[antlr-interest] Handling 'finally' in language target that doesn't have 'finally'

[Gavin Lambert]

At 09:27 20/01/2009, Sam Harwell wrote:
 >For the sake of creating an ANTLR target (I think that's
 >the real goal?), you have a few options. If you simply
 >want to disallow finally code in the grammar when the
 >C++ target is used, make the template for finally code
 >spit out the following if the finally code is not blank.
 >
 >#error The C++ target does not support 'finally' blocks in
 >grammars. Use RAII instead.

Sure, you could do that, but why not just make it work?

As you posted earlier, any psuedoblock:

try
{
   [a]
}
finally
{
   [b]
}

is exactly identical (in terms of behaviour) to this C++ code:

try
{
   [a]
}
catch(...)
{
   [b]
   throw;
}
{
   [b]
}

(Note the extra braces at the end; that's to keep the variable 
scoping clean.  I guess if you're ultra-paranoid, you could wrap 
the whole thing in another set, since technically the first is a 
single statement while the second is two statements.  But that's 
usually not an issue.)

Sure, it's a little messier than using RAII, but that's just an 
incentive to use RAII wherever possible.  But if you really want 
finally-like behaviour without RAII (eg. to support the ANTLR 
grammar keyword), then it's trivial to add.  (It's even more 
trivial when using a template or codegen like ANTLR, since the 
duplication of [b] will only appear in the generated file, not the 
source file.  And the generated code is ugly anyway, so it doesn't 
matter so much.)