rettv = alloc_tv(); if (rettv != NULL) { rettv->v_type = VAR_STRING; rettv->vval.v_string = s; } else vim_free(s); return rettv; } /* * Free the memory for a variable type-value. */ void free_tv(typval_T *varp) { if (varp != NULL) { switch (varp->v_type) { case VAR_FUNC: func_unref(varp->vval.v_string); // FALLTHROUGH case VAR_STRING: vim_free(varp->vval.v_string); break; case VAR_PARTIAL: partial_unref(varp->vval.v_partial); break; case VAR_BLOB: blob_unref(varp->vval.v_blob); break; case VAR_LIST: list_unref(varp->vval.v_list); break; case VAR_DICT: dict_unref(varp->vval.v_dict); break; case VAR_JOB: #ifdef FEAT_JOB_CHANNEL job_unref(varp->vval.v_job); break; #endif case VAR_CHANNEL: #ifdef FEAT_JOB_CHANNEL channel_unref(varp->vval.v_channel); break; #endif case VAR_NUMBER: case VAR_FLOAT: case VAR_UNKNOWN: case VAR_VOID: case VAR_BOOL: case VAR_SPECIAL: break; } vim_free(varp); } } /* * Free the memory for a variable value and set the value to NULL or 0. */ void clear_tv(typval_T *varp) { if (varp != NULL) { switch (varp->v_type) { case VAR_FUNC: func_unref(varp->vval.v_string); // FALLTHROUGH case VAR_STRING: VIM_CLEAR(varp->vval.v_string); break; case VAR_PARTIAL: partial_unref(varp->vval.v_partial); varp->vval.v_partial = NULL; break; case VAR_BLOB: blob_unref(varp->vval.v_blob); varp->vval.v_blob = NULL; break; case VAR_LIST: list_unref(varp->vval.v_list); varp->vval.v_list = NULL; break; case VAR_DICT: dict_unref(varp->vval.v_dict); varp->vval.v_dict = NULL; break; case VAR_NUMBER: case VAR_BOOL: case VAR_SPECIAL: varp->vval.v_number = 0; break; case VAR_FLOAT: #ifdef FEAT_FLOAT varp->vval.v_float = 0.0; break; #endif case VAR_JOB: #ifdef FEAT_JOB_CHANNEL job_unref(varp->vval.v_job); varp->vval.v_job = NULL; #endif break; case VAR_CHANNEL: #ifdef FEAT_JOB_CHANNEL channel_unref(varp->vval.v_channel); varp->vval.v_channel = NULL; #endif case VAR_UNKNOWN: case VAR_VOID: break; } varp->v_lock = 0; } } /* * Set the value of a variable to NULL without freeing items. */ void init_tv(typval_T *varp) { if (varp != NULL) vim_memset(varp, 0, sizeof(typval_T)); } /* * Get the number value of a variable. * If it is a String variable, uses vim_str2nr(). * For incompatible types, return 0. * tv_get_number_chk() is similar to tv_get_number(), but informs the * caller of incompatible types: it sets *denote to TRUE if "denote" * is not NULL or returns -1 otherwise. */ varnumber_T tv_get_number(typval_T *varp) { int error = FALSE; return tv_get_number_chk(varp, &error); // return 0L on error } varnumber_T tv_get_number_chk(typval_T *varp, int *denote) { varnumber_T n = 0L; switch (varp->v_type) { case VAR_NUMBER: return varp->vval.v_number; case VAR_FLOAT: #ifdef FEAT_FLOAT emsg(_("E805: Using a Float as a Number")); break; #endif case VAR_FUNC: case VAR_PARTIAL: emsg(_("E703: Using a Funcref as a Number")); break; case VAR_STRING: if (varp->vval.v_string != NULL) vim_str2nr(varp->vval.v_string, NULL, NULL, STR2NR_ALL, &n, NULL, 0, FALSE); return n; case VAR_LIST: emsg(_("E745: Using a List as a Number")); break; case VAR_DICT: emsg(_("E728: Using a Dictionary as a Number")); break; case VAR_BOOL: case VAR_SPECIAL: return varp->vval.v_number == VVAL_TRUE ? 