Re: [Boost-users] Cost of throwing exceptions - are they just non-local goto?

From: Paul Johnson (gt54-boost_at_[hidden])
Date: 2005-06-24 03:04:45

Tan Kwee Heong wrote:
> Hi,
> 
> I wish to reconcile the information provided from Scott Meyer's More
> Effective C++ Item 15, with that
> of Peter Dimov. Could someone post some URLs/referencess of Peter's
> comments, for
> my benefit?
> 
> Thanks,
> 
> "Vladimir Prus" <ghost_at_[hidden]> wrote in message
> news:d9dopb$3tg$1_at_sea.gmane.org...
> 
>>Thomas Costa wrote:
>>
>>1. As Peter Dimov said (IIRC), exceptions are just non-local goto. In this
>>case, non-local goto is fine.

<disclaimer>
I didn't see Peter's comments.
</disclaimer>

When you execute a goto, you still have to unwind any enclosing blocks 
in your function. If you jump out of a (nested) block, the compiler 
destructs any objects which are local to those blocks. In this sense, 
gotos and exceptions are similar.

Scott Meyer estimates that returning from a function via an exception 
could be ~1000x slower than a normal return. However, exiting a block 
via a goto is probably only marginally slower than a normal exit, since 
the block still has to be unwound anyway (clean up stack/call 
destructors) even in a normal exit.

PJ