Tcl Source Code

matching range is found and substituted.
If \fB\-all\fR is specified, then
.QW &
and
.QW \e\fIn\fR
sequences are handled for each substitution using the information
from the corresponding match.
.TP
\fB\-command\fR
.VS 8.7
Changes the handling of the substitution string so that it no longer treats
.QW &
and
.QW \e
as special characters, but instead uses them as a non-empty list of words.
Each time a substitution is processed, another complete Tcl word is appended
to that list for each substitution value (the first such argument represents
the overall matched substring, the subsequent arguments will be one per
capturing sub-RE, much as are returned from \fBregexp\fR \fB\-inline\fR) and
the overall list is then evaluated as a Tcl command call. If the command
finishes successfully, the result of command call is substituted into the
resulting string.
.RS
.PP
If \fB\-all\fR is not also given, the command callback will be invoked at most
once (exactly when the regular expression matches). If \fB\-all\fR is given,
the command callback will be invoked for each matched location, in sequence.
The exact location indices that matched are not made available to the script.
.PP
See \fBEXAMPLES\fR below for illustrative cases.
.RE
.VE 8.7
.TP
\fB\-expanded\fR
.
Enables use of the expanded regular expression syntax where
whitespace and comments are ignored.  This is the same as specifying
the \fB(?x)\fR embedded option (see the \fBre_syntax\fR manual page).
.TP

# Now we apply the substitution to get a subst-string that
# will perform the computational parts of the conversion. Note
# that newline is handled specially through \fBstring map\fR since
# backslash-newline is a special sequence.
set quoted [subst [string map {\en {\e\eu000a}} \e
        [\fBregsub\fR -all $RE $string $substitution]]]
.CE
.PP
.VS 8.7
The above operation can be done using \fBregsub \-command\fR instead, which is
often faster. (A full pre-computed \fBstring map\fR would be faster still, but
the cost of computing the map for a transformation as complex as this can be
quite large.)
.PP
.CS
# This RE is just a character class for everything "bad"
set RE {[][{};#\e\e\e$\es\eu0080-\euffff]}

# This encodes what the RE described above matches
proc encodeChar {ch} {
    # newline is handled specially since backslash-newline is a
    # special sequence.
    if {$ch eq "\en"} {
        return "\e\eu000a"
    }
    # No point in writing this as a one-liner
    scan $ch %c charNumber
    format "\e\eu%04x" $charNumber
}

set quoted [\fBregsub\fR -all -command $RE $string encodeChar]
.CE
.PP
Decoding a URL-encoded string using \fBregsub \-command\fR, a lambda term and
the \fBapply\fR command.
.PP
.CS
# Match one of the sequences in a URL-encoded string that needs
# fixing, converting + to space and %XX to the right character
# (e.g., %7e becomes ~)
set RE {(\e+)|%([0-9A-Fa-f]{2})}

# Note that -command uses a command prefix, not a command name
set decoded [\fBregsub\fR -all -command $RE $string {apply {{- p h} {
    # + is a special case; handle directly
    if {$p eq "+"} {
        return " "
    }
    # convert hex to a char
    scan $h %x charNumber
    format %c $charNumber
}}}]
.CE
.VE 8.7
.SH "SEE ALSO"
regexp(n), re_syntax(n), subst(n), string(n)
.SH KEYWORDS
match, pattern, quoting, regular expression, substitution
'\" Local Variables:
'\" mode: nroff
'\" End:

Tcl_RegsubObjCmd(
    ClientData dummy,		/* Not used. */
    Tcl_Interp *interp,		/* Current interpreter. */
    int objc,			/* Number of arguments. */
    Tcl_Obj *const objv[])	/* Argument objects. */
{
    int idx, result, cflags, all, wlen, wsublen, numMatches, offset;
    int start, end, subStart, subEnd, match, command, numParts, numArgs;
    Tcl_RegExp regExpr;
    Tcl_RegExpInfo info;
    Tcl_Obj *resultPtr, *subPtr, *objPtr, *startIndex = NULL;
    Tcl_Obj **args = NULL, **parts;
    Tcl_UniChar ch, *wsrc, *wfirstChar, *wstring, *wsubspec, *wend;

