flex-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Left Right <olegsivo...@gmail.com>
Subject Re: SwfEncoder.java#minBits - very strange code
Date Fri, 22 Aug 2014 18:23:51 GMT
More to the point though: I don't know whose version of 16.16 fixed
point is correct. Adobe specification is very vague on this subject. I
managed to recreate the behaviour of Adobe's (and presently, Apache
Flex) compilers, but I'm not sure of implications. It seems like at
very large number (something that almost never happens in real-life
SWFs) Adobe's implementation might do something unexpected, like,
maybe change sign, or overflow and start from the small numbers
again... The difference is in that Haxe format.swf library writes
16.16 fp with the sign bit one bit further on the left (i.e. when Haxe
writes -2, it'll write #b101, but the code above expects it to be
#b11.

Best,

Oleg

On Fri, Aug 22, 2014 at 9:16 PM, Left Right <olegsivokon@gmail.com> wrote:
> This code is used in swfdump (this is how I found it), but I didn't
> know whether it's used by mxmlc.
>
> One more thing though, this entire function is exactly equivalent to
> http://docs.oracle.com/javase/7/docs/api/java/lang/Integer.html#numberOfLeadingZeros%28int%29
> (except one needs to do 32 - Integer.numberOfLeadingZeros(i),
> obviously).
>
> Best,
>
> Oleg
>
> On Tue, Aug 19, 2014 at 12:06 AM, Alex Harui <aharui@adobe.com> wrote:
>> Have you proven that MXMLC actually uses SwfEncoder.java to create a SWF?
>>
>> Have you looked at the code in Falcon that does this?  That code might
>> work better.
>>
>> -Alex
>>
>> On 8/18/14 3:07 AM, "Left Right" <olegsivokon@gmail.com> wrote:
>>
>>>http://pastebin.com/KjvksPDX
>>>
>>>I think, I'm getting closer: when Flex compiler code encodes, and when
>>>Flash player decodes /negative/ 16.16 fixed point numbers, it is
>>>mistaken by one bit, so the whole part becomes doubled.
>>>
>>>On Sun, Aug 17, 2014 at 5:26 PM, Left Right <olegsivokon@gmail.com> wrote:
>>>> Sorry, the table looks crooked, here's a better view:
>>>> http://pastebin.com/rhw9sFtQ
>>>>
>>>> Best,
>>>>
>>>> Oleg
>>>>
>>>> On Sun, Aug 17, 2014 at 5:23 PM, Left Right <olegsivokon@gmail.com>
>>>>wrote:
>>>>>
>>>>>https://git-wip-us.apache.org/repos/asf/flex-sdk/repo?p=flex-sdk.git;a=b
>>>>>lob;f=modules/swfutils/src/java/flash/swf/SwfEncoder.java;h=03a100dda929
>>>>>89d537b00b96033d614c73c47801;hb=HEAD#l320
>>>>>
>>>>> This is the code I'm talking about.
>>>>>
>>>>> What is strange about it: it doesn't do what the comment above it
>>>>> says. For example, it says that you need 2 bits to write unsigned
>>>>> integer 1 - which is obviously false: you need only one bit. In fact,
>>>>> it overestimates the number of bits needed to be written by one,
>>>>> whenever all bits are set.
>>>>>
>>>>> (You can also point fingers at the... strange technique the author
>>>>> used for assertions, but this is less important at the moment).
>>>>>
>>>>> Why am I confused about this code: even though it's wrong, the results
>>>>> I see in the player agree with the results it produce. I.e. Flash
>>>>> player, when it decodes the Matrix record, uses some procedure, which
>>>>> expects this extra bit (I'm still struggling to understand how does it
>>>>> do it, but it sure works outside the spec).
>>>>>
>>>>> I'm trying to implement a SWF linker, something that assembles SWF
>>>>> files from a different format. I used HXSWFML library for that (it's
>>>>> written in Haxe), which, in turn, uses haxe.format.swf library. The
>>>>> problem is that when I write the Matrix record using these libraries,
>>>>> I do it to the spec, but the player reads the record in some wrong
>>>>> way, which I can't understand. Somehow Flex compiler manages to
