Re: %destructor and unused values

[John P. Hartmann]

Thanks, Akim.

Right, so it is the presence of a destructor that enables this
checking.  Might it be worth mentioning in the documentation?

The example in 3.8.7 shows how to use

%destructor { } <character>

it probably confused me.  The explanation certainly doesn't tell me
all the ramifications.

-----

Trying a token that does need a destructor I chose a struct term,
which is passed as an lvalue (that is, the entire struct, not a
pointer).  Thus struct term does not require a destructor.  However,
it contains an optional pointer to a string that must be destroyed and
the token clearly identifies which cases have a pointer.  Clearly one
cannot expect Bison to understand such subtlety.

I had to add a variant to the %union (struct term termv) so that those
cases that actually need to be destroyed were identified as such.  Of
course this upsets the applecart in a slew of function calls, but

#define termv term

took care of that.  Thus the struct yylval that the lexxer sees
remains unchanged.

(Yes I use a struct yylval instead of a union; then incorrect usage
soon results in a null pointer being dereferenced rather than one
struct being mistaken for another.)

On 28 January 2013 13:35, Akim Demaille <address@hidden> wrote:
> Hi John,
>
> If you declared destructors for parsehow and parsecase,
> it is understood as "these symbols allocated memory
> that needs to be reclaimed before the symbol is destroyed".
>
> Bison can "see" that you are not using these symbols
> in the action, so you are certainly leaking them.  The
> parser will _not_ reclaim them: %destructor is there
> for error recovery, not to reclaim the memory
> of rhs symbols in an action reduction.