windowscodecs questions.
Max TenEyck Woodbury
max at mtew.isa-geek.net
Wed Oct 24 11:16:12 CDT 2012
Background: I am trying to improve the performance/reliability of
GuildWars2 under wine. In particular, PNG handling seems very
slow, so I am looking at windowscodecs and I seem to be
misunderstanding some things. I have read the PNG spec.
First: The CRC check. The spec says that a 'chunk' contains a four
byte size of data field, a four byte 'chunk identifier', the data
and a CRC32 of the data.
I found some code that reads a generic chunk. I see where the size
and type are read, where a correctly sized record is allocated and
where the data is read into the record.
There is a 'FIXME' that asks about checking the CRC, but no code to
actually read or check the CRC.
I added code to read and check the CRC to my working copy of the git
tree. That threw off the block synchronization, so I think there is
code someplace else that either checks or skips the CRC, but I have
not found it after a few hours searching for it. An indication on
where to look and other advice would be very helpful at this time.
Second: At least two of the methods GuildWars2 wants to use are stubbed.
To implement those methods, the frame section of the WIC object should
contain an array (or something with similar properties) of pointers to
chunks. From what I have read about WIC, other formats could use a
similar array. Some could even use an array of pointers to frames.
Before I go messing with that level of change, I think I should listen
to other peoples opinions of the subject.
Third: On a very broad level, the whole OLE implementation seems to be
very messed up. In particular, the usual practice for doing
inheritance in C does not seem to be being used. That practice is to
have the elements common to the base and extending object be placed at
the beginning of the implementing data structures. While the member
names need not be the same in all instances, the function, type and
order of the common elements must be the same for economical
implementation.
This has been done for the basic object; all start with a pointer to
the 'vtable'. What I think should follow is the critical section lock
structure and, for 'IUnknown', the reference count. What I see is that
these common elements are placed more or less at random in each
extended object.
Is there any reason, other than inertia, that at least these two
fields should not be moved to the beginning of implementing objects.
This can be done one object at a time and would allow the changed
objects to use a common implementation of 'AddRef' at least, and
another common routine or macro that handles the critical part of
'Release'.
More information about the wine-devel
mailing list