    static const char *const options[] = {
	"-all",		"-command",	"-expanded",	"-line",
	"-linestop",	"-lineanchor",	"-nocase",	"-start",
	"--",		NULL
    };
    enum options {
	REGSUB_ALL,	 REGSUB_COMMAND,    REGSUB_EXPANDED, REGSUB_LINE,
	REGSUB_LINESTOP, REGSUB_LINEANCHOR, REGSUB_NOCASE,   REGSUB_START,
	REGSUB_LAST
    };

    cflags = TCL_REG_ADVANCED;
    all = 0;
    offset = 0;
    command = 0;
    resultPtr = NULL;

    for (idx = 1; idx < objc; idx++) {
	const char *name;
	int index;

	name = TclGetString(objv[idx]);
	if (name[0] != '-') {
	    break;
	}
	if (Tcl_GetIndexFromObj(interp, objv[idx], options, "option",
		TCL_EXACT, &index) != TCL_OK) {
	    goto optionError;
	}
	switch ((enum options) index) {
	case REGSUB_ALL:
	    all = 1;
	    break;
	case REGSUB_NOCASE:
	    cflags |= TCL_REG_NOCASE;
	    break;
	case REGSUB_COMMAND:
	    command = 1;
	    break;
	case REGSUB_EXPANDED:
	    cflags |= TCL_REG_EXPANDED;
	    break;
	case REGSUB_LINE:
	    cflags |= TCL_REG_NEWLINE;
	    break;

	TclGetIntForIndexM(NULL, startIndex, stringLength, &offset);
	Tcl_DecrRefCount(startIndex);
	if (offset < 0) {
	    offset = 0;
	}
    }

    if (all && (offset == 0) && (command == 0)
	    && (strpbrk(TclGetString(objv[2]), "&\\") == NULL)
	    && (strpbrk(TclGetString(objv[0]), "*+?{}()[].\\|^$") == NULL)) {
	/*
	 * This is a simple one pair string map situation. We make use of a
	 * slightly modified version of the one pair STR_MAP code.
	 */

	goto regsubDone;
    }

    regExpr = Tcl_GetRegExpFromObj(interp, objv[0], cflags);
    if (regExpr == NULL) {
	return TCL_ERROR;
    }

    if (command) {
	/*
	 * In command-prefix mode, we require that the third non-option
	 * argument be a list, so we enforce that here. Afterwards, we fetch
	 * the RE compilation again in case objv[0] and objv[2] are the same
	 * object. (If they aren't, that's cheap to do.)
	 */

	if (Tcl_ListObjLength(interp, objv[2], &numParts) != TCL_OK) {
	    return TCL_ERROR;
	}
	if (numParts < 1) {
	    Tcl_SetObjResult(interp, Tcl_NewStringObj(
		    "command prefix must be a list of at least one element",
		    -1));
	    Tcl_SetErrorCode(interp, "TCL", "OPERATION", "REGSUB",
		    "CMDEMPTY", NULL);
	    return TCL_ERROR;
	}
	regExpr = Tcl_GetRegExpFromObj(interp, objv[0], cflags);
    }

    /*
     * Make sure to avoid problems where the objects are shared. This can
     * cause RegExpObj <> UnicodeObj shimmering that causes data corruption.
     * [Bug #461322]
     */

    if (objv[1] == objv[0]) {
	objPtr = Tcl_DuplicateObj(objv[1]);
    } else {
	objPtr = objv[1];
    }
    wstring = Tcl_GetUnicodeFromObj(objPtr, &wlen);
    if (objv[2] == objv[0]) {
	subPtr = Tcl_DuplicateObj(objv[2]);
    } else {
	subPtr = objv[2];
    }
    if (!command) {
	wsubspec = Tcl_GetUnicodeFromObj(subPtr, &wsublen);
    }

    result = TCL_OK;

    /*
     * The following loop is to handle multiple matches within the same source
     * string; each iteration handles one match and its corresponding
     * substitution. If "-all" hasn't been specified then the loop body only

	 * result variable.
	 */

	Tcl_RegExpGetInfo(regExpr, &info);
	start = info.matches[0].start;
	end = info.matches[0].end;
	Tcl_AppendUnicodeToObj(resultPtr, wstring + offset, start);

