windows x64 support in _tkagg.so

[ddale]

Original report at SourceForge, opened Wed Oct 6 06:04:18 2010

The tkagg backend doesn't natively support the 64-bits version of windows (not ia64 but amd64 architecture).
This is due the fact that long unsigned and __int64 are not the same types on this host.

The attached patch, based on the available commented source in src/_tkagg.cpp allows for a correct handling of Windows 64-bits platform.

SourceForge Comments

On Tue Oct 19 14:38:50 2010, cjgohlke wrote:

There are two things to do:

1. verify that the new scanf format "%zu" does work on all non msvc compilers used to build matplotlib, especially older gcc versions and operating systems. I do not have the means to do this.

2. create a test case that fails without the patch on 64 bit Windows. I have been using mpl on win-amd64 for a year and never ran into this issue (Win 7, 8 GB RAM). From the documentation available, using size_t and "%Iu" is the correct way when using msvc. I verified that mpl still works on win32 and win-amd64 with this change.

Maybe a patch that only changes the scanf format specifier when compiled on Windows is better at this point, even though less straightforward.

On Tue Oct 19 12:21:08 2010, oeloio wrote:

if no one else have, i 'll try your patch next month.
thanks for your help

On Tue Oct 19 11:04:06 2010, mdboom wrote:

I don't have access to a 64-bit Windows installation to test with. Do any of the other developers? Any way you could get a third-party to verify that this fix works?

On Thu Oct 7 14:25:23 2010, cjgohlke wrote:

Regarding my patch: it is probably better to check for _MSC_VER instead of _WIN32:

-#if defined(_WIN32)
+#if defined(_MSC_VER)

On Thu Oct 7 02:55:09 2010, oeloio wrote:

forget what i said, i was thinking about soth. else.
let me know what you chose to implement.

On Thu Oct 7 02:50:40 2010, cjgohlke wrote:

Never heard of ptr_type and its scanf format. I got the length specifiers for size_t from Wikipedia http://en.wikipedia.org/wiki/Printf. However MSDN does not seem to know about 'I' (capital i) http://msdn.microsoft.com/en-us/library/xdb9w69d%28v=VS.90%29.aspx.

On Thu Oct 7 01:53:50 2010, oeloio wrote:

Thanks for your answer.
Then, should we not be directly using a ptr_type if we choose to adjust the sscanf format ?

On Wed Oct 6 22:29:41 2010, cjgohlke wrote:

This should be fixed. How about using size_t to store the pointers and adjusting the sscanf format?

Index: src/_tkagg.cpp

--- src/_tkagg.cpp (revision 8732)
+++ src/_tkagg.cpp (working copy)
@@ -32,8 +32,12 @@
#endif
}

+#if defined(_WIN32)
+# define SIZE_T_FORMAT "%Iu"
+#else
+# define SIZE_T_FORMAT "%zu"
+#endif

typedef struct
{
PyObject_HEAD
@@ -51,7 +55,7 @@
// vars for blitting
PyObject* bboxo;

• unsigned long aggl, bboxl;
• size_t aggl, bboxl;
  bool has_bbox;
  agg::int8u destbuffer;
  double l, b, r, t;
  @@ -81,7 +85,7 @@
  return TCL_ERROR;
  }
  / get array (or object that can be converted to array) pointer */
• if (sscanf(argv[2], "%lu", &aggl) != 1)
• if (sscanf(argv[2], SIZE_T_FORMAT, &aggl) != 1)
  {
  Tcl_AppendResult(interp, "error casting pointer", (char *) NULL);
  return TCL_ERROR;
  @@ -107,7 +111,7 @@
  }

/* check for bbox/blitting */

• if (sscanf(argv[4], "%lu", &bboxl) != 1)
• if (sscanf(argv[4], SIZE_T_FORMAT, &bboxl) != 1)
  {
  Tcl_AppendResult(interp, "error casting pointer", (char *) NULL);
  return TCL_ERROR;

SourceForge History

• On Sat Oct 9 20:07:44 2010, by efiring: assigned_to: 100
• On Wed Oct 6 06:04:19 2010, by oeloio: File Added: 389037: patch_wx64_tkagg.diff

[dmcdougall]

Closing: commit 13384a1 claims to have resolved this.

@cgohlke Let me know if I'm wrong and I will re-open this.