adding cache manager documentation (draft)

This commit is contained in:
David Turner 2000-11-23 03:01:07 +00:00
parent f4f55e2159
commit 36d67f89c9
1 changed files with 267 additions and 0 deletions

267
docs/cache/cache.txt vendored Normal file
View File

@ -0,0 +1,267 @@
The FreeType 2 cache sub-system explained
(c) 2000 David Turner
-----------------------------------------------
Introduction :
--------------
this document describes the caching sub-system that comes
with the FreeType library, version 2.0. Note that unlike
the rest of the library, this code is still in beta stage
and might still suffer slight changes in the future.
Its basic design shouldn't evolve though and is explained
in this paper.
I. Requirements and Design Goals:
---------------------------------
The FT2 cache sub-system was designed to implement caching
of glyph images. However, it is extremely flexible and can
be easily extended to cache other kind of data like metrics,
character maps, coverage tables, etc..
II. Base Concepts:
------------------
1. The cache manager object:
at the heart of the caching sub-system is a single object
called the "cache manager". It is used to deal with FT_Face
and FT_Size objects, as well as to manager a LRU list of
abstract "cache nodes".
a. caching FT_Face and FT_Size objects:
each FT_Face object created by FreeType 2 can take from
a few hundred bytes to several tens of kilobytes, depending
on the original font's file format as well as its content.
there is no easy way to compute the size of a given FT_Face
object, so it's always a good idea to assume that it is
large and to want to limit the number of live face objects
as much as possible.
similarly, each FT_Face can have one or more FT_Size childs,
whose byte size depends heavily on the font format.
the first purpose of the cache manager is to provide a
small cache for FT_Face and FT_Size objects. Basically,
an application can use it as follows:
- each font face is described to the cache manager
through a typeless pointer, called a FTC_FaceID.
the cache manager itself doesn't interpret or use
the value of FTC_FaceIDs directly. Rather, it passes
them to a user-provided function called a
"face requester". see the defintion of the
FTC_Face_Requester type in <freetype/ftcache.h>
for details..
the face requester is in charge of translating a given
face into into a real new FT_Face object that is
returned to the cache manager. The latter will keep
the face object alive as long as it needs to.
the face requester is unique and must be passed
to the function named FTC_Manager_New used to
create/initialise a new cache manager.
- to lookup a given FT_Face, call the function
FTC_Manager_Lookup_Face as in the following code:
FTC_Manager_Lookup_Face( manager,
face_id,
&face );
if the corresponding FT_Face object is kept in
the cache manager's list, it will be returned
directly. Otherwise, this function will call
the user-provided face requester to create
a new FT_Face object, add it to the manager's
list to finally return it.
FT_Face objects are always destroyed by the cache
manager. An application that uses the cache
sub-system should never call FT_Done_Face !!
- to lookup a given FT_Size and FT_Face, call the
function FTC_Manager_Lookup_Size, as in:
FTC_Manager_Lookup_Size( manager,
ftc_font,
&face,
&size );
where "ftc_font" is a pointer to a FTC_Font descriptor
(a structure containing a FTC_FaceIDs and character
dimensions corresponding to the desired FT_Size).
note that the function returns both a FT_Face and
a FT_Size object. You don't need to call
FTC_Manager_Lookup_Face before it !!
also note that returned FT_Size objects are always
destroyed by the cache manager. A client application
that uses it should never call FT_Done_Size !!
the big advantage of using FTC_FaceIDs is that is
makes the caching sub-system completely independent
of the way font files are installed / listed / managed
in your application. In most implementations, a FTC_FaceID
is really a pointer to an application-specific structure
that describe the source font file + face index.
b - manage a MRU list of abstract "cache nodes":
the second role of the cache manager is to hold and manager
a list of abstract "cache nodes". The list is always sorted
in most-recently-used order. The manager always ensure that
the total size of nodes in memory doesn't over-reach a
certain threshold, by eliminating "old" nodes when
necessary.
the cache manager doesn't know much about the cache nodes:
- it knows how to move them in its list
- it knows how to destroy them when they're too old
- it knows how to "size" them (i.e. compute their byte
size in memory)
2. Cache objects:
the cache manager doesn't create new cache nodes however, this
is the charge of what are called "cache objects".
Basically, each cache object is in charge of managing cache
nodes of a certain type. Its role is to:
- provide a simple description of its cache nodes to the
manager (i.e. through a FTC_CacheNode_Class structure)
- provide a high-level API that can be called by client
applications to lookup cache nodes of the corresponding
type.
this function usually creates new nodes when they're not
available yet.
- also, and even though this is completely transparent to
the applications and the cache manager, each cache object
manages "node sets", where each set contains cache nodes
usually correspond to the same font face + font size.
For example, the cache sub-system currently comes with two
distinct cache classes:
- a FTC_Image_Cache, which is used to cache FT_Glyph images
(with one FT_Glyph per cache node).
- a FTC_SBit_Cache, which is used to cache small glyph bitmaps
("sbit" means "embedded bitmaps" in digital typography).
the small bitmaps glyph is useful because storing one glyph
image per cache node isn't memory efficient when the data
associated to each node is very small. Indeed, each cache
node has a minimal size of 20 bytes, which is huge when
your data is an 8x8 monochrome bitmap :-)
Hence, a FTC_SBit_Cache is capable of storing several
contiguous sbits in a single cache node, resulting in much
higher cached glyphs / total cache size.
an application can lookup a FT_Glyph image with a FTC_Image_Cache
by calling:
error = FTC_Image_Cache_Lookup( image_cache,
ftc_font,
glyph_index,
&ft_glyph );
or a FTC_SBit (small bitmap descriptor) by calling:
error = FTC_SBit_Cache_Lookup( sbit_cache,
ftc_font,
glyph_index,
&ftc_sbit );
III. Extending the cache sub-system:
It is possible to extend the current cache sub-system by
providing your own cache class and register it in the cache
manager. That might be useful to cache other kind of data
in the sub-system, like glyph metrics (without images),
To do it, you'll need to read the cache sub-system public
header files rather heavily :-) Fortunately, they're pretty
well commented and should guide you to your goal.
Note that the cache sub-system already provides two "abstract
cache" classes that can be re-used by your own implementation:
1. The abstract "FTC_GlyphCache" class:
this code is used to implement an abstract "glyph cache",
i.e. one that simply maps one glyph data per cache node.
it is sub-classed by FTC_Image_Cache, whose implementation
only consists in simple code to store a FT_Glyph in each
cache node.
you could sub-class it in your application to store glyph
images in a different format, for example.
see the files <freetype/cache/ftcglyph.h> and
"src/cache/ftcglyph.h" for details.
2. The abstract "FTC_ChunkCache" class:
this code is used to implement an abstract "glyph chunk cache".
it's very similar to a FTC_GlyphCache, except that it is capable
of storing several glyph-specific elements per cache node.
it is sub-classed by FTC_SBit_Cache, whose implementation
only consists in code to store a FTC_SBitRec record in each
node element.
you could sub-class it in your application to store small
glyph data, like metrics, glyph names, wathever.
see the files <freetype/cache/ftcchunk.h> and
"src/cache/ftcchunk.h" for details..
Note that the two abstract caches are rather complex because
they use "glyph sets". Each glyph set corresponds to a single
font face + font size combination. both caches are also
glyph-specific, though it is perfectly possible to use
broader selection criterion, here are a few examples:
- caching language coverage maps corresponding to
a given font face + language combination
- caching charmaps, layout tables, and other global
data..
- caching (font_face + font_size) specific "latin1"
ascender + descender
as you can see, a lot is possible with this design :-)