Segment wildcard counts by $string[0]

Looks like the commit introduced an issues:
Notice: Undefined variable: table in MemCacheDrupal->wildcards() (line 248 of sites/all/modules/memcache/memcache.inc).
I guess the variable $table is deprecated in favour of $this->bin

The attached patch replaces all $table variables with $this->bin.

*grml* forgot to change status

Thanks for committing so fast.
Unfortunately I didn't catch all issues.

The variable $wildcard_invalidate seems to be replaced by $this->invalidate.

Attached fixes that and also one trailing space.

Next minor issue.
Pisco told me about this strict warning: Strict warning: Only variables should be passed by reference in MemCacheDrupal->wildcards() (line 248 of …/modules/memcache/memcache.inc).

It's caused because reset() expects a reference to a variable. Thus a function return like from array_keys() provokes the warning.

Found another construct that wasn't properly ported to D7. Unfortunately the code was written defensive enough to prevent warnings ;)

Ok, so either I'm not getting this or this patch makes no sense at all. The wildcard tests are currently failing for me and I've tried to fix them, but didn't get far except adding like a million debug statements ;)

The following code attempts to get the shortest prefix:

      // Wildcard flushes per table are keyed by a substring equal to the
      // shortest wildcard clear on the table so far. So if the shortest
      // wildcard was "links:foo:", and the cid we're checking for is
      // "links:bar:bar", then the key will be "links:bar:".
      $keys = array_keys($this->wildcard_flushes[$this->bin]);
      $wildcard_length = strlen(reset($keys));

- This only gets the shortest prefix because the code below is broken and completely clears the array whenever a shorter wildcard is found, see below.
- Second and even more important, this never actually verifies that the prefix matches the cid. If the cid is 'abc' and the prefix which happens to be the first is 'c', then we're using the prefix 'a' for the wildcard lookup. Whut?

The following lines of code empties the wildcard_flushes array for the given bin if the key length is shorter:

        if ($length < $key_length) {
          $this->wildcard_flushes[$this->bin] = array();
        }

- Because we never actually verify that the prefixes have something in common (e.g. 'a' and 'abc' as opposed to 'a' and 'cba'), this always resets the wildcards array when any shorter wildcard is found, e.g. when you have the wildcards 'abc' and 'bca' and then get a new wildcard 'c' you end up with just 'c'.

- Also, even if the wildcards are common, e.g. 'abc' and 'ab', we loose the information about the number of flushes. which means that with the following sequence of calls, the last cache_get() does *not* fail: cache_set('test_string:1', 1), cache_clear_all('test_string'), cache_set('test_string:1', 2), cache_clear_all('test_'), cache_get(''test_string:1'). This is because flushes of our cache is 1 but we also only have a single wildcard with a single flush registered. This is the reason the tests currently fail, the other issues aren't detected.

It is not trivial to fix and correctly implement this. I'm wondering if we should actually roll back this patch and start with a much more simple solution that uses a fixed prefix size of 1.

Ok, here is a patch that adds the full bin clear (Note that I also had to set $this->cache_flush directly, the whole code is screaming for some refactoring that moves things like that in separate methods).

I also added some tests to test this.

Back-port to 6.x committed.