summaryrefslogtreecommitdiffstats
path: root/generic
diff options
context:
space:
mode:
authordkf <dkf@noemail.net>2001-11-29 15:38:40 (GMT)
committerdkf <dkf@noemail.net>2001-11-29 15:38:40 (GMT)
commit9a75cd4ae0bd7f173fa9169d5124bea9d0f163d7 (patch)
treef91eaf3ac2ad9690548bde56c7c74d94bb5957e2 /generic
parentdd035d93838e1f65e1759a5b91d7e0253bc3fc74 (diff)
downloadtcl-9a75cd4ae0bd7f173fa9169d5124bea9d0f163d7.zip
tcl-9a75cd4ae0bd7f173fa9169d5124bea9d0f163d7.tar.gz
tcl-9a75cd4ae0bd7f173fa9169d5124bea9d0f163d7.tar.bz2
* generic/tclBinary.c: Added caching scheme to reduce number of object
allocations when doing scans of large repetitive binary strings. FossilOrigin-Name: 3d0f0e218b6277d9d18b792f838d5b6689c8dd7f
Diffstat (limited to 'generic')
-rw-r--r--generic/tclBinary.c133
1 files changed, 114 insertions, 19 deletions
diff --git a/generic/tclBinary.c b/generic/tclBinary.c
index 9fd9f4d..54f8ff0 100644
--- a/generic/tclBinary.c
+++ b/generic/tclBinary.c
@@ -10,7 +10,7 @@
* See the file "license.terms" for information on usage and redistribution
* of this file, and for a DISCLAIMER OF ALL WARRANTIES.
*
- * RCS: @(#) $Id: tclBinary.c,v 1.8 2001/08/23 14:22:49 dkf Exp $
+ * RCS: @(#) $Id: tclBinary.c,v 1.9 2001/11/29 15:38:40 dkf Exp $
*/
#include <math.h>
@@ -26,6 +26,26 @@
#define BINARY_NOCOUNT -2 /* No count was specified in format. */
/*
+ * The following defines the maximum number of different (integer)
+ * numbers placed in the object cache by 'binary scan' before it bails
+ * out and switches back to Plan A (creating a new object for each
+ * value.) Theoretically, it would be possible to keep the cache
+ * about for the values that are already in it, but that makes the
+ * code slower in practise when overflow happens, and makes little
+ * odds the rest of the time (as measured on my machine.) It is also
+ * slower (on the sample I tried at least) to grow the cache to hold
+ * all items we might want to put in it; presumably the extra cost of
+ * managing the memory for the enlarged table outweighs the benefit
+ * from allocating fewer objects. This is probably because as the
+ * number of objects increases, the likelihood of reuse of any
+ * particular one drops, and there is very little gain from larger
+ * maximum cache sizes (the value below is chosen to allow caching to
+ * work in full with conversion of bytes.) - DKF
+ */
+
+#define BINARY_SCAN_MAX_CACHE 260
+
+/*
* Prototypes for local procedures defined in this file:
*/
@@ -36,7 +56,8 @@ static int FormatNumber _ANSI_ARGS_((Tcl_Interp *interp, int type,
static void FreeByteArrayInternalRep _ANSI_ARGS_((Tcl_Obj *objPtr));
static int GetFormatSpec _ANSI_ARGS_((char **formatPtr,
char *cmdPtr, int *countPtr));
-static Tcl_Obj * ScanNumber _ANSI_ARGS_((unsigned char *buffer, int type));
+static Tcl_Obj * ScanNumber _ANSI_ARGS_((unsigned char *buffer,
+ int type, Tcl_HashTable **numberCachePtr));
static int SetByteArrayFromAny _ANSI_ARGS_((Tcl_Interp *interp,
Tcl_Obj *objPtr));
static void UpdateStringOfByteArray _ANSI_ARGS_((Tcl_Obj *listPtr));
@@ -996,12 +1017,16 @@ Tcl_BinaryObjCmd(dummy, interp, objc, objv)
case BINARY_SCAN: {
int i;
Tcl_Obj *valuePtr, *elementPtr;
+ Tcl_HashTable numberCacheHash;
+ Tcl_HashTable *numberCachePtr;
if (objc < 4) {
Tcl_WrongNumArgs(interp, 2, objv,
"value formatString ?varName varName ...?");
return TCL_ERROR;
}
+ numberCachePtr = &numberCacheHash;
