#read_entry in ActiveSupport::Cache::FileStore should log details of the exception when an exception is thrown.

In order that temp filenames generated from it will fit in 255 chars. See https://github.com/rails/rails/issues/4907

This reverts commit abe915f2.

This broke all existing keys and it's wrong anyway. The array is just there as a convenience option for building the string. It's intentional that [ "stuff"] and "stuff" generates the same key.

minor

In short:
expand_cache_key(element) should not equal expand_cache_key([element])

This way a fragment cache key for an index page with only a single
element in the collection is different than a fragment cache for a
typical show page for that element.

Add ActiveSupport::Cache::NullStore to expose caching interface without actually caching for development and test environments.

`nil` and `false` both expand to `""` (empty string), while `true` expands to
`"true"`; `false` should expand to `"false"`

`cache_key` method is never called when the argument is a 1-element array
with something that responds to `cache_key`

tests predicates according to the boolean interpretation of their return value, not expecting specifically true or false

Respecting Genuine assert_equal false Edition

Revert "tests predicates according to the boolean interpretation of their return value, not expecting specifically true or false"

Reason: there were some genuine tests for false when
reading values, going to revise those ones.

This reverts commit a539a5e3.

tests predicates according to the boolean interpretation of their return value, not expecting specifically true or false

Tests were failing on Travis-CI.

This reverts commit 79d01a8f, reversing
changes made to b838570b.

Added fix so that file store does not raise an exception when cache dir does not exist yet. This can happen if a delete_matched is called before anything is saved in the cache.

Updated existing test that fails with this fix (8d63678d).  Before the fix the test was giving a false positive, because file_store.key_file_path would return an empty filename (i.e. test/tmp_cache/4D0/F4D rather than test/tmp_cache/4D0/F4D/xxxx…).

Even though the fix referenced above divides the filename into directories to prevent it from being too long, it seems that 1000 characters will always raise an error, so reducing the key size to 900).

Fixed issue in file store where it could create a filename that was too long for the file system.  (https://github.com/rails/rails/issues/3072)

Port from 3-1-stable

do not test explicit equality of predicate methods, they should be allowed to return truthy or falsey objects

Fix FileStore cache incorrectly regenerating its key from a pathname when a regexp is used in expire_fragment

[#5850 state:committed]
Signed-off-by: NSantiago Pastorino <santiago@wyeworks.com>

code gardening: we have assert_(nil|blank|present), more concise, with better default failure messages - let's use them