# Version 0.05 alpha $Revision: 1.6 $ $Date: 2001/06/24 17:11:26 $
=head1 TO DO
=over 4
=item *
LIST_CACHE doesn't work with ties to most DBM implementations, because
Memouze tries to save a listref, and DB_File etc. can only store
strings. This should at least be documented. Maybe Memoize could
detect the problem at TIE time and throw a fatal error.
20010623 This was added sometime prior to 20001025.
Try out MLDBM here and document it if it works.
=item *
We should extend the benchmarking module to allow
timethis(main, { MEMOIZED => [ suba, subb ] })
What would this do? It would time C<main> three times, once with
C<suba> and C<subb> unmemoized, twice with them memoized.
Why would you want to do this? By the third set of runs, the memo
tables would be fully populated, so all calls by C<main> to C<suba>
and C<subb> would return immediately. You would be able to see how
much of C<main>'s running time was due to time spent computing in
C<suba> and C<subb>. If that was just a little time, you would know
that optimizing or improving C<suba> and C<subb> would not have a
large effect on the performance of C<main>. But if there was a big
difference, you would know that C<suba> or C<subb> was a good
candidate for optimization if you needed to make C<main> go faster.
Done.
=item *
Perhaps C<memoize> should return a reference to the original function
as well as one to the memoized version? But the programmer could
always construct such a reference themselves, so perhaps it's not
necessary. We save such a reference anyway, so a new package method
could return it on demand even if it wasn't provided by C<memoize>.
We could even bless the new function reference so that it could have
accessor methods for getting to the original function, the options,
the memo table, etc.
Naah.
=item *
The TODISK feature is not ready yet. It will have to be rather
complicated, providing options for which disk method to use (GDBM?
DB_File? Flat file? Storable? User-supplied?) and which stringizing
method to use (FreezeThaw? Marshal? User-supplied?)
Done!
=item *
Maybe an option for automatic expiration of cache values? (`After one
day,' `After five uses,' etc.) Also possibly an option to limit the
number of active entries with automatic LRU expiration.
You have a long note to Mike Cariaso that outlines a good approach
that you sent on 9 April 1999.
What's the timeout stuff going to look like?
EXPIRE_TIME => time_in_sec
EXPIRE_USES => num_uses
MAXENTRIES => n
perhaps? Is EXPIRE_USES actually useful?
19990916: Memoize::Expire does EXPIRE_TIME and EXPIRE_USES.
MAXENTRIES can come later as a separate module.
=item *
Put in a better example than C<fibo>. Show an example of a
nonrecursive function that simply takes a long time to run.
C<getpwuid> for example? But this exposes the bug that you can't say
C<memoize('getpwuid')>, so perhaps it's not a very good example.
Well, I did add the ColorToRGB example, but it's still not so good.
These examples need a lot of work. C<factorial> might be a better
example than C<fibo>.
=item *
Add more regression tests for normalizers.
=item *
Maybe resolve normalizer function to code-ref at memoize time instead
of at function call time for efficiency? I think there was some
reason not to do this, but I can't remember what it was.
=item *
Add more array value tests to the test suite.
Does it need more now?
=item *
Fix that `Subroutine u redefined ... line 484' message.
Fixed, I think.
=item *
Get rid of any remaining *{$ref}{CODE} or similar magic hashes.
=item *
There should be an option to dump out the memoized values or to
otherwise traverse them.
What for?
Maybe the tied hash interface taskes care of this anyway?
=item *
Include an example that caches DNS lookups.
=item *
Make tie for Storable (Memoize::Storable)
A prototype of Memoize::Storable is finished. Test it and add to the
test suite.
Done.
=item *
Make tie for DBI (Memoize::DBI)
=item *
I think there's a bug. See `###BUG'.
=item *
Storable probably can't be done, because it doesn't allow updating.
Maybe a different interface that supports readonly caches fronted by a
writable in-memory cache? A generic tied hash maybe?
FETCH {
if (it's in the memory hash) {
return it
} elsif (it's in the readonly disk hash) {
return it
} else {
not-there
}
}
STORE {
put it into the in-memory hash
}
Maybe `save' and `restore' methods?
It isn't working right because the destructor doesn't get called at
the right time.
This is fixed. `use strict vars' would have caught it immediately. Duh.
=item *
Don't forget about generic interface to Storable-like packages
20010627 It would appear that you put this into 0.51.
=item *
Maybe add in TODISK after all, with TODISK => 'filename' equivalent to
SCALAR_CACHE => [TIE, Memoize::SDBM_File, $filename, O_RDWR|O_CREAT, 0666],
LIST_CACHE => MERGE
=item *
Maybe the default for LIST_CACHE should be MERGE anyway.