+ Tcl_InitHashTable(numberCachePtr, TCL_ONE_WORD_KEYS);
buffer = Tcl_GetByteArrayFromObj(objv[2], &length);
format = Tcl_GetString(objv[3]);
cursor = buffer;
@@ -1018,6 +1043,9 @@ Tcl_BinaryObjCmd(dummy, interp, objc, objv)
unsigned char *src;
if (arg >= objc) {
+ if (numberCachePtr != NULL) {
+ Tcl_DeleteHashTable(numberCachePtr);
+ }
goto badIndex;
}
if (count == BINARY_ALL) {
@@ -1051,6 +1079,9 @@ Tcl_BinaryObjCmd(dummy, interp, objc, objv)
NULL, valuePtr, TCL_LEAVE_ERR_MSG);
arg++;
if (resultPtr == NULL) {
+ if (numberCachePtr != NULL) {
+ Tcl_DeleteHashTable(numberCachePtr);
+ }
Tcl_DecrRefCount(valuePtr); /* unneeded */
return TCL_ERROR;
}
@@ -1063,6 +1094,9 @@ Tcl_BinaryObjCmd(dummy, interp, objc, objv)
char *dest;
if (arg >= objc) {
+ if (numberCachePtr != NULL) {
+ Tcl_DeleteHashTable(numberCachePtr);
+ }
goto badIndex;
}
if (count == BINARY_ALL) {
@@ -1104,6 +1138,9 @@ Tcl_BinaryObjCmd(dummy, interp, objc, objv)
NULL, valuePtr, TCL_LEAVE_ERR_MSG);
arg++;
if (resultPtr == NULL) {
+ if (numberCachePtr != NULL) {
+ Tcl_DeleteHashTable(numberCachePtr);
+ }
Tcl_DecrRefCount(valuePtr); /* unneeded */
return TCL_ERROR;
}
@@ -1118,6 +1155,9 @@ Tcl_BinaryObjCmd(dummy, interp, objc, objv)
static char hexdigit[] = "0123456789abcdef";
if (arg >= objc) {
+ if (numberCachePtr != NULL) {
+ Tcl_DeleteHashTable(numberCachePtr);
+ }
goto badIndex;
}
if (count == BINARY_ALL) {
@@ -1159,6 +1199,9 @@ Tcl_BinaryObjCmd(dummy, interp, objc, objv)
NULL, valuePtr, TCL_LEAVE_ERR_MSG);
arg++;
if (resultPtr == NULL) {
+ if (numberCachePtr != NULL) {
+ Tcl_DeleteHashTable(numberCachePtr);
+ }
Tcl_DecrRefCount(valuePtr); /* unneeded */
return TCL_ERROR;
}
@@ -1191,13 +1234,17 @@ Tcl_BinaryObjCmd(dummy, interp, objc, objv)
scanNumber:
if (arg >= objc) {
+ if (numberCachePtr != NULL) {
+ Tcl_DeleteHashTable(numberCachePtr);
+ }
goto badIndex;
}
if (count == BINARY_NOCOUNT) {
if ((length - offset) < size) {
goto done;
}
- valuePtr = ScanNumber(buffer+offset, cmd);
+ valuePtr = ScanNumber(buffer+offset, cmd,
+ &numberCachePtr);
offset += size;
} else {
if (count == BINARY_ALL) {
@@ -1209,7 +1256,8 @@ Tcl_BinaryObjCmd(dummy, interp, objc, objv)
valuePtr = Tcl_NewObj();
src = buffer+offset;
for (i = 0; i < count; i++) {
- elementPtr = ScanNumber(src, cmd);
+ elementPtr = ScanNumber(src, cmd,
+ &numberCachePtr);
src += size;
Tcl_ListObjAppendElement(NULL, valuePtr,
elementPtr);
@@ -1221,6 +1269,9 @@ Tcl_BinaryObjCmd(dummy, interp, objc, objv)
NULL, valuePtr, TCL_LEAVE_ERR_MSG);
arg++;
if (resultPtr == NULL) {
+ if (numberCachePtr != NULL) {
+ Tcl_DeleteHashTable(numberCachePtr);
+ }
Tcl_DecrRefCount(valuePtr); /* unneeded */
return TCL_ERROR;
}
@@ -1251,6 +1302,9 @@ Tcl_BinaryObjCmd(dummy, interp, objc, objv)
}
case '@': {
if (count == BINARY_NOCOUNT) {
+ if (numberCachePtr != NULL) {
+ Tcl_DeleteHashTable(numberCachePtr);
+ }
goto badCount;
}
if ((count == BINARY_ALL) || (count > length)) {
@@ -1261,6 +1315,9 @@ Tcl_BinaryObjCmd(dummy, interp, objc, objv)
break;
}
default: {
+ if (numberCachePtr != NULL) {
+ Tcl_DeleteHashTable(numberCachePtr);
+ }
errorString = str;
goto badfield;
}
@@ -1274,6 +1331,9 @@ Tcl_BinaryObjCmd(dummy, interp, objc, objv)
done:
Tcl_ResetResult(interp);
Tcl_SetLongObj(Tcl_GetObjResult(interp), arg - 4);
+ if (numberCachePtr != NULL) {
+ Tcl_DeleteHashTable(numberCachePtr);
+ }
break;
}
}
@@ -1465,15 +1525,21 @@ FormatNumber(interp, type, src, cursorPtr)
* This object has a ref count of zero.
