/* Copyright (C) 1992, 1996, 1997 Aladdin Enterprises.  All rights reserved.
  
  This file is part of GNU Ghostscript.
  
  GNU Ghostscript is distributed in the hope that it will be useful, but
  WITHOUT ANY WARRANTY.  No author or distributor accepts responsibility to
  anyone for the consequences of using it or for whether it serves any
  particular purpose or works at all, unless he says so in writing.  Refer to
  the GNU General Public License for full details.
  
  Everyone is granted permission to copy, modify and redistribute GNU
  Ghostscript, but only under the conditions described in the GNU General
  Public License.  A copy of this license is supposed to have been given to
  you along with GNU Ghostscript so you can know your rights and
  responsibilities.  It should be in a file named COPYING.  Among other
  things, the copyright notice and this notice must be preserved on all
  copies.
  
  Aladdin Enterprises is not affiliated with the Free Software Foundation or
  the GNU Project.  GNU Ghostscript, as distributed by Aladdin Enterprises,
  does not depend on any other GNU software.
*/

/* dstack.h */
/* Definitions for the dictionary stack */
#include "istack.h"

/* Define the dictionary stack and systemdict. */
extern ref_stack d_stack;
extern ref ref_systemdict;
#define systemdict (&ref_systemdict)

/* Define the dictionary stack pointers. */
typedef s_ptr ds_ptr;
typedef const_s_ptr const_ds_ptr;
#define dsbot (d_stack.bot)
#define dsp (d_stack.p)
#define dstop (d_stack.top)

/* Macro to ensure enough room on the dictionary stack */
#define check_dstack(n)\
  if ( dstop - dsp < (n) )\
    { d_stack.requested = (n); return_error(e_dictstackoverflow); }

/* Check whether a dictionary is one of the permanent ones on the d-stack. */
bool dict_is_permanent_on_dstack(P1(const ref *));

/*
 * Switching between Level 1 and Level 2 involves inserting and removing
 * globaldict on the dictionary stack.  Instead of truly inserting and
 * removing entries, we replace globaldict by a copy of systemdict in
 * Level 1 mode.  min_dstack_size, the minimum number of entries, does not
 * change depending on language level; the countdictstack and dictstack
 * operators must take this into account.
 */
extern uint min_dstack_size;

/*
 * The dictionary stack is implemented as a linked list of blocks;
 * operators that access the entire d-stack must take this into account.
 * These are:
 *	countdictstack  dictstack
 * In addition, name lookup requires searching the entire stack, not just
 * the top block, and the underflow check for the dictionary stack
 * (`end' operator) is not just a check for underflowing the top block.
 */

/*
 * Cache a value for fast checking of def operations.
 * If the top entry on the dictionary stack is a writable dictionary,
 * dsspace is the space of the dictionary; if it is a non-writable
 * dictionary, dsspace = -1.  Then def is legal precisely if
 * r_space(pvalue) <= dsspace.  Note that in order for this trick to work,
 * the result of r_space must be a signed integer; some compilers treat
 * enums as unsigned, probably in violation of the ANSI standard.
 */
extern int dsspace;
#define dtop_can_store(pvalue) ((int)r_space(pvalue) <= dsspace)
/*
 * Cache values for fast name lookup.  If the top entry on the dictionary
 * stack is a readable dictionary with packed keys, dtop_keys, dtop_npairs,
 * and dtop_values are keys.value.packed, npairs, and values.value.refs
 * for that dictionary; otherwise, these variables point to a dummy
 * empty dictionary.
 */
extern const ref_packed *dtop_keys;
extern uint dtop_npairs;
extern ref *dtop_values;
/*
 * Reset the cached top values.  Every routine that alters the
 * dictionary stack (including changing the protection or size of the
 * top dictionary on the stack) must call this.
 */
void dict_set_top(P0());

/*
 * Define a special fast entry for name lookup in the interpreter.
 * The key is known to be a name; search the entire dict stack.
 * Return the pointer to the value slot.
 * If the name isn't found, just return 0.
 */
ref *dict_find_name_by_index(P1(uint nidx));
#define dict_find_name(pnref) dict_find_name_by_index(name_index(pnref))

/* Define the hashing function for names. */
/* We don't have to scramble the index, because */
/* indices are assigned in a scattered order (see name_ref in iname.c). */
#define dict_name_index_hash(nidx) (nidx)

/*
 * Define an extra-fast macro for name lookup, optimized for
 * a single-probe lookup in the top dictionary on the stack.
 * Amazingly enough, this seems to hit over 90% of the time
 * (aside from operators, of course, which are handled either with
 * the special cache pointer or with 'bind').
 */
#define dict_find_name_by_index_inline(nidx,htemp)\
  (dtop_keys[htemp = dict_hash_mod_inline(dict_name_index_hash(nidx),\
     dtop_npairs) + 1] == pt_tag(pt_literal_name) + (nidx) ?\
   dtop_values + htemp : dict_find_name_by_index(nidx))
/*
 * Define a similar macro that only checks the top dictionary on the stack.
 */
#define if_dict_find_name_by_index_top(nidx,htemp,pvslot)\
  if ( ((dtop_keys[htemp = dict_hash_mod_inline(dict_name_index_hash(nidx),\
	 dtop_npairs) + 1] == pt_tag(pt_literal_name) + (nidx)) ?\
	((pvslot) = dtop_values + (htemp), 1) :\
	0)\
     ) 

/*
Notes on dictionary lookup performance
--------------------------------------

We mark heavily used operations with a * below; moderately heavily used
operations with a +.

