This is not necessarily the current version of this TIP.This is not necessarily the current version of this TIP.
| TIP: | 80 |
| Title: | Additional Options for 'lsearch' |
| Version: | $Revision: 1.4 $ |
| Authors: |
Tom Wilkason <tom dot wilkason at home dot com> Tom Wilkason <tom dot wilkason at cox dot net> |
| State: | Draft |
| Type: | Project |
| Tcl-Version: | 8.4 |
| Vote: | Pending |
| Created: | Wednesday, 02 January 2002 |
| Discussions To: | news:comp.lang.tcl |
This TIP proposes additional options for the lsearch command to return and work with all matching items in the return rather than the first matching item.
The lsearch function works well for finding the first item in a list that matches a pattern. However it is often useful to find all of the items in the list that match a pattern. This TIP proposes adding options to return the entire list of matches. With this capability, additional options are proposed to return the data rather than the indices (since you often want to work the the data anyway), and to add an option to return the logical exclusion of the matching items (i.e. those that don't match the search pattern).
I propose the following options be added to lsearch:
Option: -start index
Initiates the list search starting at index, which can be any valid list index (such as 0 , end , end-1 ...)
Option: -all
Returns a list of all indices that match the search condition (rather than the first one). The indices are returned low to high order. For a no match condition, a {} (empty result) is returned. If the the -all or -data switches are not specified, a -1 is returned for a no match condition just as it is done now.
Option: -data
Returns a list of the data that matches the search condition rather than the index (or indices with the -all option). A {} is returned for a no match condition. The data is returned in proper list order. This option is useful when you want to iterate over the returned data anyway. e.g.
foreach item [lsearch -all -data -glob $someList *stuff] {
# deal with item
}
Option: -not
Negates the sense of the search condition (i.e. what doesn't match). When used with the -data or -all options, the return set will be the items that do match. If all items match then a {} is returned. Without the -all option, the first item in the list that does not match will be returned.
These can be combined as needed and yield some powerful capabilities when iterating over sub-lists (esp. with the new lset command).
Changes to the Tcl_LsearchObjCmd command in generic/tclCmdIL.c are needed along with documentation and test code. The changes to the 8.4 head version of tclCmdIL.c are available here.
/*
*----------------------------------------------------------------------
*
* Tcl_LsearchObjCmd --
*
* This procedure is invoked to process the "lsearch" Tcl command.
* See the user documentation for details on what it does.
*
* Results:
* A standard Tcl result.
*
* Side effects:
* See the user documentation.
*
*----------------------------------------------------------------------
*/
int
Tcl_LsearchObjCmd(clientData, interp, objc, objv)
ClientData clientData; /* Not used. */
Tcl_Interp *interp; /* Current interpreter. */
int objc; /* Number of arguments. */
Tcl_Obj *CONST objv[]; /* Argument values. */
{
char *bytes, *patternBytes;
int i, match, mode, index, result, listc, length, elemLen, returnData=0;
int useStart=-1, offset, allData=0;
int dataType, isIncreasing, lower, upper, patInt, objInt, notMatch=FALSE;
double patDouble, objDouble;
Tcl_Obj *patObj, **listv, *listPtr, *startPtr;
static CONST char *options[] = {
"-all", "-ascii", "-data", "-decreasing", "-dictionary",
"-exact", "-increasing","-integer", "-glob", "-not",
"-real", "-regexp", "-sorted", "-start", NULL
};
enum options {
LSEARCH_ALL, LSEARCH_ASCII, LSEARCH_DATA, LSEARCH_DECREASING, LSEARCH_DICTIONARY,
LSEARCH_EXACT,LSEARCH_INCREASING, LSEARCH_INTEGER, LSEARCH_GLOB, LSEARCH_NOT,
LSEARCH_REAL, LSEARCH_REGEXP, LSEARCH_SORTED, LSEARCH_START
};
enum datatypes {
ASCII, DICTIONARY, INTEGER, REAL
};
enum modes {
EXACT, GLOB, REGEXP, SORTED
};
mode = GLOB;
dataType = ASCII;
isIncreasing = 1;
/* Note: This counts options as possible list|patterns */
if (objc < 3) {
Tcl_WrongNumArgs(interp, 1, objv, "?options? list pattern");
return TCL_ERROR;
}
for (i = 1; i < objc-2; i++) {
if (Tcl_GetIndexFromObj(interp, objv[i], options, "option", 0, &index)
!= TCL_OK) {
return TCL_ERROR;
}
switch ((enum options) index) {
case LSEARCH_ASCII: /* -ascii */
dataType = ASCII;
break;
case LSEARCH_NOT: /* -not */
notMatch = TRUE;
break;
case LSEARCH_ALL: /* -all */
allData = TRUE;
listPtr = Tcl_NewListObj(0, (Tcl_Obj **) NULL);
break;
case LSEARCH_DATA: /* -data */
returnData = TRUE;
break;
case LSEARCH_START: /* -start index*/
useStart = ++i; /* Pull off next as index for offset */
break;
case LSEARCH_DECREASING: /* -decreasing */
isIncreasing = 0;
break;
case LSEARCH_DICTIONARY: /* -dictionary */
dataType = DICTIONARY;
break;
case LSEARCH_EXACT: /* -exact */
mode = EXACT;
break;
case LSEARCH_INCREASING: /* -increasing */
isIncreasing = 1;
break;
case LSEARCH_INTEGER: /* -integer */
dataType = INTEGER;
break;
case LSEARCH_GLOB: /* -glob */
mode = GLOB;
break;
case LSEARCH_REAL: /* -real */
dataType = REAL;
break;
case LSEARCH_REGEXP: /* -regexp */
mode = REGEXP;
break;
case LSEARCH_SORTED: /* -sorted */
mode = SORTED;
break;
}
}
/*
* -start option processing:
* Ensure we get a unique copy of command line arg for start index
*/
if (useStart > 0)
{
startPtr = Tcl_DuplicateObj(objv[useStart]);
}
/*
* Make sure the list argument is a list object and get its length and
* a pointer to its array of element pointers.