=item *
There's some terrible bug probably related to use under threaded perl,
possibly connected with line 56:
my $wrapper = eval "sub { unshift \@_, qq{$cref}; goto &_memoizer; }";
I think becayse C<@_> is lexically scoped in threadperl, the effect of
C<unshift> never makes it into C<_memoizer>. That's probably a bug in
Perl, but maybe I should work around it. Can anyone provide more
information here, or lend me a machine with threaded Perl where I can
test this theory? Line 59, currently commented out, may fix the
problem.
20010623 Working around this in 0.65, but it still blows.
=item *
Maybe if the original function has a prototype, the module can use
that to select the most appropriate default normalizer. For example,
if the prototype was C<($)>, there's no reason to use `join'. If it's
C<(\@)> then it can use C<join $;,@$_[0];> instead of C<join $;,@_;>.
=item *
Ariel Scolnikov suggests using the change counting problem as an
example. (How many ways to make change of a dollar?)
=item *
Jonathan Roy found a use for `unmemoize'. If you're using the
Storable glue, and your program gets SIGINT, you find that the cache
data is not in the cache, because Perl normally writes it all out at
once from a DESTROY method, and signals skip DESTROY processing. So
you could add
$sig{INT} = sub { unmemoize ... };
=item *
This means it would be useful to have a method to return references to
all the currently-memoized functions so that you could say
$sig{INT} = sub { for $f (Memoize->all_memoized) {
unmemoize $f;
}
}
=item *
19990917 There should be a call you can make to get back the cache
itself. If there were, then you could delete stuff from it to
manually expire data items.
=item *
19990925 Randal says that the docs for Memoize;:Expire should make it
clear that the expired entries are never flushed all at once. He
asked if you would need to do that manually. I said:
Right, if that's what you want. If you have EXISTS return false,
it'll throw away the old cached item and replace it in the cache
with a new item. But if you want the cache to actually get smaller,
you have to do that yourself.
I was planning to build an Expire module that implemented an LRU
queue and kept the cache at a constant fixed size, but I didn't get
to it yet. It's not clear to me that the automatic exptynig-out
behavior is very useful anyway. The whole point of a cache is to
trade space for time, so why bother going through the cache to throw
away old items before you need to?
Randal then pointed out that it could discard expired items at DESTRoY
or TIEHASH time, which seemed like a good idea, because if the cache
is on disk you might like to keep it as small as possible.
=item *
19991219 Philip Gwyn suggests this technique: You have a load_file
function that memoizes the file contexts. But then if the file
changes you get the old contents. So add a normalizer that does
return join $;, (stat($_[0])[9]), $_[0];
Now when the modification date changes, the true key returned by the
normalizer is different, so you get a cache miss and it loads the new
contents. Disadvantage: The old contents are still in the cache. I
think it makes more sense to have a special expiration manager for
this. Make one up and bundle it.
19991220 I have one written: Memoize::ExpireFile. But how can you
make this work when the function might have several arguments, of
which some are filenames and some aren't?
=item *
19991219 There should be an inheritable TIEHASH method that does the
argument processing properly.
19991220 Philip Gwyn contributed a patch for this.
20001231 You should really put this in. Jonathan Roy uncovered a
problem that it will be needed to solve. Here's the problem: He has:
memoize "get_items",
LIST_CACHE => ["TIE", "Memoize::Expire",
LIFETIME => 86400,
TIE => ["DB_File", "debug.db", O_CREAT|O_RDWR, 0666]
];
This won't work, because memoize is trying to store listrefs in a
DB_File. He owuld have gotten a fatal error if he had done this:
memoize "get_items",
LIST_CACHE => ["TIE", "DB_File", "debug.db", O_CREAT|O_RDWR, 0666]'
But in this case, he tied the cache to Memoize::Expire, which is *not*
scalar-only, and the check for scalar-only ties is missing from
Memoize::Expire. The inheritable method can take care of this.
20010623 I decided not to put it in. Instead, we avoid the problem by
getting rid of TIE. The HASH option does the same thing, and HASH is
so simple to support that a module is superfluous.
=item *
20001130 Custom cache manager that checks to make sure the function
return values actually match the memoized values.
=item *
20001231 Expiration manager that watches cache performance and
accumulates statistics. Variation: Have it automatically unmemoize
the function if performance is bad.
=item *
20010517 Option to have normalizer *modify* @_ for use by memoized
function. This would save code and time in cases like the one in the
manual under 'NORMALIZER', where both f() and normalize_f() do the
same analysis and make the same adjustments to the hash. If the
normalizer could make the adjustments and save the changes in @_, you
wouldn't have to do it twice.
=item *
20010623 Add CLEAR methods to tied hash modules.
=item *
20010623 You get a warning if you try to use DB_File as LIST_CACHE,
because it won't store lists. But if you use it as the underlying
cache with an expiration manager in the middle, no warning---the
expiration manager doesn't know it's managing a list cache, and
memoize doesn't know that DB_File is underlying. Is this fixable?
Probably not, but think about it.
=item *
There was probably some other stuff that I forgot.
=back
|
