[MINC-development] "Amusing" mincblur issue...

[Andrew Janke]

Hrm, seems like this is more endemic than I originally thought...

Get this:

   http://mavis.anu.edu.au/minc-beta/parsetest.tar.gz

Make it. (or use my prebuilt intrepid ones) then do this:

(on 64 bit)

   ben:parsetest$ ./parsetest64 -a -b; ./parsetest64 -b -a
   result a:1 b:1
   result a:1 b:0

(on 32 bit)

   mavis:parsetest$ ./parsetest32 -a -b; ./parsetest64 -b -a
   result a:1 b:1
   result a:1 b:1

oops.

As for a fix I went digging to see what the tcl/tk people do (given
that this is their library to start with) and found this:

   http://tktoolkit.cvs.sourceforge.net/viewvc/tktoolkit/tk/generic/tkArgv.c?revision=1.17&view=markup

Specifically:

  173 	switch (infoPtr->type) {
  174 	case TK_ARGV_CONSTANT:
  175 	    *((int *) infoPtr->dst) = PTR2INT(infoPtr->src);
  176 	    break;


Interesting, Macros...  and what pray tell are these macros?

   http://tktoolkit.cvs.sourceforge.net/viewvc/tktoolkit/tk/generic/tkInt.h?view=markup

Gives us this little bit of blah:

   69 /*
   70  * Macros used to cast between pointers and integers (e.g. when
storing an int
   71  * in ClientData), on 64-bit architectures they avoid gcc
warning about "cast
   72  * to/from pointer from/to integer of different size".
   73  */
   74
   75 #if !defined(INT2PTR) && !defined(PTR2INT)
   76 #   if defined(HAVE_INTPTR_T) || defined(intptr_t)
   77 #	define INT2PTR(p) ((void*)(intptr_t)(p))
   78 #	define PTR2INT(p) ((int)(intptr_t)(p))
   79 #   else
   80 #	define INT2PTR(p) ((void*)(p))
   81 #	define PTR2INT(p) ((int)(p))
   82 #   endif
   83 #endif
   84 #if !defined(UINT2PTR) && !defined(PTR2UINT)
   85 #   if defined(HAVE_UINTPTR_T) || defined(uintptr_t)
   86 #	define UINT2PTR(p) ((void*)(uintptr_t)(p))
   87 #	define PTR2UINT(p) ((unsigned int)(uintptr_t)(p))
   88 #   else
   89 #	define UINT2PTR(p) ((void*)(p))
   90 #	define PTR2UINT(p) ((unsigned int)(p))
   91 #   endif
   92 #endif

So do we just do this:

ben:libsrc$ cvs diff ParseArgv.c
Index: ParseArgv.c
===================================================================
RCS file: /private-cvsroot/minc/libsrc/ParseArgv.c,v
retrieving revision 6.9
diff -u -b -B -r6.9 ParseArgv.c
--- ParseArgv.c	12 Aug 2008 05:00:23 -0000	6.9
+++ ParseArgv.c	27 Nov 2008 00:05:45 -0000
@@ -27,6 +27,8 @@
 #include <math.h>
 #include <ParseArgv.h>

+#define PTR2INT(p) ((int)(p))
+
 /*
  * Default table of argument descriptors.  These are normally available
  * in every application.
@@ -174,7 +176,7 @@
       infoPtr = matchPtr;
       switch (infoPtr->type) {
       case ARGV_CONSTANT:
-         *((long *) infoPtr->dst) = (long) infoPtr->src;
+         *((int *) infoPtr->dst) = PTR2INT(infoPtr->src);
          break;
       case ARGV_INT:
          nargs = (int) infoPtr->src;

Which then gives up this result:

   ben:parsetest$ ./parsetest -a -b
   result a:1 b:1
   ben:parsetest$ ./parsetest -b -a
   result a:1 b:1

Which is "correct", but now spits a compiler warning for minc (amongst
others). The question is will this cause other bugs to creep in.
blah.

Thoughts?


a