flex/bison fixing memory leaks with unexpected tokens - memory-leaks

Question

I have a flex bison application. For a few of my tokens, I copy out the yytext from flex using strdup. This works great except when there is an error of an unexpected token.
simple example
flex.l:
...
[a-zA-Z0-9]+ { lval.string = strdup(yytext); return IDENT };
[\{\}] { return yytext[0] };
...
and
parse.y
...
%destructor { free($$); } IDENT
%destructor { free($$->name); free($$->type); free($$); } tag
...
tag: IDENT '{' IDENT '}'
{
struct tag *mytag = malloc(sizeof(struct tag));
mytag->name = $1;
mytag->type = $3;
$<tag>$ = mytag;
}
...
Now suppose I hand it the input:
blah blah blah
The lexer will send up the first IDENT token, which gets pushed onto the stack. After the first token it's expecting a bracket token, but instead gets another IDENT token. This is a syntax error. The destructor will be called on the first IDENT token, but not on the second one (the unexpected one). I haven't been able to find a way to destruct the unexpected token. Does anyone know how I should do it?

Answer 1

I found that appropriate use of the 'error' token in flex prompts it to correctly call the destructor function. Go me!
parse.y
...
%destructor { free($$); } IDENT
%destructor { free($$->name); free($$->type); free($$); } tag
...
tags: tag tags | error tags | ;
tag: IDENT '{' IDENT '}'
{
struct tag *mytag = malloc(sizeof(struct tag));
mytag->name = $1;
mytag->type = $3;
$<tag>$ = mytag;
}
...