*/
result = Tcl_ListObjGetElements(interp, objv[objc - 2], &listc, &listv);
if (result != TCL_OK) {
return result;
}
/*
* Retrieve user specified start offset.
*/
if (useStart > 0)
{
result = TclGetIntForIndex(interp, startPtr, /*end*/ listc-1, &offset);
Tcl_DecrRefCount(startPtr); /* free unneeded obj */
if (result != TCL_OK) {
return result;
} else if (offset < 0) {
offset = 0;
}
} else {
offset = 0;
}
/*
* Process the pattern
*/
patObj = objv[objc - 1];
patternBytes = NULL;
if ((enum modes) mode == EXACT || (enum modes) mode == SORTED) {
switch ((enum datatypes) dataType) {
case ASCII:
case DICTIONARY:
patternBytes = Tcl_GetStringFromObj(patObj, &length);
break;
case INTEGER:
result = Tcl_GetIntFromObj(interp, patObj, &patInt);
if (result != TCL_OK) {
return result;
}
break;
case REAL:
result = Tcl_GetDoubleFromObj(interp, patObj, &patDouble);
if (result != TCL_OK) {
return result;
}
break;
}
} else {
patternBytes = Tcl_GetStringFromObj(patObj, &length);
}
/*
* Set default index value to -1, indicating failure; if we find the
* item in the course of our search, index will be set to the correct
* value.
*/
index = -1;
match = 0;
if ((enum modes) mode == SORTED && allData == FALSE) {
/* If the data is sorted, we can do a more intelligent search */
lower = offset-1 /*-1*/;
upper = listc;
while (lower + 1 != upper) {
i = (lower + upper)/2;
switch ((enum datatypes) dataType) {
case ASCII: {
bytes = Tcl_GetString(listv[i]);
match = strcmp(patternBytes, bytes);
break;
}
case DICTIONARY: {
bytes = Tcl_GetString(listv[i]);
match = DictionaryCompare(patternBytes, bytes);
break;
}
case INTEGER: {
result = Tcl_GetIntFromObj(interp, listv[i], &objInt);
if (result != TCL_OK) {
return result;
}
if (patInt == objInt) {
match = 0;
} else if (patInt < objInt) {
match = -1;
} else {
match = 1;
}
break;
}
case REAL: {
result = Tcl_GetDoubleFromObj(interp, listv[i],
&objDouble);
if (result != TCL_OK) {
return result;
}
if (patDouble == objDouble) {
match = 0;
} else if (patDouble < objDouble) {
match = -1;
} else {
match = 1;
}
break;
}
}
if (match == 0) {
/*
* Normally, binary search is written to stop when it
* finds a match. If there are duplicates of an element in
* the list, our first match might not be the first occurance.
* Consider: 0 0 0 1 1 1 2 2 2
* To maintain consistancy with standard lsearch semantics,
* we must find the leftmost occurance of the pattern in the
* list. Thus we don't just stop searching here. This
* variation means that a search always makes log n
* comparisons (normal binary search might "get lucky" with
* an early comparison).