1 : 0; case VAR_JOB: #ifdef FEAT_JOB_CHANNEL emsg(_("E910: Using a Job as a Number")); break; #endif case VAR_CHANNEL: #ifdef FEAT_JOB_CHANNEL emsg(_("E913: Using a Channel as a Number")); break; #endif case VAR_BLOB: emsg(_("E974: Using a Blob as a Number")); break; case VAR_UNKNOWN: case VAR_VOID: internal_error_no_abort("tv_get_number(UNKNOWN)"); break; } if (denote == NULL) // useful for values that must be unsigned n = -1; else *denote = TRUE; return n; } #ifdef FEAT_FLOAT float_T tv_get_float(typval_T *varp) { switch (varp->v_type) { case VAR_NUMBER: return (float_T)(varp->vval.v_number); case VAR_FLOAT: return varp->vval.v_float; case VAR_FUNC: case VAR_PARTIAL: emsg(_("E891: Using a Funcref as a Float")); break; case VAR_STRING: emsg(_("E892: Using a String as a Float")); break; case VAR_LIST: emsg(_("E893: Using a List as a Float")); break; case VAR_DICT: emsg(_("E894: Using a Dictionary as a Float")); break; case VAR_BOOL: emsg(_("E362: Using a boolean value as a Float")); break; case VAR_SPECIAL: emsg(_("E907: Using a special value as a Float")); break; case VAR_JOB: # ifdef FEAT_JOB_CHANNEL emsg(_("E911: Using a Job as a Float")); break; # endif case VAR_CHANNEL: # ifdef FEAT_JOB_CHANNEL emsg(_("E914: Using a Channel as a Float")); break; # endif case VAR_BLOB: emsg(_("E975: Using a Blob as a Float")); break; case VAR_UNKNOWN: case VAR_VOID: internal_error_no_abort("tv_get_float(UNKNOWN)"); break; } return 0; } #endif /* * Get the string value of a variable. * If it is a Number variable, the number is converted into a string. * tv_get_string() uses a single, static buffer. YOU CAN ONLY USE IT ONCE! * tv_get_string_buf() uses a given buffer. * If the String variable has never been set, return an empty string. * Never returns NULL; * tv_get_string_chk() and tv_get_string_buf_chk() are similar, but return * NULL on error. */ char_u * tv_get_string(typval_T *varp) { static char_u mybuf[NUMBUFLEN]; return tv_get_string_buf(varp, mybuf); } char_u * tv_get_string_buf(typval_T *varp, char_u *buf) { char_u *res = tv_get_string_buf_chk(varp, buf); return res != NULL ? res : (char_u *)""; } /* * Careful: This uses a single, static buffer. YOU CAN ONLY USE IT ONCE! */ char_u * tv_get_string_chk(typval_T *varp) { static char_u mybuf[NUMBUFLEN]; return tv_get_string_buf_chk(varp, mybuf); } char_u * tv_get_string_buf_chk(typval_T *varp, char_u *buf) { switch (varp->v_type) { case VAR_NUMBER: vim_snprintf((char *)buf, NUMBUFLEN, "%lld", (varnumber_T)varp->vval.v_number); return buf; case VAR_FUNC: case VAR_PARTIAL: emsg(_("E729: using Funcref as a String")); break; case VAR_LIST: emsg(_("E730: using List as a String")); break; case VAR_DICT: emsg(_("E731: using Dictionary as a String")); break; case VAR_FLOAT: #ifdef FEAT_FLOAT emsg(_(e_float_as_string)); break; #endif case VAR_STRING: if (varp->vval.v_string != NULL) return varp->vval.v_string; return (char_u *)""; case VAR_BOOL: case VAR_SPECIAL: STRCPY(buf, get_var_special_name(varp->vval.v_number)); return buf; case VAR_BLOB: emsg(_("E976: using Blob as a String")); break; case VAR_JOB: #ifdef FEAT_JOB_CHANNEL { job_T *job = varp->vval.v_job; char *status; if (job == NULL) return (char_u *)"no process"; status = job->jv_status == JOB_FAILED ? "fail" : job->jv_status >= JOB_ENDED ? "dead" : "run"; # ifdef UNIX vim_snprintf((char *)buf, NUMBUFLEN, "process %ld %s", (long)job->jv_pid, status); # elif defined(MSWIN) vim_snprintf((char *)buf, NUMBUFLEN, "process %ld %s", (long)job->jv_proc_info.dwProcessId, status); # else // fall-back vim_snprintf((char *)buf, NUMBUFLEN, "process ? %s", status); # endif return buf; } #endif break; case VAR_CHANNEL: #ifdef FEAT_JOB_CHANNEL { channel_T *channel = varp->vval.v_channel; char *status = channel_status(channel, -1); if (channel == NULL) vim_snprintf((char *)buf, NUMBUFLEN, "channel %s", status); else vim_snprintf((char *)buf, NUMBUFLEN, "channel %d %s", channel->ch_id, status); return buf; } #endif break; case VAR_UNKNOWN: case VAR_VOID: emsg(_(e_inval_string)); break; } return NULL; } /* * Turn a typeval into a string. Similar to tv_get_string_buf() but uses * string() on Dict, List, etc. */ static char_u * tv_stringify(typval_T *varp, char_u *buf) { if (varp->v_type == VAR_LIST || varp->v_type == VAR_DICT || varp->v_type == VAR_BLOB || varp->v_type == VAR_FUNC || varp->v_type == VAR_PARTIAL || varp->v_type == VAR_FLOAT) { typval_T tmp; f_string(varp, &tmp); tv_get_string_buf(&tmp, buf); clear_tv(varp); *varp = tmp; return tmp.vval.v_string; } return tv_get_string_buf(varp, buf); } /* * Return TRUE if typeval "tv" and its value are set to be locked (immutable). * Also give an error message, using "name" or _("name") when use_gettext is * TRUE. */ static int tv_check_lock(typval_T *tv, char_u *name, int use_gettext) { int lock = 0; switch (tv->v_type) { case VAR_BLOB: if (tv->vval.v_blob != NULL) lock = tv->vval.v_blob->bv_lock; break; case VAR_LIST: if (tv->vval.v_list != NULL) lock = tv->vval.v_list->lv_lock; break; case VAR_DICT: if (tv->vval.v_dict != NULL) lock = tv->vval.v_dict->dv_lock; break; default: break; } return var_check_lock(tv->v_lock, name, use_gettext) || (lock != 0 && var_check_lock(lock, name, use_gettext)); } /* * Copy the values from typval_T "from" to typval_T "to". * When needed allocates string or increases reference count. * Does not make a copy of a list, blob or dict but copies the reference! * It is OK for "from" and "to" to point to the same item. This is used to * make a copy later. */ void copy_tv(typval_T *from, typval_T *to) { to->v_type = from->v_type; to->v_lock = 0; switch (from->v_type) { case VAR_NUMBER: case VAR_BOOL: case VAR_SPECIAL: to->vval.v_number = from->vval.v_number; break; case VAR_FLOAT: #ifdef FEAT_FLOAT to->vval.v_float = from->vval.v_float; break; #endif case VAR_JOB: #ifdef FEAT_JOB_CHANNEL to->vval.v_job = from->vval.v_job; if (to->vval.v_job != NULL) ++to->vval.v_job->jv_refcount; break; #endif case VAR_CHANNEL: #ifdef FEAT_JOB_CHANNEL to->vval.v_channel = from->vval.v_channel; if (to->vval.v_channel != NULL) ++to->vval.v_channel->ch_refcount; break; #endif case VAR_STRING: case VAR_FUNC: if (from->vval.v_string == NULL) to->vval.v_string = NULL; else { to->vval.v_string = vim_strsave(from->vval.v_string); if (from->v_type == VAR_FUNC) func_ref(to->vval.v_string); } break; case VAR_PARTIAL: if (from->vval.v_partial == NULL) to->vval.v_partial = NULL; else { to->vval.v_partial = from->vval.v_partial; ++to->vval.v_partial->pt_refcount; } break; case VAR_BLOB: if (from->vval.v_blob == NULL) to->vval.v_blob = NULL; else { to->vval.v_blob = from->vval.v_blob; ++to->vval.v_blob->bv_refcount; } break; case VAR_LIST: if (from->vval.v_list == NULL) to->vval.v_list = NULL; else { to->vval.v_list = from->vval.v_list; ++to->vval.v_list->lv_refcount; } break; case VAR_DICT: if (from->vval.v_dict == NULL) to->vval.v_dict = NULL; else { to->vval.v_dict = from->vval.v_dict; ++to->vval.v_dict->dv_refcount; } break; case VAR_UNKNOWN: case VAR_VOID: internal_error_no_abort("copy_tv(UNKNOWN)"); break; } } /* * Make a copy of an item. * Lists and Dictionaries are also copied. A deep copy if "deep" is set. * For deepcopy() "copyID" is zero for a full copy or the ID for when a * reference to an already copied list/dict can be used. * Returns FAIL or OK. */ int item_copy( typval_T *from, typval_T *to, int deep, int copyID) { static int recurse = 0; int ret = OK; if (recurse >= DICT_MAXNEST) { emsg(_("E698: variable nested too deep for making a copy")); return FAIL; } ++recurse; switch (from->v_type) { case VAR_NUMBER: case VAR_FLOAT: case VAR_STRING: case VAR_FUNC: case VAR_PARTIAL: case VAR_BOOL: case VAR_SPECIAL: case VAR_JOB: case VAR_CHANNEL: copy_tv(from, to); break; case VAR_LIST: to->v_type = VAR_LIST; to->v_lock = 0; if (from->vval.v_list == NULL) to->vval.v_list = NULL; else if (copyID != 0 && from->vval.v_list->lv_copyID == copyID) { // use the copy made earlier to->vval.v_list = from->vval.v_list->lv_copylist; ++to->vval.v_list->lv_refcount; } else to->vval.v_list = list_copy(from->vval.v_list, deep, copyID); if (to->vval.v_list == NULL) ret = FAIL; break; case VAR_BLOB: ret = blob_copy(from->vval.v_blob, to); break; case VAR_DICT: to->v_type = VAR_DICT; to->v_lock = 0; if (from->vval.v_dict == NULL) to->vval.v_dict = NULL; else if (copyID != 0 && from->vval.v_dict->dv_copyID == copyID) { // use the copy made earlier to->vval.v_dict = from->vval.v_dict->dv_copydict; ++to->vval.v_dict->dv_refcount; } else to->vval.v_dict = dict_copy(from->vval.v_dict, deep, copyID); if (to->vval.v_dict == NULL) ret = FAIL; break; case VAR_UNKNOWN: case VAR_VOID: internal_error_no_abort("item_copy(UNKNOWN)"); ret = FAIL; } --recurse; return ret; } void echo_one(typval_T *rettv, int with_space, int *atstart, int *needclr) { char_u *tofree; char_u numbuf[NUMBUFLEN]; char_u *p = echo_string(rettv, &tofree, numbuf, get_copyID()); if (*atstart) { *atstart = FALSE; // Call msg_start() after eval1(), evaluating the expression // may cause a message to appear. if (with_space) { // Mark the saved text as finishing the line, so that what // follows is displayed on a new line when scrolling back // at the more prompt. msg_sb_eol(); msg_start(); } } else if (with_space) msg_puts_attr(" ", echo_attr); if (p != NULL) for ( ; *p != NUL && !got_int; ++p) { if (*p == '\n' || *p == '\r' || *p == TAB) { if (*p != TAB && *needclr) { // remove any text still there from the command msg_clr_eos(); *needclr = FALSE; } msg_putchar_attr(*p, echo_attr); } else { if (has_mbyte) { int i = (*mb_ptr2len)(p); (void)msg_outtrans_len_attr(p, i, echo_attr); p += i - 1; } else (void)msg_outtrans_len_attr(p, 1, echo_attr); } } vim_free(tofree); } /* * ":echo expr1 ..." print each argument separated with a space, add a * newline at the end. * ":echon expr1 ..." print each argument plain. */ void ex_echo(exarg_T *eap) { char_u *arg = eap->arg; typval_T rettv; char_u *p; int needclr = TRUE; int atstart = TRUE; int did_emsg_before = did_emsg; int called_emsg_before = called_emsg; if (eap->skip) ++emsg_skip; while (*arg != NUL && *arg != '|' && *arg != '\n' && !got_int) { // If eval1() causes an error message the text from the command may // still need to be cleared. E.g., "echo 22,44". need_clr_eos = needclr; p = arg; if (eval1(&arg, &rettv, !eap->skip) == FAIL) { /* * Report the invalid expression unless the expression evaluation * has been cancelled due to an aborting error, an interrupt, or an * exception. */ if (!aborting() && did_emsg == did_emsg_before && called_emsg == called_emsg_before) semsg(_(e_invexpr2), p); need_clr_eos = FALSE; break; } need_clr_eos = FALSE; if (!eap->skip) echo_one(&rettv, eap->cmdidx == CMD_echo, &atstart, &needclr); clear_tv(&rettv); arg = skipwhite(arg); } eap->nextcmd = check_nextcmd(arg); if (eap->skip) --emsg_skip; else { // remove text that may still be there from the command if (needclr) msg_clr_eos(); if (eap->cmdidx == CMD_echo) msg_end(); } } /* * ":echohl {name}". */ void ex_echohl(exarg_T *eap) { echo_attr = syn_name2attr(eap->arg); } /* * Returns the :echo attribute */ int get_echo_attr(void) { return echo_attr; } /* * ":execute expr1 ..." execute the result of an expression. * ":echomsg expr1 ..." Print a message * ":echoerr expr1 ..." Print an error * Each gets spaces around each argument and a newline at the end for * echo commands */ void ex_execute(exarg_T *eap) { char_u *arg = eap->arg; typval_T rettv; int ret = OK; char_u *p; garray_T ga; int len; int save_did_emsg; ga_init2(&ga, 1, 80); if (eap->skip) ++emsg_skip; while (*arg != NUL && *arg != '|' && *arg != '\n') { ret = eval1_emsg(&arg, &rettv, !eap->skip); if (ret == FAIL) break; if (!eap->skip) { char_u buf[NUMBUFLEN]; if (eap->cmdidx == CMD_execute) { if (rettv.v_type == VAR_CHANNEL || rettv.v_type == VAR_JOB) { emsg(_(e_inval_string)); p = NULL; } else p = tv_get_string_buf(&rettv, buf); } else p = tv_stringify(&rettv, buf); if (p == NULL) { clear_tv(&rettv); ret = FAIL; break; } len = (int)STRLEN(p); if (ga_grow(&ga, len + 2) == FAIL) { clear_tv(&rettv); ret = FAIL; break; } if (ga.ga_len) ((char_u *)(ga.ga_data))[ga.ga_len++] = ' '; STRCPY((char_u *)(ga.ga_data) + ga.ga_len, p); ga.ga_len += len; } clear_tv(&rettv); arg = skipwhite(arg); } if (ret != FAIL && ga.ga_data != NULL) { if (eap->cmdidx == CMD_echomsg || eap->cmdidx == CMD_echoerr) { // Mark the already saved text as finishing the line, so that what // follows is displayed on a new line when scrolling back at the // more prompt. msg_sb_eol(); } if (eap->cmdidx == CMD_echomsg) { msg_attr(ga.ga_data, echo_attr); out_flush(); } else if (eap->cmdidx == CMD_echoerr) { // We don't want to abort following commands, restore did_emsg. save_did_emsg = did_emsg; emsg(ga.ga_data); if (!force_abort) did_emsg = save_did_emsg; } else if (eap->cmdidx == CMD_execute) do_cmdline((char_u *)ga.ga_data, eap->getline, eap->cookie, DOCMD_NOWAIT|DOCMD_VERBOSE); } ga_clear(&ga); if (eap->skip) --emsg_skip; eap->nextcmd = check_nextcmd(arg); } /* * Skip over the name of an option: "&option", "&g:option" or "&l:option". * "arg" points to the "&" or '+' when called, to "option" when returning. * Returns NULL when no option name found. Otherwise pointer to the char * after the option name. */ char_u * find_option_end(char_u **arg, int *opt_flags) { char_u *p = *arg; ++p; if (*p == 'g' && p[1] == ':') { *opt_flags = OPT_GLOBAL; p += 2; } else if (*p == 'l' && p[1] == ':') { *opt_flags = OPT_LOCAL; p += 2; } else *opt_flags = 0; if (!ASCII_ISALPHA(*p)) return NULL; *arg = p; if (p[0] == 't' && p[1] == '_' && p[2] != NUL && p[3] != NUL) p += 4; // termcap option else while (ASCII_ISALPHA(*p)) ++p; return p; } /* * Display script name where an item was last set. * Should only be invoked when 'verbose' is non-zero. */ void last_set_msg(sctx_T script_ctx) { char_u *p; if (script_ctx.sc_sid != 0) { p = home_replace_save(NULL, get_scriptname(script_ctx.sc_sid)); if (p != NULL) { verbose_enter(); msg_puts(_("\n\tLast set from ")); msg_puts((char *)p); if (script_ctx.sc_lnum > 0) { msg_puts(_(line_msg)); msg_outnum((long)script_ctx.sc_lnum); } verbose_leave(); vim_free(p); } } } /* * Compare "typ1" and "typ2". Put the result in "typ1". */ int typval_compare( typval_T *typ1, // first operand typval_T *typ2, // second operand exptype_T type, // operator int ic) // ignore case { int i; varnumber_T n1, n2; char_u *s1, *s2; char_u buf1[NUMBUFLEN], buf2[NUMBUFLEN]; int type_is = type == EXPR_IS || type == EXPR_ISNOT; if (type_is && typ1->v_type != typ2->v_type) { // For "is" a different type always means FALSE, for "notis" // it means TRUE. n1 = (type == EXPR_ISNOT); } else if (typ1->v_type == VAR_BLOB || typ2->v_type == VAR_BLOB) { if (type_is) { n1 = (typ1->v_type == typ2->v_type && typ1->vval.v_blob == typ2->vval.v_blob); if (type == EXPR_ISNOT) n1 = !n1; } else if (typ1->v_type != typ2->v_type || (type != EXPR_EQUAL && type != EXPR_NEQUAL)) { if (typ1->v_type != typ2->v_type) emsg(_("E977: Can only compare Blob with Blob")); else emsg(_(e_invalblob)); clear_tv(typ1); return FAIL; } else { // Compare two Blobs for being equal or unequal. n1 = blob_equal(typ1->vval.v_blob, typ2->vval.v_blob); if (type == EXPR_NEQUAL) n1 = !n1; } } else if (typ1->v_type == VAR_LIST || typ2->v_type == VAR_LIST) { if (type_is) { n1 = (typ1->v_type == typ2->v_type && typ1->vval.v_list == typ2->vval.v_list); if (type == EXPR_ISNOT) n1 = !n1; } else if (typ1->v_type != typ2->v_type || (type != EXPR_EQUAL && type != EXPR_NEQUAL)) { if (typ1->v_type != typ2->v_type) emsg(_("E691: Can only compare List with List")); else emsg(_("E692: Invalid operation for List")); clear_tv(typ1); return FAIL; } else { // Compare two Lists for being equal or unequal. n1 = list_equal(typ1->vval.v_list, typ2->vval.v_list, ic, FALSE); if (type == EXPR_NEQUAL) n1 = !n1; } } else if (typ1->v_type == VAR_DICT || typ2->v_type == VAR_DICT) { if (type_is) { n1 = (typ1->v_type == typ2->v_type && typ1->vval.v_dict == typ2->vval.v_dict); if (type == EXPR_ISNOT) n1 = !n1; } else if (typ1->v_type != typ2->v_type || (type != EXPR_EQUAL && type != EXPR_NEQUAL)) { if (typ1->v_type != typ2->v_type) emsg(_("E735: Can only compare Dictionary with Dictionary")); else emsg(_("E736: Invalid operation for Dictionary")); clear_tv(typ1); return FAIL; } else { // Compare two Dictionaries for being equal or unequal. n1 = dict_equal(typ1->vval.v_dict, typ2->vval.v_dict, ic, FALSE); if (type == EXPR_NEQUAL) n1 = !n1; } } else if (typ1->v_type == VAR_FUNC || typ2->v_type == VAR_FUNC || typ1->v_type == VAR_PARTIAL || typ2->v_type == VAR_PARTIAL) { if (type != EXPR_EQUAL && type != EXPR_NEQUAL && type != EXPR_IS && type != EXPR_ISNOT) { emsg(_("E694: Invalid operation for Funcrefs")); clear_tv(typ1); return FAIL; } if ((typ1->v_type == VAR_PARTIAL && typ1->vval.v_partial == NULL) || (typ2->v_type == VAR_PARTIAL && typ2->vval.v_partial == NULL)) // when a partial is NULL assume not equal n1 = FALSE; else if (type_is) { if (typ1->v_type == VAR_FUNC && typ2->v_type == VAR_FUNC) // strings are considered the same if their value is // the same n1 = tv_equal(typ1, typ2, ic, FALSE); else if (typ1->v_type == VAR_PARTIAL && typ2->v_type == VAR_PARTIAL) n1 = (typ1->vval.v_partial == typ2->vval.v_partial); else n1 = FALSE; } else n1 = tv_equal(typ1, typ2, ic, FALSE); if (type == EXPR_NEQUAL || type == EXPR_ISNOT) n1 = !n1; } #ifdef FEAT_FLOAT /* * If one of the two variables is a float, compare as a float. * When using "=~" or "!~", always compare as string. */ else if ((typ1->v_type == VAR_FLOAT || typ2->v_type == VAR_FLOAT) && type != EXPR_MATCH && type != EXPR_NOMATCH) { float_T f1, f2; f1 = tv_get_float(typ1); f2 = tv_get_float(typ2); n1 = FALSE; switch (type) { case EXPR_IS: case EXPR_EQUAL: n1 = (f1 == f2); break; case EXPR_ISNOT: case EXPR_NEQUAL: n1 = (f1 != f2); break; case EXPR_GREATER: n1 = (f1 > f2); break; case EXPR_GEQUAL: n1 = (f1 >= f2); break; case EXPR_SMALLER: n1 = (f1 < f2); break; case EXPR_SEQUAL: n1 = (f1 <= f2); break; case EXPR_UNKNOWN: case EXPR_MATCH: default: break; // avoid gcc warning } } #endif /* * If one of the two variables is a number, compare as a number. * When using "=~" or "!~", always compare as string. */ else if ((typ1->v_type == VAR_NUMBER || typ2->v_type == VAR_NUMBER) && type != EXPR_MATCH && type != EXPR_NOMATCH) { n1 = tv_get_number(typ1); n2 = tv_get_number(typ2); switch (type) { case EXPR_IS: case EXPR_EQUAL: n1 = (n1 == n2); break; case EXPR_ISNOT: case EXPR_NEQUAL: n1 = (n1 != n2); break; case EXPR_GREATER: n1 = (n1 > n2); break; case EXPR_GEQUAL: n1 = (n1 >= n2); break; case EXPR_SMALLER: n1 = (n1 < n2); break; case EXPR_SEQUAL: n1 = (n1 <= n2); break; case EXPR_UNKNOWN: case EXPR_MATCH: default: break; // avoid gcc warning } } else { s1 = tv_get_string_buf(typ1, buf1); s2 = tv_get_string_buf(typ2, buf2); if (type != EXPR_MATCH && type != EXPR_NOMATCH) i = ic ? MB_STRICMP(s1, s2) : STRCMP(s1, s2); else i = 0; n1 = FALSE; switch (type) { case