ctypes and setjmp

[Richard Jones]

Currently ctypes can't play well with any C code that requires use of setjmp
as part of its API.

libpng is one of those libraries.

Can anyone think of a reasonable solution to this? Perhaps ctypes might be
patched to offer setjmp support in foreign function definitions?


Richard

 

[Thomas Heller]

Currently ctypes can't play well with any C code that requires use of setjmp
as part of its API.

libpng is one of those libraries.

Can anyone think of a reasonable solution to this? Perhaps ctypes might be
patched to offer setjmp support in foreign function definitions?


Richard

I didn't know that setjmp/longjmp is actually used by production libraries
for error handling.

How is this pattern used in C? Do you call setjump() before each api call,
or do you call setjump once, and then do all the api calls? What do you do
when setjmp() returns != 0? Exit the program? Log a message? How do you
determine which call failed? How do you pass the jmp_buf that you passed to
setjmp() to the api call?

For ctypes, the only solution I can think of is to invent a new calling
convention, which will call setjmp() first internally before calling the
libpng api function...

Thomas

 

[Richard Jones]

I didn't know that setjmp/longjmp is actually used by production libraries
for error handling.

To be honest, I didn't even know what setjmp/longjmp *were* until I hit its
use in libpng, and I've been programming C for over a decade now

How is this pattern used in C? Do you call setjump() before each api
call, or do you call setjump once, and then do all the api calls?

I believe you call setjmp just before calling the API function in question.

What do you do when setjmp() returns != 0?

I'd say that in ctypes land you'd have to provide a function to call in that
situation. ctypes would call that function and then continue on. It's up to
that function to raise an exception or otherwise flag the error. So in
rough terms, you'd have ctypes do:

if (error_handler != NULL) r = 0;
else r = setjmp();
if (r == 0) api_call();
if (r == 1) error_handler();

For ctypes, the only solution I can think of is to invent a new calling
convention, which will call setjmp() first internally before calling the
libpng api function...

Yeah, that's what I figured.


Richard

 

[Gabriel Genellina]

I didn't know that setjmp/longjmp is actually used by production libraries
for error handling.

Using setjmp to report errors in a general library is, uhm, a bit
crazy at least!
C programmers are certainly crazy people

How is this pattern used in C? Do you call setjump() before each api call,
or do you call setjump once, and then do all the api calls? What do you do
when setjmp() returns != 0? Exit the program? Log a message? How do you
determine which call failed? How do you pass the jmp_buf that you passed to
setjmp() to the api call?

The setjmp must be on scope when a longjmp is called. Since there is
no endjmp/canceljmp, this in the practice means that you have to call
setjmp in *every* function that calls another one which may issue a longjmp.
There are other alternatives in plain C (cexcept:
http://cexcept.sourceforge.net/) but that wont help you with ctypes.

For ctypes, the only solution I can think of is to invent a new calling
convention, which will call setjmp() first internally before calling the
libpng api function...

May be reasonable - a non-zero in setjmp would raise a Python
exception. But perhaps a new calling convention is not needed: as you
need a way to indicate *which* argument is the jmp_buf, just a new
argument type would suffice. If you encounter one such thing in the
function.argtypes list, that means a setjmp is needed before calling
the actual function.



Gabriel Genellina
Softlab SRL





__________________________________________________
Preguntá. Respondé. Descubrí.
Todo lo que querías saber, y lo que ni imaginabas,
está en Yahoo! Respuestas (Beta).
¡Probalo ya!
http://www.yahoo.com.ar/respuestas

 

[Richard Jones]

May be reasonable - a non-zero in setjmp would raise a Python
exception. But perhaps a new calling convention is not needed: as you
need a way to indicate *which* argument is the jmp_buf, just a new
argument type would suffice. If you encounter one such thing in the
function.argtypes list, that means a setjmp is needed before calling
the actual function.

The ctypes C code which handles invoking the foreign function call must be
responsible for calling setjmp(). You can't call that in Python code - the
C stack is manipulated all over the place for each line of Python code
executed. You have to both call and set the jmp_buf in the same C code call
from Python.

Thus my suggestion of configuring a handler function on the foreign function
wrapper object alongside the argtypes and restype. If one is defined then
ctypes automatically calls setjmp and passes control to that handler
function if setjmp returns non-zero.

This page:

http://www.massey.ac.nz/~mgwalker/misc/create-images.html

has some example programs that show setjmp/longjmp in action. Read the whole
thing as his first examples aren't correct


Richard