distorm3 has problems compiling/running on OSX 64bits [PATCH]

[GoogleCodeExporter]

What steps will reproduce the problem?
1. Compilation fails
2. Once fixed distorm3 segfaults if running the executable in 64bits

The patch will:
-Allow distorm3 to successfully build on OSX
-Build universal binaries for 32bit & 64bit environments
-Successfully execute in 64bits

The build process for OSX goes like:
-Enter "build/mac"
-Run "make"
-Change to "Python/"
-Run "python setup.py install"

That should build a universal (i386, x86_64) binary for the Python library and 
get it installed. Patch is attached that fixes the issues described

Original issue reported on code.google.com by ero.carr...@gmail.com on 10 Oct 2010 at 11:50

Attachments:

• distorm_ox_patch.diff

[GoogleCodeExporter]

Hey Ero,
Thanks for the patch,
I will add it to source this weekend,
unfortunately I can't check it out myself...

Original comment by distorm@gmail.com on 10 Oct 2010 at 9:52

[GoogleCodeExporter]

Hey, I fixed a few other things and also my previous "fixes" weren't thoroughly 
tested and I had broken some things.
Use this new patch instead of the old one (I'll try to delete the old one). The 
new one includes all changes.

Original comment by ero.carr...@gmail.com on 12 Oct 2010 at 3:34

Attachments:

• distorm_ox_patch_20101012.diff

[GoogleCodeExporter]

Ero,
I didn't understand the rationale of changing the 'code' pointer related lines.

Original comment by distorm@gmail.com on 16 Oct 2010 at 10:38

[GoogleCodeExporter]

"addressof()" does not return a valid address in OSX 64bit. "pointer()" does. 
The problem is that "pointer" returns and instance that does not seem to allow 
for simple addition to the pointer value to move it forward in the data buffer 
being decoded. 
Hence I create copies of the sliced buffer and get a pointer() for each, 
effectively achieving the same behavior.

I would agree that would cleaner to operate in pointers as the original code 
intended to, but that does not work on OSX 64bit. I didn't test if it was a 
problem generic to all 64bit platforms or only on Mac. My main interest was to 
get it working, if you feel the problem is rather something with ctypes 
implementation then maybe we can let those guys know. I didn't have time to 
look into the issue in more depth.

Original comment by ero.carr...@gmail.com on 17 Oct 2010 at 11:59

[GoogleCodeExporter]

The question if it really creates a buffer every time, cause otherwise it's not 
so good performance-wise.
I will check it out, thanks.

Original comment by distorm@gmail.com on 18 Oct 2010 at 8:24

• Changed state: Fixed

[GoogleCodeExporter]

Ok, I made it byref, instead of pointer. Seems to be better according to ctypes 
docs.
And the create_code_buffer to be a problem.

Waiting for your test-confirmation on Mac.

Original comment by distorm@gmail.com on 20 Oct 2010 at 2:12

[GoogleCodeExporter]

Hey,

The changes look good & work on OSX. I would only propose to tweak it a bit 
more so that the "create_string_buffer" is called only once and then we work 
with the byref() on that object. I didn't know that byref() can take the offset 
as a second optional argument, that's really convenient. Although according to 
the documentation that was just introduced in Python 2.6 will prevent distorm 
from working with Python versions older than 2.5.
I'm just pasting the relevant part of the diff next as it's just a few minor 
changes.

diff -x .svn -urN distorm-read-only/Python/__init__.py 
distorm-read-only-osxfix/Python/__init__.py
--- distorm-read-only/Python/__init__.py        2010-10-20 04:13:38.000000000 
+0200
+++ distorm-read-only-osxfix/Python/__init__.py 2010-10-26 17:23:24.000000000 
+0200
@@ -374,7 +374,8 @@
         raise ValueError("Invalid decode type value: %r" % (dt,))

     codeLen         = len(code)
-    p_code          = byref(create_string_buffer(code))
+    code_buf        = create_string_buffer(code)
+    p_code          = byref(code_buf)
     result          = (_DecodedInst * MAX_INSTRUCTIONS)()
     p_result        = byref(result)
     instruction_off = 0
@@ -405,7 +406,7 @@
         if delta <= 0:
             break
         codeOffset = codeOffset + delta
-        p_code     = byref(create_string_buffer(code), instruction_off)
+        p_code     = byref(code_buf, instruction_off)
         codeLen    = codeLen - delta

 def Decode(offset, code, type = Decode32Bits):
@@ -712,7 +713,8 @@
         raise ValueError("Invalid decode type value: %r" % (dt,))

     codeLen         = len(code)
-    p_code          = byref(create_string_buffer(code))
+    code_buf        = create_string_buffer(code)
+    p_code          = byref(code_buf)
     result          = (_DInst * MAX_INSTRUCTIONS)()
     instruction_off = 0

@@ -738,7 +740,7 @@
         if delta <= 0:
             break
         codeOffset = codeOffset + delta
-        p_code     = byref(create_string_buffer(code), instruction_off)
+        p_code     = byref(code_buf, instruction_off)
         codeLen    = codeLen - delta

 def Decompose(offset, code, type = Decode32Bits):

Original comment by ero.carr...@gmail.com on 26 Oct 2010 at 4:48

[GoogleCodeExporter]

That's nice indeed! Thanks a lot.

Original comment by distorm@gmail.com on 31 Oct 2010 at 8:10