Improving comment.

This commit is contained in:
Werner Lemberg 2005-03-11 09:14:21 +00:00
parent 465a53243f
commit 1dbcbabf6d
2 changed files with 54 additions and 50 deletions

View File

@ -1,6 +1,6 @@
2005-03-10 David Turner <david@freetype.org>
* src/tools/glnames.py: adding comment explaining the compression
* src/tools/glnames.py: Add comment to explain the compression
being used for the Adobe Glyph List.
2005-03-10 Werner Lemberg <wl@gnu.org>

View File

@ -4758,17 +4758,17 @@ class StringTable:
write( line + "\n };\n\n\n" )
#
# here's an explanation about the way we now store the Adobe Glyph List.
# First of all, we store the list as a tree. Consider for example that
# you want to store the following name mapping:
# We now store the Adobe Glyph List in compressed form. The list is put
# into a data structure called `trie' (because it has a tree-like
# appearance). Consider, for example, that you want to store the
# following name mapping:
#
# A => 1
# Aacute => 6
# Abalon => 2
# Abstract => 4
#
# it's possible to store them in a tree, as in:
# It is possible to store the entries as follows.
#
# A => 1
# |
@ -4776,58 +4776,62 @@ class StringTable:
# |
# +-b
# |
# +-alone => 2
# +-alon => 2
# |
# +-stract => 4
#
# we see that each node in the tree has:
# We see that each node in the trie has:
#
# - one or more 'letters'
# - one or more `letters'
# - an optional value
# - zero or more child nodes
#
# you can build such a tree with:
# The first step is to call
#
# root = StringNode( "", 0 )
# for word in map.values():
# root.add( word, map[word] )
#
# this will create a large tree where each node has only one letter
# then call:
# which creates a large trie where each node has only one children.
#
# Executing
#
# root = root.optimize()
#
# which will optimize the tree by mergin the letters of successive
# nodes whenever possible
# optimizes the trie by merging the letters of successive nodes whenever
# possible.
#
# now, each node of the tree is stored as follows in the table:
# Each node of the trie is stored as follows.
#
# - first the node's letters, according to the following scheme:
# - First the node's letter, according to the following scheme. We
# use the fact that in the AGL no name contains character codes > 127.
#
# name bitsize description
# ----------------------------------------------------------------
# notlast 1 Set to 1 if this is not the last letter
# in the word.
# ascii 7 The letter's ASCII value.
#
# - The letter is followed by a children count and the value of the
# current key (if any). Again we can do some optimization because all
# AGL entries are from the BMP; this means that 16 bits are sufficient
# to store its Unicode values. Additionally, no node has more than
# 127 children.
#
# name bitsize description
# -----------------------------------------
# notlast 1 set to 1 if this is not the last letter
# in the word
# ascii 7 the letter's ASCII value
# hasvalue 1 Set to 1 if a 16-bit Unicode value follows.
# num_children 7 Number of childrens. Can be 0 only if
# `hasvalue' is set to 1.
# value 16 Optional Unicode value.
#
# - then, the children count and optional value:
# - A node is finished by a list of 16bit absolute offsets to the
# children, which must be sorted in increasing order of their first
# letter.
#
# name bitsize description
# -----------------------------------------
# hasvalue 1 set to 1 if a 16-bit Unicode value follows
# num_children 7 number of childrens. can be 0 only if
# 'hasvalue' is set to 1
# if (hasvalue)
# value 16 optional Unicode value
# For simplicity, all 16bit quantities are stored in big-endian order.
#
# - followed by the list of 16-bit absolute offsets to the children.
# Children must be sorted in increasing order of their first letter.
#
# All 16-bit quantities are stored in big-endian. If you don't know why,
# you've never debugged this kind of code ;-)
#
# Finally, the root node has first letter = 0, and no value.
# The root node has first letter = 0, and no value.
#
class StringNode:
def __init__( self, letter, value ):