1 # Version 0.05 alpha $Revision: 1.6 $ $Date: 2001/06/24 17:11:26 $
9 LIST_CACHE doesn't work with ties to most DBM implementations, because
10 Memouze tries to save a listref, and DB_File etc. can only store
11 strings. This should at least be documented. Maybe Memoize could
12 detect the problem at TIE time and throw a fatal error.
14 20010623 This was added sometime prior to 20001025.
16 Try out MLDBM here and document it if it works.
20 We should extend the benchmarking module to allow
22 timethis(main, { MEMOIZED => [ suba, subb ] })
24 What would this do? It would time C<main> three times, once with
25 C<suba> and C<subb> unmemoized, twice with them memoized.
27 Why would you want to do this? By the third set of runs, the memo
28 tables would be fully populated, so all calls by C<main> to C<suba>
29 and C<subb> would return immediately. You would be able to see how
30 much of C<main>'s running time was due to time spent computing in
31 C<suba> and C<subb>. If that was just a little time, you would know
32 that optimizing or improving C<suba> and C<subb> would not have a
33 large effect on the performance of C<main>. But if there was a big
34 difference, you would know that C<suba> or C<subb> was a good
35 candidate for optimization if you needed to make C<main> go faster.
41 Perhaps C<memoize> should return a reference to the original function
42 as well as one to the memoized version? But the programmer could
43 always construct such a reference themselves, so perhaps it's not
44 necessary. We save such a reference anyway, so a new package method
45 could return it on demand even if it wasn't provided by C<memoize>.
46 We could even bless the new function reference so that it could have
47 accessor methods for getting to the original function, the options,
54 The TODISK feature is not ready yet. It will have to be rather
55 complicated, providing options for which disk method to use (GDBM?
56 DB_File? Flat file? Storable? User-supplied?) and which stringizing
57 method to use (FreezeThaw? Marshal? User-supplied?)
63 Maybe an option for automatic expiration of cache values? (`After one
64 day,' `After five uses,' etc.) Also possibly an option to limit the
65 number of active entries with automatic LRU expiration.
67 You have a long note to Mike Cariaso that outlines a good approach
68 that you sent on 9 April 1999.
70 What's the timeout stuff going to look like?
72 EXPIRE_TIME => time_in_sec
73 EXPIRE_USES => num_uses
76 perhaps? Is EXPIRE_USES actually useful?
78 19990916: Memoize::Expire does EXPIRE_TIME and EXPIRE_USES.
79 MAXENTRIES can come later as a separate module.
83 Put in a better example than C<fibo>. Show an example of a
84 nonrecursive function that simply takes a long time to run.
85 C<getpwuid> for example? But this exposes the bug that you can't say
86 C<memoize('getpwuid')>, so perhaps it's not a very good example.
88 Well, I did add the ColorToRGB example, but it's still not so good.
89 These examples need a lot of work. C<factorial> might be a better
94 Add more regression tests for normalizers.
98 Maybe resolve normalizer function to code-ref at memoize time instead
99 of at function call time for efficiency? I think there was some
100 reason not to do this, but I can't remember what it was.
104 Add more array value tests to the test suite.
106 Does it need more now?
110 Fix that `Subroutine u redefined ... line 484' message.
116 Get rid of any remaining *{$ref}{CODE} or similar magic hashes.
120 There should be an option to dump out the memoized values or to
121 otherwise traverse them.
125 Maybe the tied hash interface taskes care of this anyway?
129 Include an example that caches DNS lookups.
133 Make tie for Storable (Memoize::Storable)
135 A prototype of Memoize::Storable is finished. Test it and add to the
142 Make tie for DBI (Memoize::DBI)
146 I think there's a bug. See `###BUG'.
150 Storable probably can't be done, because it doesn't allow updating.
151 Maybe a different interface that supports readonly caches fronted by a
152 writable in-memory cache? A generic tied hash maybe?
155 if (it's in the memory hash) {
157 } elsif (it's in the readonly disk hash) {
165 put it into the in-memory hash
168 Maybe `save' and `restore' methods?
170 It isn't working right because the destructor doesn't get called at
173 This is fixed. `use strict vars' would have caught it immediately. Duh.
177 Don't forget about generic interface to Storable-like packages
179 20010627 It would appear that you put this into 0.51.
183 Maybe add in TODISK after all, with TODISK => 'filename' equivalent to
185 SCALAR_CACHE => [TIE, Memoize::SDBM_File, $filename, O_RDWR|O_CREAT, 0666],
190 Maybe the default for LIST_CACHE should be MERGE anyway.
194 There's some terrible bug probably related to use under threaded perl,
195 possibly connected with line 56:
197 my $wrapper = eval "sub { unshift \@_, qq{$cref}; goto &_memoizer; }";
199 I think becayse C<@_> is lexically scoped in threadperl, the effect of
200 C<unshift> never makes it into C<_memoizer>. That's probably a bug in
201 Perl, but maybe I should work around it. Can anyone provide more
202 information here, or lend me a machine with threaded Perl where I can
203 test this theory? Line 59, currently commented out, may fix the
206 20010623 Working around this in 0.65, but it still blows.
