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

[Rick Mann]

On Jan 19, 2009, at 11:53:57, Gavin Lambert wrote:

> There's actually no technical reason (from a language standpoint)  
> why "finally" couldn't be supported in C/C++ language targets (apart  
> from the fact that there isn't a C++ target yet for v3).
>
> I don't know if it *is* actually supported, but it could be.


Hmm. I thought I had a pretty good grasp of C++. It lacks a "finally"  
keyword, which means you implement automatic-code-execution-on-scope- 
exit with RAII. This requires instantiating objects on the stack at  
appropriate places. This is fine, if sometimes cumbersome, but I  
wanted to avoid letting someone put "finally" in their grammar.

On Jan 19, 2009, at 11:56:47, Wacek Kusnierczyk wrote:

> Rick Mann wrote:
>> In a language like C++, which stupidly doesn't support "finally",
>>
> you might want to read what the designers of c++ have to say in this
> matter before you make such judgments:
>
> http://www.research.att.com/~bs/bs_faq2.html#finally

Yeah, I love C++, really, but I disagree with the creators/sustainers  
on a few issues. This is one. I love that I can use RAII, and often it  
is better, but frequently you have one-off cleanup that is much more  
compactly written in a finally clause. There's no good reason not to  
have both.

On Jan 19, 2009, at 12:00:29, Sam Harwell wrote:

> I assume you're trying to write a new target? If so, you can do
> something like this, with nested try/catch and a duplication of the
> finally code:
>
> try {
>  try {
>    <tryCode>
>  } catch ( ExceptionType1 ) {
>    <handler>
>  }
> } catch ( ... ) {
>  <finallyCode>
>  throw;
> }
> <finallyCode>

An interesting approach, but a bit icky, don't you think? If you can  
guarantee a superclass to the exceptions, it could be a bit tidier,  
perhaps, but really, C++? Can't we just have finally already? :-)

Anyway, I guess I have my answer. Thanks!

-- 
Rick