EXPR_IS: case EXPR_EQUAL: n1 = (i == 0); break; case EXPR_ISNOT: case EXPR_NEQUAL: n1 = (i != 0); break; case EXPR_GREATER: n1 = (i > 0); break; case EXPR_GEQUAL: n1 = (i >= 0); break; case EXPR_SMALLER: n1 = (i < 0); break; case EXPR_SEQUAL: n1 = (i <= 0); break; case EXPR_MATCH: case EXPR_NOMATCH: n1 = pattern_match(s2, s1, ic); if (type == EXPR_NOMATCH) n1 = !n1; break; default: break; // avoid gcc warning } } clear_tv(typ1); typ1->v_type = VAR_NUMBER; typ1->vval.v_number = n1; return OK; } char_u * typval_tostring(typval_T *arg) { char_u *tofree; char_u numbuf[NUMBUFLEN]; char_u *ret = NULL; if (arg == NULL) return vim_strsave((char_u *)"(does not exist)"); ret = tv2string(arg, &tofree, numbuf, 0); // Make a copy if we have a value but it's not in allocated memory. if (ret != NULL && tofree == NULL) ret = vim_strsave(ret); return ret; } #endif // FEAT_EVAL /* * Perform a substitution on "str" with pattern "pat" and substitute "sub". * When "sub" is NULL "expr" is used, must be a VAR_FUNC or VAR_PARTIAL. * "flags" can be "g" to do a global substitute. * Returns an allocated string, NULL for error. */ char_u * do_string_sub( char_u *str, char_u *pat, char_u *sub, typval_T *expr, char_u *flags) { int sublen; regmatch_T regmatch; int i; int do_all; char_u *tail; char_u *end; garray_T ga; char_u *ret; char_u *save_cpo; char_u *zero_width = NULL; // Make 'cpoptions' empty, so that the 'l' flag doesn't work here save_cpo = p_cpo; p_cpo = empty_option; ga_init2(&ga, 1, 200); do_all = (flags[0] == 'g'); regmatch.rm_ic = p_ic; regmatch.regprog = vim_regcomp(pat, RE_MAGIC + RE_STRING); if (regmatch.regprog != NULL) { tail = str; end = str + STRLEN(str); while (vim_regexec_nl(®match, str, (colnr_T)(tail - str))) { // Skip empty match except for first match. if (regmatch.startp[0] == regmatch.endp[0]) { if (zero_width == regmatch.startp[0]) { // avoid getting stuck on a match with an empty string i = mb_ptr2len(tail); mch_memmove((char_u *)ga.ga_data + ga.ga_len, tail, (size_t)i); ga.ga_len += i; tail += i; continue; } zero_width = regmatch.startp[0]; } /* * Get some space for a temporary buffer to do the substitution * into. It will contain: * - The text up to where the match is. * - The substituted text. * - The text after the match. */ sublen = vim_regsub(®match, sub, expr, tail, FALSE, TRUE, FALSE); if (ga_grow(&ga, (int)((end - tail) + sublen - (regmatch.endp[0] - regmatch.startp[0]))) == FAIL) { ga_clear(&ga); break; } // copy the text up to where the match is i = (int)(regmatch.startp[0] - tail); mch_memmove((char_u *)ga.ga_data + ga.ga_len, tail, (size_t)i); // add the substituted text (void)vim_regsub(®match, sub, expr, (char_u *)ga.ga_data + ga.ga_len + i, TRUE, TRUE, FALSE); ga.ga_len += i + sublen - 1; tail = regmatch.endp[0]; if (*tail == NUL) break; if (!do_all) break; } if (ga.ga_data != NULL) STRCPY((char *)ga.ga_data + ga.ga_len, tail); vim_regfree(regmatch.regprog); } ret = vim_strsave(ga.ga_data == NULL ? str : (char_u *)ga.ga_data); ga_clear(&ga); if (p_cpo == empty_option) p_cpo = save_cpo; else // Darn, evaluating {sub} expression or {expr} changed the value. free_string_option(save_cpo); return ret; }