context.txt

开放源码的编译器open watcom 1.6.0版的源代码

THE WINHELP |CONTEXT INTERNAL FILE
==================================

The |CONTEXT file is used by WinHelp as a lookup table to obtain the offset
of a topic based on a hash value.  See b_tree.txt for a general discussion
of b-trees.  The |CONTEXT file is used to resolve hotlinks.

Hash Values
-----------

Hotlinks within the help text are *not* stored as the offset of the topic
to which they point.  Rather, hotlinks are stored as hash values which
are generated from the context string for the topic...usually.  I'm not
sure about hot spots embedded in bitmaps (.SHG files), because those files
store both the hash value and the context string.  Whatever the case,
it's clear that both 'hc31' and 'shed' (and possibly WinHelp) share a
common notion of what the hashing function is.

Fortunately, that hashing function isn't too difficult.  Characters are
assigned a numeric value as follows:

Character	1  ... 9   0   .   _   a   b ...  z
---------------------------------------------------
Value		1  ... 9  10  12  13  17  18 ... 42

I don't yet know which characters, if any, get the values 11, 14, 15 or 16,
or 0.

Then the string is converted to a DWORD via a simple algorithm:
	1)  Set the hash value to zero, current character to first
	    character of the context string.
	2)  Multiply hash value by 43 (0x2B).
	3)  Add value of current character to hash value.
	4)  Throw away all but the low 32 bits.
	5)  If characters remain in the string, increment current
	    character and go to step 2.
	    
If two context strings will receive the same hash value (unlikely but
possible) a warning is issued by the help compiler and only the first
string in recorded in the |CONTEXT table.

The |CONTEXT B-tree
-------------------

The index pages of the |CONTEXT b-tree use the 4-byte hash values as key
values.

The leaf pages of the tree store 8-byte records.  The first four bytes
are a hash value; the last four bytes are the offset of the corresponding
topic within the |TOPIC file.  These offsets are in the form of "Character"
offsets; see topic.txt.