Tcl Source Code

I quote the doc:
       TCL_ARGV_CONSTANT
              The argument does not take any following value
              argument. If this argument is present, the int pointed
              to by the srcPtr field is copied to the dstPtr
              field. The clientData field is ignored.

But today, while programming an extension, I noticed that it's not
"the int pointed to by the srcPtr field" but the pointer itself being
put in dstPtr. When compiling the attached "test.c" and testing it, I get:

  $ gcc -shared -o test.{so,c} -DUSE_TCL_STUBS -I/usr/include -L/usr/lib64 -ltclstub8.6
  $ tclsh
  > load ./test.so
  > test1 -opt
  515178680
  > test2 -opt
  10


After looking at the source code, the solution is either applying

diff --git a/tclIndexObj.c b/tclIndexObj.c
index 30c33f1..07792c5 100644
--- a/tclIndexObj.c
+++ b/tclIndexObj.c
@@ -1207,7 +1207,7 @@ Tcl_ParseArgsObjv(
 	infoPtr = matchPtr;
 	switch (infoPtr->type) {
 	case TCL_ARGV_CONSTANT:
-	    *((int *) infoPtr->dstPtr) = PTR2INT(infoPtr->srcPtr);
+	    *((int *) infoPtr->dstPtr) = *((int *) infoPtr->srcPtr);
 	    break;
 	case TCL_ARGV_INT:
 	    if (objc == 0) {


or changing the documentation to mean that it's the value being copied.

Documentation fixed in: [b970ae4ac5]. Will be in next release.

There actually is a test-case, testing for the actual current behavior: [artifact/76c4d17a5f6cca82?ln=7367]

Actually, I don't know if TCL_ARGV_CONSTANT is useful in real life, but the problem is: If we change the actual behavior then current extensions using TCL_ARGV_CONSTANT will be broken by this changes. Therefore I would suggest to change the documentation.

Thanks for reporting this! Maybe it - indeed - would be more logical to change the implementation, but it's too dangerous IMHO