	/*
	 * In command-prefix mode, the substitutions are added as quoted
	 * arguments to the subSpec to form a command, that is then executed
	 * and the result used as the string to substitute in. Actually,
	 * everything is passed through Tcl_EvalObjv, as that's much faster.
	 */

	if (command) {
	    if (args == NULL) {
		Tcl_ListObjGetElements(interp, subPtr, &numParts, &parts);
		numArgs = numParts + info.nsubs + 1;
		args = ckalloc(sizeof(Tcl_Obj*) * numArgs);
		memcpy(args, parts, sizeof(Tcl_Obj*) * numParts);
	    }

	    for (idx = 0 ; idx <= info.nsubs ; idx++) {
		subStart = info.matches[idx].start;
		subEnd = info.matches[idx].end;
		if ((subStart >= 0) && (subEnd >= 0)) {
		    args[idx + numParts] = Tcl_NewUnicodeObj(
			    wstring + offset + subStart, subEnd - subStart);
		} else {
		    args[idx + numParts] = Tcl_NewObj();
		}
		Tcl_IncrRefCount(args[idx + numParts]);
	    }

	    /*
	     * At this point, we're locally holding the references to the
	     * argument words we added for this time round the loop, and the
	     * subPtr is holding the references to the words that the user
	     * supplied directly. None are zero-refcount, which is important
	     * because Tcl_EvalObjv is "hairy monster" in terms of refcount
	     * handling, being able to optionally add references to any of its
	     * argument words. We'll drop the local refs immediately
	     * afterwarsds; subPtr is handled in the main exit stanza.
	     */

	    result = Tcl_EvalObjv(interp, numArgs, args, 0);
	    for (idx = 0 ; idx <= info.nsubs ; idx++) {
		TclDecrRefCount(args[idx + numParts]);
	    }
	    if (result != TCL_OK) {
		if (result == TCL_ERROR) {
		    Tcl_AppendObjToErrorInfo(interp, Tcl_ObjPrintf(
			    "\n    (%s substitution computation script)",
			    options[REGSUB_COMMAND]));
		}
		goto done;
	    }

	    Tcl_AppendObjToObj(resultPtr, Tcl_GetObjResult(interp));
	    Tcl_ResetResult(interp);

	    /*
	     * Refetch the unicode, in case the representation was smashed by
	     * the user code.
	     */

	    wstring = Tcl_GetUnicodeFromObj(objPtr, &wlen);

	    offset += end;
	    if (end == 0 || start == end) {
		/*
		 * Always consume at least one character of the input string
		 * in order to prevent infinite loops, even when we
		 * technically matched the empty string; we must not match
		 * again at the same spot.
		 */

		if (offset < wlen) {
		    Tcl_AppendUnicodeToObj(resultPtr, wstring + offset, 1);
		}
		offset++;
	    }
	    if (all) {
		continue;
	    } else {
		break;
	    }
	}

	/*
	 * Append the subSpec argument to the variable, making appropriate
	 * substitutions. This code is a bit hairy because of the backslash
	 * conventions and because the code saves up ranges of characters in
	 * subSpec to reduce the number of calls to Tcl_SetVar.
	 */

  done:
    if (objPtr && (objv[1] == objv[0])) {
	Tcl_DecrRefCount(objPtr);
    }
    if (subPtr && (objv[2] == objv[0])) {
	Tcl_DecrRefCount(subPtr);
    }
    if (args) {
	ckfree(args);
    }
    if (resultPtr) {
	Tcl_DecrRefCount(resultPtr);
    }
    return result;
}


    package require tcltest 2
    namespace import -force ::tcltest::*
}

unset -nocomplain foo

testConstraint exec [llength [info commands exec]]

