C api and exception handling

Carl Banks pavlovevidence at gmail.com
Mon Nov 2 18:01:44 CET 2009

On Nov 2, 7:16 am, "lallous" <lall... at lgwm.org> wrote:
> Hello,
> Is there is a way, using the Python C api, to install an exception handler
> that:
> - will be triggered when an exception occurs
> - analyze the reason of the exception
> - correct the situation and try again (something like exception handling on
> windows where the exception handler can retrieve the registers
> context->faulting instruction->fix situation if needed->restart execution
> from the same point)

Python has no concept of "retrying", at either the Python or C API
level.  You might be able to do something Evil in C to get this effect
but I don't recommend it, it'll be fundamentally averse to how Python
works and future versions are likely to break it.

> Since I will be asked: "what are you trying to achieve?", this is what I
> want:
> func_call("hello") <- no exceptions, good code: function is defined and
> called properly
> SomeUndefinedFunction("x", "y") <- undefined function call will trigger an
> exception. I want my python/C exception handler to inspect the reason of the
> exception, if it was a call to an undefined function call then redirect the
> execution to a certain method, say: ExecuteByName("SomeUndefinedFunction",
> "x", "y")
> I know if I create a small class with getattr hooked, what I want can be
> achieved.

I'd do it that way.  There is ordinarily no way to hook into a plain
function call like SomeUndefinedFunction() in Python; if you go around
hacking up the interpreter to do that users will be highly confused
and surprised.

OTOH, hooking into attributes is pretty well-known.  When a person
sees attribute notation they know there's an opportunity to do weird
stuff.  When a strange function is called, they will be like, "oh,
someone overrode __getattr__".

> But can it be done otherwise (without using a class and instead relying on
> exception handlers and correcting the exception)?

Just forget about exception handling.  If you REALLY insist on doing
this, and I highly recommend against it, the best chance you have is
to try to hook into the importing process and load a module that uses
a custom dictionary object.

Carl Banks

More information about the Python-list mailing list