>>>>> produce results coherent with the player... Below is my dissection of
>>>>> a generated PlaceOjbect3 tag, containing a Matrix record, and its
>>>>> interpretation using haxe.format.swf and swfdump Flex utility:
>>>>>
>>>>>    | id          |  dummy | length                                 |
>>>>> options                          |
>>>>>
>>>>>|-------------+--------+----------------------------------------+-------
>>>>>---------------------------|
>>>>>    | bf          |     11 | 17 00 00 00                            |
>>>>> 26 00                            |
>>>>>    | 70          |     63 | 23                                     |
>>>>> hasname, hasmatrix, hascharacter |
>>>>>    | 00010001 10 | 111111 | 00010111 000000000 000000000 000000000 |
>>>>> 00100110 00000000                |
>>>>>
>>>>>    | depth             | characterid       | matrix
>>>>>     | name                 |
>>>>>
>>>>>|-------------------+-------------------+-------------------------------
>>>>>+----------------------|
>>>>>    | 01 00             | 02 00             | a1 c5 c6 88 17 f7 3e c4
>>>>> 8b 98 | 73 6c 69 64 65 30 00 |
>>>>>    | 1                 | 2                 |
>>>>>     | "slide0"             |
>>>>>    | 00000001 00000000 | 00000010 00000000 | 10100001 11000101
>>>>> 11000110    |                      |
>>>>>
>>>>>    | hasscale | nscalebits | scalex    | scaley    | hasrotate |
>>>>> nrotatebits | rotateskew0 |
>>>>>
>>>>>|----------+------------+-----------+-----------+-----------+-----------
>>>>>--+-------------|
>>>>>    |          |         a1 | c5        | c6        |           |
>>>>>    88 | 17          |
>>>>>    | true     |          8 |           |           | true      |
>>>>>     8 |             |
>>>>>    | 1        |      01000 | 01 110001 | 01 110001 | 1         |
>>>>> 0 1000 | 1000 0001   |
>>>>>    |          |            |           |           |           |
>>>>>       |             |
>>>>>
>>>>>    | rotateskew1 | ntranslatebits | translatex      | translatey
>>>>> | padding |
>>>>>
>>>>>|-------------+----------------+-----------------+------------------+---
>>>>>------|
>>>>>    | f7          |             3e | c4 8b 98        |
>>>>> |         |
>>>>>    |             |             14 | 8034 (401.7)    | 4467 (223.35)
>>>>> |         |
>>>>>    | 0111 1111   |         0111 0 | 0111110 1100010 | 0 10001011 10011
>>>>> |     000 |
>>>>>
>>>>> The calculations I made by hand agree with haxe.format.swf, and with
>>>>> my understanding of the SWF file specification. Exactly, they give me
>>>>> this:
>>>>>
>>>>> Matrix:
>>>>> HasScale = 1
>>>>> ScaleX = 0.44140625
>>>>> ScaleY = 0.44140625
>>>>> HasRotate = 1
>>>>> RotateSkew0 = 0.50390625
>>>>> RotateSkew1 = 0.49609375
>>>>> TranslateX = 8034 twips = 401.7 px
>>>>> TranslateY = 4467 twips = 223.35 px
>>>>>
>>>>> Swfdump gives me:
>>>>>
>>>>>   <PlaceObject2 depth='1' matrix='s1.8934174,1.8934174
>>>>> r-4.135132,2.1351318 t8034,4467'/>
>>>>>
>>>>> If I trace the matrix from inside the player, it gives me:
>>>>>
>>>>> (a=1.8934173583984375, b=-4.1351318359375, c=2.1351318359375,
>>>>> d=1.8934173583984375, tx=401.7, ty=223.35)
>>>>>
>>>>> A result I cannot explain: the matrix was symmetrical to begin with,
>>>>> how on Earth does it arrive at different scaleX and scaleY values?
>>>>>
>>>>> Sorry for the long post. It looks to me, like this is a bug in SWF
>>>>> format implementation in Adobe Flash player, which eventually made its
>>>>> way into Flex compiler too. I would be interested to understand in
>>>>> what way exactly does it make this mistake - no matter how I tried, I
>>>>> can't generate these numbers :/
>>>>>
>>>>> Best,
>>>>>
>>>>> Oleg
>>

Mime
View raw message