struct pointer returned by function seems partially messed up in the returned cdata

[Sarvi Shanmugham]

The following is contained in the header file as well as what is passed to CFFI in the py file.

struct token* add_number_expr_after(struct token* where, int num, struct expression** ret);

struct token* add_addr_space_attr_begin(struct token* tok);        //problem function

struct token* add_addr_space_attr_end(struct token* tok);

I have a C function below that I am wrapping. I added some printfs to make sure i see what porinters are being passed/returned by the C function

struct token* add_addr_space_attr_begin(struct token* tok)

{

        struct token* ret = create_built_in_token_before(tok, "__attribute__");

    printf("\nadd_addr_space_attr_begin: %p",ret);

        ret = create_char_token_after(ret, '(');

    printf("\nadd_addr_space_attr_begin: %p",ret);

        return ret;

}

I then stepped into the python debugger to step the piece of code that invokes it

> /nobackup/sarvi/spark1/rules.py(29)fixit()

-> tok = warn_info.symbol1.start_tok

(Pdb) n

> /nobackup/sarvi/spark1/rules.py(30)fixit()

-> if api_mod.check_addr_space(warn_info.symbol1, 1):

(Pdb) n

> /nobackup/sarvi/spark1/rules.py(32)fixit()

-> tok = api_mod.add_addr_space_attr_begin(tok)

(Pdb) n

add_addr_space_attr_begin: 0x2a9a81bfd0

add_addr_space_attr_begin: 0x2a9a81c010> /nobackup/sarvi/spark1/rules.py(33)fixit()

-> tok = api_mod.add_number_expr_after(tok, 1, exprptr)

(Pdb) print tok

<cdata 'struct token *' 0xffffffff9a81c010>

The value I see being returned seems like a valid pointer right to the end where it does 

return ret

But the tok that I print from the python debugger seems off on the top.

<cdata 'struct token *' 0xffffffff9a81c010>

CFFI version

>>> import cffi

>>> cffi.__version__

'0.7.2'

>>>

I can't seem to figure what is going wrong. From the looks of it looks like the function is returning a valid pointer

but the cdata object returned by CFFI has the pointer address wrong. 

Has anyone seen this before? How did you resolve it?

Sarvi 

[Sarvi Shanmugham]

I upgarded to 0.8.1. and now my build hangs at the following

LD_LIBRARY_PATH=/ws/sarvi-sjc/mytools/lib:./DebugLNX: PYTHONPATH=./DebugLNX/py_install/lib/python2.7/site-packages: /ws/sarvi-sjc/mytools/bin/python setup.py install --prefix=./DebugLNX/py_install

In file included from ./expression.h:14,

                 from ./parse.h:13,

                 from __pycache__/_cffi_lib_x9cd4e6d7x4f3c2bc3.c:159:

./allocate.h:31:1: warning: multi-line comment

./allocate.h:40:1: warning: multi-line comment

In file included from ./expression.h:14,

                 from ./parse.h:13,

                 from __pycache__/_cffi_token_x415b8b8dx3952ec87.c:159:

./allocate.h:31:1: warning: multi-line comment

./allocate.h:40:1: warning: multi-line comment

__pycache__/_cffi_token_x415b8b8dx3952ec87.c: In function `_cffi_var_includepath':

__pycache__/_cffi_token_x415b8b8dx3952ec87.c:2280: warning: assignment from incompatible pointer type

In file included from ./expression.h:14,

                 from ./parse.h:13,

                 from __pycache__/_cffi_symbol_x3642b1a5xb28472c0.c:157:

./allocate.h:31:1: warning: multi-line comment

./allocate.h:40:1: warning: multi-line comment

Sarvi

[Sarvi Shanmugham]

The hang was case of looped import/ffi.include between 2 python wrapper files.

Fixed that. 

Now I am still seeing crash due to returned "struct token *" pointer changing from  0x2a9a81c010 to 0xffffffff9a81c010 between the return in the C function and cdata returned by the cffi wrapper to the C function.

Any idea what this might be caused by? Is this bug in CFFI or am I doing something wrong.

Interestingly an almost similar function than returns "struct token *" with a similar wrapper seems to be returning cdata/pointer correctly.

Any help.

Sarvi

[Sarvi Shanmugham]

I went through gdb to walk the path of return and I see the value from the "return ret" to "result=", go from 0x2a9a824010 to 0xffffffff9a824010

add_addr_space_attr_begin: 0x2a9a823f90

118             ret = create_char_token_after(ret, '(');

(gdb) n

119         printf("\nadd_addr_space_attr_begin: %p",ret);

(gdb) n

add_addr_space_attr_begin: 0x2a9a823fd0

120             return ret;

(gdb) print ret

$2 = (struct token *) 0x2a9a824010

(gdb) s

121     }

(gdb) s

_cffi_f_add_addr_space_attr_begin (self=<value optimized out>, arg0=<value optimized out>) at __pycache__/_cffi_api_mod_x382563ffxa7974013.c:404

404       _cffi_save_errno();

(gdb) list

399       }

400

401       Py_BEGIN_ALLOW_THREADS

402       _cffi_restore_errno();

403       { result = add_addr_space_attr_begin(x0); }

404       _cffi_save_errno();

405       Py_END_ALLOW_THREADS

406

407       return _cffi_from_c_pointer((char *)result, _cffi_type(1));

408     }

(gdb) print result

$3 = (struct token *) 0xffffffff9a824010

(gdb) print x0

$4 = (struct token *) 0x2a9a822f50

[Tyler Wade]

Hi Sarvi,

Your pointer is being truncated to 32-bits. Are you sure that the signature of the function is passed to ffi.verify (either directly or by #including the appropriate header)? If it isn't, the function won’t be declared in the C code that cffi generates and the compiler will assume it returns an int.

[Sarvi Shanmugham]

ok. looks like the problem was the verify and the generated code did not include the header file associated with the API. 

So this had generated a bunch of warnings but still compiled fine.

But the failed during execution.  Well, hopefully this help others.

Sarvi