Re: [RFA] Integrated hashtables take 2

[Per Bothner <per at bothner dot com>]

How about:

struct cpp_identifier
{
  int length;
  const char* pointer;
};

struct tree_identifier
{
  struct tree_common common;
  struct cpp_identifier ident;
};
#define IDENTIFIER_LENGTH(NODE) \
  (IDENTIFIER_NODE_CHECK (NODE)->identifier.ident.length)
#define IDENTIFIER_POINTER(NODE) \
  (IDENTIFIER_NODE_CHECK (NODE)->identifier.ident.pointer)

Then we have two choices:

(a) The hashtable contains pointers to the cpp_identifier
part.  For the front-end to get the tree_identifier it has
to subtract the offset of ident in tree_identifier.
There would be a cpp_alloc_identifier function, which the
front-end could define as:

struct cpp_identifier *
cpp_alloc_identifier ()
{
  tree id = make_node (IDENTIFIER_NODE);
  return &id->ident;
}

(b) The hashtable contains (void*) pointers to the tree_identifier
object.  The front-end registers with cpplib (i.e. a global
int variable) the offset of the cpp_identifier part, or
a function that converts the (void*) pointer to tree_identifier
to one that return the cpp_identifier part.  (That latter is
cleaner but slower.)
-- 
	--Per Bothner
per@bothner.com   http://www.bothner.com/per/