Re: memcache(3) 1.2.0 released...

On Wed, 2005-01-12 at 21:46 +0300, Antony Dovgal wrote:
> On Wed, 12 Jan 2005 10:35:54 -0800
> John McCaskey <johnm@xxxxxxxx> wrote:
> 
> > The one major complaint I have with the current php oo api's (and
> > possibly the Ruby api) are that they simply return FALSE (nil for
> > Ruby?) for an unfound key, thus breaking the ability to store
> > arbitrary binary data as I now cannot store a FALSE, or a binary
> > equivalent in memcache.
> 
> Why?
> Just don't store anything and you'll get the same result - FALSE.

That doesn't help though.  What if I have a database or other slow
storage where I'm getting boolean values out in a large quantity and
both TRUE and FALSE are valid values that I need to retrieve and use.

If I get a cache miss it indicates to me that I need to hit this storage
to find out whether the value is TRUE or FALSE.  However, if I get a
FALSE out then I don't need to hit the storage I should just use the
FALSE as my value.

If the get() command however returns a FALSE for either my stored FALSE
value or for a miss, then I will always have to assume its a miss and
hit the slow storage everytime I retrieve a FALSE.  Am I missing
something here?  I can't see it if so.

> 
> Or store FALSE and get empty string as get() value (it seems quite 
> logical to me that "" == FALSE just because they _are_ equal).
> 
> I don't see any problems here, but I appreciate any 
> help/suggestions/improvements.
> 
> > Perhaps, I'll always return as an array('key'=>'value', ...,
> > 'keyn'=>'valuen') and the way to determine a miss is simply that the
> > key does not exist at all, and then NULL or FALSE are valid data from
> > the cache.  
> 
> You've missed something =)
> 
> <?php
> $m = memcache_connect("localhost"); 
> $m->set("var", FALSE); 
> $m->get(Array("var", "nonexistent_key"));
> ?>
> ----
> bool(true)
> array(1) {
>   ["var"]=>
>   string(0) ""
> }

Thats great, but it's also exactly what I explained as my plan right?
This does definately improve the usuability of your PECL client for me
though.

> ----
> 
> But thanks, I just realized that I forgot to add this syntax to the docs.
> 
> > If I did this as a php5 module, maybe I'd just throw an exception...
> > but alas I need to use php4.
> 
> Throwing an exception when there is no such key seems a bit scary to me.
> I used to think that exceptions are throwed in case of some errors, but
> missing key doesn't look as error to me.
> 

I agree with that, it was just a quick thought.  But you are very right,
it would probably be a bit of an abuse of exceptions.

-- 
John A. McCaskey
Software Development Engineer
Klir Technologies, Inc.
johnm@xxxxxxxx
206.902.2027