*/
index = i;
upper = i;
} else if (match > 0) {
if (isIncreasing) {
lower = i;
} else {
upper = i;
}
} else {
if (isIncreasing) {
upper = i;
} else {
lower = i;
}
}
}
} else {
for (i = offset; i < listc; i++) {
match = 0;
switch ((enum modes) mode) {
case SORTED:
case EXACT: {
switch ((enum datatypes) dataType) {
case ASCII: {
bytes = Tcl_GetStringFromObj(listv[i], &elemLen);
if (length == elemLen) {
match = (memcmp(bytes, patternBytes,
(size_t) length) == 0);
}
break;
}
case DICTIONARY: {
bytes = Tcl_GetString(listv[i]);
match =
(DictionaryCompare(bytes, patternBytes) == 0);
break;
}
case INTEGER: {
result = Tcl_GetIntFromObj(interp, listv[i],
&objInt);
if (result != TCL_OK) {
return result;
}
match = (objInt == patInt);
break;
}
case REAL: {
result = Tcl_GetDoubleFromObj(interp, listv[i],
&objDouble);
if (result != TCL_OK) {
return result;
}
match = (objDouble == patDouble);
break;
}
}
break;
}
case GLOB: {
match = Tcl_StringMatch(Tcl_GetString(listv[i]),
patternBytes);
break;
}
case REGEXP: {
match = Tcl_RegExpMatchObj(interp, listv[i], patObj);
if (match < 0) {
return TCL_ERROR;
}
break;
}
}
/* Invert match condition for -not */
if (notMatch) {
match = (match != 0 ? 0 : 1);
}
/* Process the possible match for this element */
if (match != 0) {
if (allData == TRUE) {
if (returnData) {
/* Append data */
Tcl_ListObjAppendElement(interp, listPtr,listv[i]);
} else {
/* Append index */
Tcl_ListObjAppendElement(interp, listPtr,Tcl_NewIntObj(i));
}
} else {
index = i;
break;
}
}
}
}
/*
* Return either a list (-all) or a single element
*/
if (allData) {
Tcl_SetObjResult(interp,listPtr);
} else {
if (returnData) {
if (index < 0) { /* Return a null */
Tcl_SetObjResult(interp,Tcl_NewObj());
} else { /* Return one datum */
Tcl_SetObjResult(interp,listv[index]);
}
} else {
Tcl_SetIntObj(Tcl_GetObjResult(interp), index);
}
}
return TCL_OK;
}
The changes to lsearch are entirely backward compatible and do no change the behaviour or performance of the command for existing options. Moreover, these changes should not impact any of the other list changes in TIP #22, TIP #33 or TIP #45.
This document has been placed in the public domain.
##
# performs a lsearch -all -data -glob search
#
proc lsearch_dataGLOB {listData pattern} {
set result [list]
foreach item $listData {
if {[string match $pattern $item]} {
lappend result $item
}
}
return $result
}
##
# performs a lsearch -all -data -regexp search
#
proc lsearch_dataRE {listData pattern} {
set result [list]
foreach item $listData {
if {[regexp $pattern $item]} {
lappend result $item
}
}
return $result
}
##
# performs a lsearch -all -glob search
#
proc lsearch_allGLOB {listData pattern} {
set result [list]
set count 0
foreach item $listData {
if {[string match $pattern $item]} {
lappend result $count
}
incr count
}
return $result
}
# Build a 2K list of data
catch {unset LIST}
time {lappend LIST someStuff} 1000
time {lappend LIST otherStuff} 1000
# Case with all data matching in a 2K list 5x speedup
puts "#C implementation [time {listx search -glob -all -data $LIST *Stuff} 100]"
#=> C implementation 6810 microseconds per iteration
puts "#tcl implementation [time {lsearch_dataGLOB $LIST *Stuff} 100]"
#=> tcl implementation 33050 microseconds per iteration
# Case with all data matching but returning indicies 6.5X speed up
puts "#C implementation [time {listx search -glob -all $LIST *Stuff} 100]"
#=> C implementation 5800 microseconds per iteration
puts "#tcl implementation [time {lsearch_allGLOB $LIST *Stuff} 100]"
#=> tcl implementation 39060 microseconds per iteration
# Case with no matching data 12X speed up
puts "#C implementation [time {listx search -glob -all -data $LIST none*} 100]"
#=> C implementation 1010 microseconds per iteration
puts "#tcl implementation [time {lsearch_dataGLOB $LIST none*} 100]"
#=> tcl implementation 13210 microseconds per iteration
# Repeat with RE, note more time spent in RE engine 2X speedup
puts "#C implementation [time {listx search -regexp -all -data $LIST Stuff} 100]"
#=> C implementation 39050 microseconds per iteration
puts "#tcl implementation [time {lsearch_dataRE $LIST Stuff} 100]"
#=> tcl implementation 76200 microseconds per iteration
# Case with no matching data 2X speedup
puts "#C implementation [time {listx search -regexp -all -data $LIST none*} 100]"
#=> C implementation 17520 microseconds per iteration
puts "#tcl implementation [time {lsearch_dataRE $LIST none} 100]"
#=> tcl implementation 33250 microseconds per iteration
This is not necessarily the current version of this TIP.