210 Maybe if the original function has a prototype, the module can use
211 that to select the most appropriate default normalizer. For example,
212 if the prototype was C<($)>, there's no reason to use `join'. If it's
213 C<(\@)> then it can use C<join $;,@$_[0];> instead of C<join $;,@_;>.
217 Ariel Scolnikov suggests using the change counting problem as an
218 example. (How many ways to make change of a dollar?)
222 Jonathan Roy found a use for `unmemoize'. If you're using the
223 Storable glue, and your program gets SIGINT, you find that the cache
224 data is not in the cache, because Perl normally writes it all out at
225 once from a DESTROY method, and signals skip DESTROY processing. So
228 $sig{INT} = sub { unmemoize ... };
233 This means it would be useful to have a method to return references to
234 all the currently-memoized functions so that you could say
236 $sig{INT} = sub { for $f (Memoize->all_memoized) {
244 19990917 There should be a call you can make to get back the cache
245 itself. If there were, then you could delete stuff from it to
246 manually expire data items.
250 19990925 Randal says that the docs for Memoize;:Expire should make it
251 clear that the expired entries are never flushed all at once. He
252 asked if you would need to do that manually. I said:
254 Right, if that's what you want. If you have EXISTS return false,
255 it'll throw away the old cached item and replace it in the cache
256 with a new item. But if you want the cache to actually get smaller,
257 you have to do that yourself.
259 I was planning to build an Expire module that implemented an LRU
260 queue and kept the cache at a constant fixed size, but I didn't get
261 to it yet. It's not clear to me that the automatic exptynig-out
262 behavior is very useful anyway. The whole point of a cache is to
263 trade space for time, so why bother going through the cache to throw
264 away old items before you need to?
266 Randal then pointed out that it could discard expired items at DESTRoY
267 or TIEHASH time, which seemed like a good idea, because if the cache
268 is on disk you might like to keep it as small as possible.
272 19991219 Philip Gwyn suggests this technique: You have a load_file
273 function that memoizes the file contexts. But then if the file
274 changes you get the old contents. So add a normalizer that does
276 return join $;, (stat($_[0])[9]), $_[0];
278 Now when the modification date changes, the true key returned by the
279 normalizer is different, so you get a cache miss and it loads the new
280 contents. Disadvantage: The old contents are still in the cache. I
281 think it makes more sense to have a special expiration manager for
282 this. Make one up and bundle it.
284 19991220 I have one written: Memoize::ExpireFile. But how can you
285 make this work when the function might have several arguments, of
286 which some are filenames and some aren't?
290 19991219 There should be an inheritable TIEHASH method that does the
291 argument processing properly.
293 19991220 Philip Gwyn contributed a patch for this.
295 20001231 You should really put this in. Jonathan Roy uncovered a
296 problem that it will be needed to solve. Here's the problem: He has:
299 LIST_CACHE => ["TIE", "Memoize::Expire",
301 TIE => ["DB_File", "debug.db", O_CREAT|O_RDWR, 0666]
304 This won't work, because memoize is trying to store listrefs in a
305 DB_File. He owuld have gotten a fatal error if he had done this:
308 LIST_CACHE => ["TIE", "DB_File", "debug.db", O_CREAT|O_RDWR, 0666]'
311 But in this case, he tied the cache to Memoize::Expire, which is *not*
312 scalar-only, and the check for scalar-only ties is missing from
313 Memoize::Expire. The inheritable method can take care of this.
315 20010623 I decided not to put it in. Instead, we avoid the problem by
316 getting rid of TIE. The HASH option does the same thing, and HASH is
317 so simple to support that a module is superfluous.
321 20001130 Custom cache manager that checks to make sure the function
322 return values actually match the memoized values.
326 20001231 Expiration manager that watches cache performance and
327 accumulates statistics. Variation: Have it automatically unmemoize
328 the function if performance is bad.
332 20010517 Option to have normalizer *modify* @_ for use by memoized
333 function. This would save code and time in cases like the one in the
334 manual under 'NORMALIZER', where both f() and normalize_f() do the
335 same analysis and make the same adjustments to the hash. If the
336 normalizer could make the adjustments and save the changes in @_, you
337 wouldn't have to do it twice.
340 20010623 Add CLEAR methods to tied hash modules.
343 20010623 You get a warning if you try to use DB_File as LIST_CACHE,
344 because it won't store lists. But if you use it as the underlying
345 cache with an expiration manager in the middle, no warning---the
346 expiration manager doesn't know it's managing a list cache, and
347 memoize doesn't know that DB_File is underlying. Is this fixable?
348 Probably not, but think about it.
351 There was probably some other stuff that I forgot.