FunctionCache.mesa
Michael Plass, June 19, 1985 9:42:48 pm PDT
Copyright © 1985 by Xerox Corporation. All rights reserved.
FunctionCache: CEDAR DEFINITIONS
~ BEGIN
Documentation:
This module provides a facility for writing cached implementations of REF-valued functions. The caches are managed with a move-to-front discipline, and trimmed according to the client supplied limits on the total number of entries or the total number of words consumed.
Cache: TYPE ~ REF CacheRep;
CacheInfo: TYPE ~ RECORD [nEntries: INT, maxEntries: INT, totalSize: INT, maxTotalSize: INT];
Create: PROC [maxEntries: INT ← 10, maxTotalSize: INTINT.LAST] RETURNS [Cache];
ClientID: TYPE ~ REF; -- to allow otherwise independent clients to share cache space.
Domain: TYPE ~ REF;
Range: TYPE ~ REF;
Eval: PROC [x: Cache, F: FuncProc, arg: Domain, match: Match, clientID: ClientID ← NIL] RETURNS [Range];
This is a convenience when the client already has an argument REF allocated; use the Lookup proc, below, to do more general matching or to avoid early argument allocation.
FuncProc: TYPE ~ PROC [Domain] RETURNS [value: Range, size: INT];
Match: TYPE ~ {refs, case, text} ← refs;
refs ~ ref equality
case ~ understands REF INT, REF REAL, REF CHAR, LIST OF REF, ROPE, ATOM; otherwise uses ref equality
text ~ like above, but ignores case on REF CHARs, ATOMs and ROPEs
Insert: PROC [x: Cache, argument: Domain, value: Range, size: INT, clientID: ClientID ← NIL];
CompareProc: TYPE ~ PROC [argument: Domain] RETURNS [good: BOOL];
Any: CompareProc; -- returns TRUE
Equal: PROC [a, b: Domain, match: Match] RETURNS [BOOL];
The fancy compare used by Eval;
Lookup: PROC [x: Cache, compare: CompareProc, clientID: ClientID ← NIL] RETURNS [value: Range, ok: BOOL];
Allows client to inspect each entry that matches clientID and decide if it meets the requirements; compare is called under a monitor, so make sure it is bulletproof. If compare returns TRUE for some entry, Lookup returns the value of that entry with ok=TRUE; otherwise returns [NIL, FALSE].
CacheEntry: TYPE ~ RECORD [clientID: ClientID, argument: Domain, value: Range, size: INT];
Obtain: PROC [x: Cache, compare: CompareProc, limit: INT ← 1, clientID: ClientID ← NIL] RETURNS [LIST OF CacheEntry];
Gains exclusive use of the entries by removing them from the cache. The result has no more than n entries.
Release: PROC [x: Cache, list: LIST OF CacheEntry];
Puts the entries back in the cache.
GetInfo: PROC [x: Cache] RETURNS [CacheInfo];
GetList: PROC [x: Cache] RETURNS [LIST OF CacheEntry];
Copies the list under the monitor.
SetLimits: PROC [x: Cache, maxEntries: INT ← 10, maxTotalSize: INTINT.LAST];
SetValueSize: PROC [x: Cache, value: Range, size: INT];
GlobalCache: PROC RETURNS [Cache];
Use this for applications where the entry sizes are large, so that all such clients can compete fairly for the available memory. Use individual caches when the entries are small; this will keep the cache lookup time down.
CacheRep: TYPE ~ PRIVATE MONITORED RECORD [
nEntries: INT,
maxEntries: INT,
totalSize: INT,
maxTotalSize: INT,
entryList: LIST OF CacheEntry
];
END.