I have 100 or so trampoline functions. I would like to know whether it is possible to automate wrapping each one inside a try/catch block.
Please be warned in advance, this is not an easy question. I will start by describing the problem with (simplified) code, and will then attempt to answer it as best I can below, so the reader may see where I am at.
Foo has a function pointer table:
EDIT: This is a C function pointer table. So it could accept static W::w
.
Signatures are here: http://svn.python.org/projects/python/trunk/Include/object.h
EDIT: I've attempted a test case here:
class Foo {
Table table;
Foo() {
// Each slot has a default lambda.
:
table->fp_53 = [](S s, A a, B b) -> int {cout<<"load me!";};
table->fp_54 = [](S s, C c, D d, E e) -> float {cout<<"load me!";};
// ^ Note: slots MAY have different signatures
// only the first parameter 'S s' is guaranteed
}
// Foo also has a method for loading a particular slot:
:
void load53() { table->fp_53 = func53; }
void load54() { table->fp_54 = func54; }
:
}
If a particular slot is 'loaded', this is what gets loaded into it:
int func53(S s, A a, B b) {
try{
return get_base(s)->f53(a,b);
}
catch(...) { return 42;}
}
float func54(S s, C c, D d, E e) {
try{
return get_base(s)->f54(c,d,e);
}
catch(...) { return 3.14;}
}
I am trying to accomplish this using lambdas, so as to bypass having to define all of these func53
separately. Something like this:
class Foo {
:
void load53() {
table->fp_53 =
[](S s, A a, B b)->int { return get_base(s)->f53(a,b); }
}
void load54() {
table->fp_54 =
[](S s, C c, D d, E e)->float { return get_base(s)->f54(c,d,e); }
}
However, this is failing to trap errors. I need to be putting a try/catch around the return statement:
try{ return get_base(s)->f53(a,b); } catch{ return 42; }
However, this creates a lot of clutter. It would be nice if I could do:
return trap( get_base(s)->f53(a,b); )
My question is: is there any way to write this trap
function (without using #define)?
This is what I've come up with so far:
I think this would pass all the necessary information:
trap<int, &Base::f53>(s,a,b)
trap's definition could then look like this:
template<typename RET, Base::Func>
static RET
trap(S s, ...) {
try {
return get_base(s)->Func(...);
}
catch {
return std::is_integral<RET>::value ? (RET)(42) : (RET)(3.14);
}
}
This may allow for a very clean syntax:
class Foo {
:
void load53() { table->fp_53 = &trap<int, &Base::f53>; }
void load54() { table->fp_54 = &trap<float, &Base::f54>; }
}
At this point I'm not even sure whether some laws have been violated. table->fp_53
must be a valid C function pointer.
Passing in the address of a nonstatic member function (&Base::f53>
) won't violate this, as it is a template parameter, and is not affecting the signature for trap
Similarly, ...
should be okay as C allows varargs.
So if this is indeed valid, can it be cleaned up?
My thoughts are:
1) maybe the ... should be moved back to the template parameter as a pack.
2) maybe it is possible to deduce the return type for trap, and save one template parameter
3) that Base::Func
template parameter is illegal syntax. And I suspect it isn't even close to something legal. Which might scupper the whole approach.
#include <utility>
template <typename T, T t>
struct trap;
template <typename R, typename... Args, R(Base::*t)(Args...)>
struct trap<R(Base::*)(Args...), t>
{
static R call(int s, Args... args)
{
try
{
return (get_base(s)->*t)(std::forward<Args>(args)...);
}
catch (...)
{
return std::is_integral<R>::value ? static_cast<R>(42)
: static_cast<R>(3.14);
}
}
};
Usage:
table->fp_53 = &trap<decltype(&Base::f53), &Base::f53>::call;
table->fp_54 = &trap<decltype(&Base::f54), &Base::f54>::call;
Note: std::forward
can still be used although Args
is not a forwarding reference itself.