*
* Side effects:
- * None.
+ * Might reuse an object in the number cache, place a new object
+ * in the cache, or delete the cache and set the reference to
+ * it (itself passed in by reference) to NULL.
*
*----------------------------------------------------------------------
*/
static Tcl_Obj *
-ScanNumber(buffer, type)
+ScanNumber(buffer, type, numberCachePtrPtr)
unsigned char *buffer; /* Buffer to scan number from. */
int type; /* Format character from "binary scan" */
+ Tcl_HashTable **numberCachePtrPtr;
+ /* Place to look for cache of scanned
+ * value objects, or NULL if too many
+ * different numbers have been scanned. */
{
long value;
@@ -1486,7 +1552,7 @@ ScanNumber(buffer, type)
*/
switch (type) {
- case 'c': {
+ case 'c':
/*
* Characters need special handling. We want to produce a
* signed result, but on some platforms (such as AIX) chars
@@ -1498,28 +1564,26 @@ ScanNumber(buffer, type)
if (value & 0x80) {
value |= -0x100;
}
- return Tcl_NewLongObj((long)value);
- }
- case 's': {
+ goto returnNumericObject;
+
+ case 's':
value = (long) (buffer[0] + (buffer[1] << 8));
goto shortValue;
- }
- case 'S': {
+ case 'S':
value = (long) (buffer[1] + (buffer[0] << 8));
shortValue:
if (value & 0x8000) {
value |= -0x10000;
}
- return Tcl_NewLongObj(value);
- }
- case 'i': {
+ goto returnNumericObject;
+
+ case 'i':
value = (long) (buffer[0]
+ (buffer[1] << 8)
+ (buffer[2] << 16)
+ (buffer[3] << 24));
goto intValue;
- }
- case 'I': {
+ case 'I':
value = (long) (buffer[3]
+ (buffer[2] << 8)
+ (buffer[1] << 16)
@@ -1534,8 +1598,39 @@ ScanNumber(buffer, type)
value -= (((unsigned int)1)<<31);
value -= (((unsigned int)1)<<31);
}
- return Tcl_NewLongObj(value);
- }
+ returnNumericObject:
+ if (*numberCachePtrPtr == NULL) {
+ return Tcl_NewLongObj(value);
+ } else {
+ register Tcl_HashTable *tablePtr = *numberCachePtrPtr;
+ register Tcl_HashEntry *hPtr;
+ int isNew;
+
+ hPtr = Tcl_CreateHashEntry(tablePtr, (char *)value, &isNew);
+ if (!isNew) {
+ return (Tcl_Obj *) Tcl_GetHashValue(hPtr);
+ }
+ if (tablePtr->numEntries > BINARY_SCAN_MAX_CACHE) {
+ /*
+ * We've overflowed the cache! Someone's parsing
+ * a LOT of varied binary data in a single call!
+ * Bail out by switching back to the old behaviour
+ * for the rest of the scan.
+ *
+ * Note that anyone just using the 'c' conversion
+ * (for bytes) cannot trigger this.
+ */
+ Tcl_DeleteHashTable(tablePtr);
+ *numberCachePtrPtr = NULL;
+ return Tcl_NewLongObj(value);
+ } else {
+ register Tcl_Obj *objPtr = Tcl_NewLongObj(value);
+ /* Don't need to fiddle with refcount... */
+ Tcl_SetHashValue(hPtr, (ClientData) objPtr);
+ return objPtr;
+ }
+ }
+
case 'f': {
float fvalue;
memcpy((VOID *) &fvalue, (VOID *) buffer, sizeof(float));