Re: RFC: generate the default semantic action

[Paul Eggert]

Akim Demaille wrote:

> This patch is prompted by C++, but I believe it’s a real improvement
> even for the other languages.  However, it does mean that the default
> action is now made twice in C for instance.  I would very much like
> to disable the pre-initialization of $$ to $1, but I’m afraid people
> might depend on this…

As I recall, in the traditional Yacc implementation the default '$$ = $1;' 
action was implemented by doing nothing (i.e., zero C code and zero machine 
instructions), as $1 was the top value on the stack, so popping and pushing the 
stack was a no-op.

I also recall that at some point Bison put in an explicit '$$ = $1;', run before 
any action, because otherwise GCC complained in some cases (I think about 
uninitialized variables).

I don't know the motivation for doing the equivalent of '$$ = $1;' both before 
all actions, and explicitly for missing actions. I'm skeptical that we should do 
both, because in that case I expect that we'll miss some useful opportunities 
for GCC to generate warnings about uninitialized variables.

User code should not depend on any of these details; it should work even if we 
went back to the Yacc way of doing things, for C. Also, I wouldn't worry about 
any C code that does depend on these details, as I doubt whether there is any. 
So for the benefit of GCC warnings, I suggest just doing things the new way.