# Used for constraining memory leak tests
testConstraint memory [llength [info commands memory]]
if {[testConstraint memory]} {
    proc memtest script {
	set end [lindex [split [memory info] \n] 3 3]
	for {set i 0} {$i < 5} {incr i} {
	    uplevel 1 $script
	    set tmp $end
	    set end [lindex [split [memory info] \n] 3 3]
	}
	expr {$end - $tmp}
    }
}

test regexp-1.1 {basic regexp operation} {
    regexp ab*c abbbc
} 1
test regexp-1.2 {basic regexp operation} {
    regexp ab*c ac
} 1

    list [catch {regsub -nocase -all a b} msg] $msg
} {1 {wrong # args: should be "regsub ?-option ...? exp string subSpec ?varName?"}}
test regexp-11.4 {regsub errors} {
    list [catch {regsub a b c d e f} msg] $msg
} {1 {wrong # args: should be "regsub ?-option ...? exp string subSpec ?varName?"}}
test regexp-11.5 {regsub errors} {
    list [catch {regsub -gorp a b c} msg] $msg
} {1 {bad option "-gorp": must be -all, -command, -expanded, -line, -linestop, -lineanchor, -nocase, -start, or --}}
test regexp-11.6 {regsub errors} {
    list [catch {regsub -nocase a( b c d} msg] $msg
} {1 {couldn't compile regular expression pattern: parentheses () not balanced}}
test regexp-11.7 {regsub errors} -setup {
    unset -nocomplain f1
} -body {
    set f1 44

} {a {}}
test regexp-26.12 {regexp with -line option} {
    regexp -all -inline -line -- {a*} "b\n"
} {{} {}}
test regexp-26.13 {regexp without -line option} {
    regexp -all -inline -- {a*} "b\n"
} {{} {}}

test regexp-27.1 {regsub -command} {
    regsub -command {.x.} {abcxdef} {string length}
} ab3ef
test regexp-27.2 {regsub -command} {
    regsub -command {.x.} {abcxdefxghi} {string length}
} ab3efxghi
test regexp-27.3 {regsub -command} {
    set x 0
    regsub -all -command {(?=.)} abcde {apply {args {incr ::x}}}
} 1a2b3c4d5e
test regexp-27.4 {regsub -command} -body {
    regsub -command {.x.} {abcxdef} error
} -returnCodes error -result cxd
test regexp-27.5 {regsub -command} {
    regsub -command {(.)(.)} {abcdef} {list ,}
} {, ab a bcdef}
test regexp-27.6 {regsub -command} {
    regsub -command -all {(.)(.)} {abcdef} {list ,}
} {, ab a b, cd c d, ef e f}
test regexp-27.7 {regsub -command representation smash} {
    set ::s {123=456 789}
    regsub -command -all {\d+} $::s {apply {n {
	expr {[llength $::s] + $n}
    }}}
} {125=458 791}
test regexp-27.8 {regsub -command representation smash} {
    set ::t {apply {n {
	expr {[llength [lindex $::t 1 1 1]] + $n}
    }}}
    regsub -command -all {\d+} "123=456 789" $::t
} {131=464 797}
test regexp-27.9 {regsub -command memory leak testing} memory {
    set ::s "123=456 789"
    set ::t {apply {n {
	expr {[llength [lindex $::t 1 1 1]] + [llength $::s] + $n}
    }}}
    memtest {
	regsub -command -all {\d+} $::s $::t
    }
} 0
test regexp-27.10 {regsub -command error cases} -returnCodes error -body {
    regsub -command . abc "def \{ghi"
} -result {unmatched open brace in list}
test regexp-27.11 {regsub -command error cases} -returnCodes error -body {
    regsub -command . abc {}
} -result {command prefix must be a list of at least one element}

# cleanup
::tcltest::cleanupTests
return

# Local Variables:
# mode: tcl
# End:

	list [catch {regsub a b c d e f} msg] $msg
    }
} {1 {wrong # args: should be "regsub ?-option ...? exp string subSpec ?varName?"}}
test regexpComp-11.5 {regsub errors} {
    evalInProc {
	list [catch {regsub -gorp a b c} msg] $msg
    }
} {1 {bad option "-gorp": must be -all, -command, -expanded, -line, -linestop, -lineanchor, -nocase, -start, or --}}
test regexpComp-11.6 {regsub errors} {
    evalInProc {
	list [catch {regsub -nocase a( b c d} msg] $msg
    }
} {1 {couldn't compile regular expression pattern: parentheses () not balanced}}
test regexpComp-11.7 {regsub errors} {
    evalInProc {