Update behind the scenes doc to match current offline behavior.

docs/behind-the-scenes.txt

@@ -6,28 +6,35 @@ Django Compressor, and that you understand how to use it in your templates.
 The goal is to explain what the main template tag, {% compress %}, does
 behind the scenes, to help you debug performance problems for instance.
 
-First step: Offline cache
+Offline cache
 -------------------------
 
-The first thing {% compress %} tries to do is get the offline cache for its
-nodelist if offline cache is activated. It doesn't parse, doesn't check the
-modified times of the files, doesn't even know which files are concerned 
-actually, since it doesn't look inside the nodelist of the template block
-enclosed by the ``compress`` template tag. The cache should return the HTML
-containing the element code for the combined file or piece of code (which,
-if the cache key exists, is supposed to already exist on disk/custom storage).
+If offline cache is activated, rhe first thing {% compress %} tries to do is
+retrieve the compressed version for its nodelist from the offline manifest 
+cache. It doesn't parse, doesn't check the modified times of the files, doesn't
+even know which files are concerned actually, since it doesn't look inside the
+nodelist of the template block enclosed by the ``compress`` template tag.
+The offline cache manifest is just a json file, stored on disk inside the
+directory that holds the compressed files. The format of the manifest is simply
+a key <=> value dictionnary, with the hash of the nodelist being the key, 
+and the HTML containing the element code for the combined file or piece of code
+being the value. Generating the offline manifest, using the ``compress`` 
+management command, also generate the combined files referenced in the manifest. 
 
-Everything stops here if the cache entry exists.
+If offline cache is activated and the nodelist hash can not be found inside the
+manifest, {% compress %} will raise an ``OfflineGenerationError``.
+
+If offline cache is de-activated, the following happens:
 
-Second step: parsing and file list
+First step: parsing and file list
 ----------------------------------
 
 A compressor instance is created, which in turns instantiates the HTML parser.
 The parser is used to determine a file or code hunk list. Each file mtime is
 checked, first in cache and then on disk/storage, and this is used to
 determine an unique cache key.
 
-Third step: Checking the "main" cache
+Second step: Checking the "main" cache
 -------------------------------------
 
 Compressor checks if it can get some info about the combined file/hunks
@@ -37,7 +44,7 @@ element code, just like in the offline step before.
 
 Everything stops here if the cache entry exists.
 
-Fourth step: Generating the combined file if needed
+Third step: Generating the combined file if needed
 ---------------------------------------------------
 
 The file is generated if necessary. All precompilers are called and all