Added Sub_Ovh implementation. #64
Conversation
I added the Sub_Ovh Implementation next is the Sub_Ovh_Un Implementation this is for ticket: CosmosOS#9
|
Can you please include a test which proves this change works as expected? |
Had to use a hack to fix stack problem with exceptions Added Sub_Ovf_Un IL Opcode
| } | ||
| // Let's check if we add overflow and if so throw OverflowException | ||
| XS.Jump(ConditionalTestEnum.NoOverflow, xSuccessLabel); | ||
| if (xSize > 4) // Hack to stop stack corruption |
There was a problem hiding this comment.
This is something we will have to consider on exception handling. When throwing in the middle of an opcode, this will probably work for most of the simple cases, but not sure about more complex ones. I'm not sure about what the CLI spec defines for the stack, still have to read that, but if the following is valid IL:
try
{
ldloc.0
ldloc.1
ldloc.2
add.ovf
add.ovf
}
catch
{
// do something
}
And we throw at the first add.ovf, the stack is not in a good state to jump to the catch block (still has the other local on the stack, which was expected to be consumed by the second add.ovf). Roslyn seems to emit good IL for this (not sure if by spec, still have to check), so this should work for now.
Also, I think we should file an issue about this or add this info to #16.
There was a problem hiding this comment.
According to ECMA-335, the throw opcode description is this:
The throw instruction throws the exception object (type O) on the stack and empties the stack.
So I think we should always empty the stack when throwing an exception, we should add some common infrastructure to throw exceptions, probably part of #16, I'll add a note there.
I added the Sub_Ovh Implementation next is the Sub_Ovh_Un Implementation this is for ticket: #9