The following operations change the dictionary stack:
	+begin, +end
	readonly (on a dictionary that is on the stack)
	noaccess (on a dictionary that is on the stack)
We implement cleardictstack as a series of ends.

The following operations change the contents of dictionaries:
	*def, +put
	undef
	restore
	.setmaxlength
We implement store in PostScript, and copy as a series of puts.  Many
other operators also do puts (e.g., ScaleMatrix in makefont,
Implementation in makepattern, ...).  Note that put can do an implicit
.setmaxlength (if it has to grow the dictionary).

The following operations look up keys on the dictionary stack:
	*(interpreter name lookup)
	load
	where

Current design
--------------

Each name has a pointer that has one of 3 states:
	- This name has no definitions.
	- This name has exactly one definition, in systemdict or userdict.
	In this case, the pointer points to the value slot.
	- This name has some other status.

We cache some pointers to the top dictionary on the stack if it is a
readable dictionary with packed keys, which allows us to do fast,
single-probe lookups in this dictionary.  We also cache a value that
allows us to do a fast check for stores into the top dictionary
(writability + space check).

Full shallow binding
--------------------

We implement shallow binding with a pointer in each name that points to
the value slot that holds the name's definition.  If the name is
undefined, or if we don't know where the slot is, the binding pointer
points to a ref with a special type t__invalid, which cannot occur
anywhere else.  "Clearing" the pointer means setting it to point to this
ref.

We also maintain a pair of pointers that bracket the value region of the
top dictionary on the stack, for fast checking in def.  If the top
dictionary is readonly or noaccess, the pointers designate an empty area.
We call this the "def region" cache.

We implement the above operations as follows:
	begin - push the dictionary on the stack; set the pointers of
		all name keys to point to the corresponding value slots.
	end - pop the stack; clear the pointers of all name keys.
	readonly - if the dictionary is the top one on the stack,
		reset the def region cache.
	noaccess - clear the pointers of all name keys.  (This is overly
		conservative, but this is a very rare operation.)
		Also reset the def region cache if the dictionary is
		the top one on the stack.
	def - if the key is a name and its pointer points within the cached
		def region, store the value through the pointer; otherwise,
		look up the key in the top dictionary, store the value,
		and if the key is a name, set its pointer to the value slot.
	put - if the key is a name and wasn't in the dictionary before,
		clear its pointer.  (Conservative, but rare.)
	undef - if the key is a name, clear its pointer.  (Overly
		conservative, but rare.)
	restore - if either the old or the new value of a change is a name
		(possibly in a packed array), clear its pointer.  This is
		conservative, but easy to detect, and probably not *too*
		conservative.
	.setmaxlength - clear all the pointers, like noaccess.
	(name lookup) - fetch the value through the pointer and dispatch
		on its type; if the type is t__invalid, do a full search
		and set the pointer.  This avoids a separate check for a
		clear pointer in the usual case where the pointer is valid.
	load - if the pointer is clear, do a search and set the pointer;
		then fetch the value.
	where - always do a full search and set the pointer.
		(Conservative, but rare.)

One place where shallow binding will result in major new overhead is the
extra push of systemdict for loading fonts.  This probably isn't a problem
in real life.

Adaptive shallow binding
------------------------

We do validity checking for the name value cache using an epoch counter.
For each dictionary D, we keep an on-stack flag F.  Each dictionary stack
entry is <D,M,F,E> where D is the actual dictionary, M is a mark vector of
V bits (V is a system constant, probably 64), F is D's former on-stack
flag, and E is the epoch at which the entry was made.  For each name K, we
keep a cache <P,E> where P is a pointer to the dictionary value slot that
holds the current value of K, and E is an epoch value; the cache is valid
if K->E >= dsp->E.  Here is what happens for each operation:

****** Still need to handle names defined only in systemdict or userdict?

To initialize:
	Epoch = 0
To clear the cache entry for K:
	*K = <ptr to invalid value, 0>
begin(D):
	*++dsp = <D, {0...}, D->F, ++Epoch>
	set D->F
value = lookup(K):
	if K->E >= dsp->E
	  value = *K->P
	else
	  do lookup as usual
	  *K = <ptr to value, Epoch>
	  set dp->M[i mod V] where dp is the dstack slot of the dictionary
	    where K was found and i is the index within that dictionary
end:
	for each i such that dsp->M[i] is set,
	  clear the cache entry for dsp->D->keys[i, i+V, ...]
	dsp->D->F = dsp->F
	--dsp
noaccess(D):
	if D->F is set,
	  clear the cache entries for all name keys of D
readonly(D):
	<< nothing >>
.setmaxlength(D,N):
	same as noaccess
restore:
	If either the old or the new value of a change is a name
	  (possibly in a packed array), clear its cache entry.  This is
	  conservative, but easy to detect, and probably not *too*
	  conservative.
def(K,V):
	if K->P points into dsp->D
	  *K->P = V
	else
	  put the new value in dsp->D
	  set *K and dsp->M[i mod V] as for a lookup
put(D,K,V):
	if K is already defined in D, do nothing special
	otherwise, if D->F isn't set, do nothing special
	otherwise, clear K's cache entry
undef(D,K):
	if D->F is set,
	  clear K's cache entry
*/