散列 DWG 文件 - 空模型不会收到相同的 hash 代码

Question

I want to use a hash function (SHA1) to determine whether two DWG models from AutoCAD 2018 are identical or not. To test if this works, I created two empty dwg files and used the SHA1 function in c# (System.Security.Cryptography.sha1) to calculate the hash code.

I was expecting resulting codes to be identical, since I only started two new empty models and then saved them directly. The only data which should differ is the name and date & time of creation (all meta data, which shouldn't play a role for the algorithm). Yet, the result was two different hash codes.

So I looked into the byte array of the two files. The only thing I could get out of it (a lot of lines which I could not interpret in any way), was that some of the meta data as date and time of creation is written in the byte array. Therefore the two files are actually not identical, eventhough the models are. So there will always be a different hash code for all hashed DWG models.

Does anyone know of this problem or a work-around for my issue?

Code used to calculate hash:

static void GetFileHash(string filePath)
        {
            var sha1 = new SHA1CryptoServiceProvider();
            byte[] bArray = File.ReadAllBytes(filePath);
            byte[] hashByte = sha1.ComputeHash(bArray);
            string hashCode = BitConverter.ToString(hashByte);

            //save file path, hash and byte array to text file
            string[] lines = { filePath, hashCode, System.Text.Encoding.Default.GetString(bArray)};               System.IO.File.WriteAllLines(@".\HashTest.txt", lines);
        }

Snippet of the byte array in which the saving time and date is captured:

A pp I nfo D ata L ist �H��

M��ρ��P� 2 2. 0. 4 9. 0. 0 ���%�דI����(o��r A utodesk DW G. T hisfileisa T r usted DWG lastsavedbyan A utodeskappli c ationo r A utodeskli c ensedappli c atio n. �����Oh�� +'��J< p r op _ setfmt _ id = " { f 2 9 f 8 5 e 0 - 4 ff 9 - 1 0 6 8 - ab 9 1 - 0 8 0 0 2 b 2 7 b 3 d 9 } " > < p r opid = " 8 " > < st r ing > M ax < / st r ing > < / p r op > < p r opid = " 1 0 " > < datetime > 2 0 2 0 - 0 7 - 1 4 T 0 9: 5 1: 3 3 < / datetime > < / p r op > < p r opid = " 2 5 8 " > < st r ing > A uto C AD 2 0 1 8 < / st r ing > < / p r op > < p r opid = " 2 5 9 " > < st r ing > O. 4 9. 0. 0 < / st r ing > < / p r op > < p r opid = " 1 2 " > < datetime > 2 0 2 0 - 0 7 - 1 4 T 0 9: 5 1: 2 9 < / datetime > < / p r op > < / p r op _ set > ���Q�βD����;��D� " < P r odu c t I nfo r mationname = \ " A uto C AD \ " build _ ve r sion = \ " O. 4 9. 0. 0 ( x 6 4 ) \ " r egist r y _ ve r sion = \ " 2 2. 0 \ " install _ id _ st Z4B43B0AEE35624CD95B910189B3D C231Z ing = \ " A C AD - 1 0 0 1: 4 0 9 \ " r egist r y _ lo c ale ID = \ " 1 0 3 3 \ " > "

Answer 1

When a new empty model is created, it will add some stuff for you like a layer "0", the model space and the paper space. Which will be added to the cad database table record and assign some id's which will be different no matter how many new model your create.

For example. when you look into the data assigned to layer "0" of 2 new empty models it will look like that:

// Layer "0" from Model 1
"((-1,(2956860066048))(0,LAYER)(5,10)(102,{ACAD_XDICTIONARY)(360,(2956924193344))(102,})(330,(2956860065824))(100,AcDbSymbolTableRecord)(100,AcDbLayerTableRecord)
(2,0)(70,0)(62,7)(6,Continuous)(290,1)(370,-3)(390,(2956860066032))(347,(2956860067552))(348,(0)))"

// Layer "0" from Model 2
"((-1,(2991422834944))(0,LAYER)(5,10)(102,{ACAD_XDICTIONARY)(360,(2991422843456))(102,})(330,(2991422834720))(100,AcDbSymbolTableRecord)(100,AcDbLayerTableRecord)
(2,0)(70,0)(62,7)(6,Continuous)(290,1)(370,-3)(390,(2991422834928))(347,(2991422836448))(348,(0)))"

Those are quite different and will be in the.dwg binary content you are comparing the hash which is unknown what it stores inside it. In order for the hash to be the same you would need to copy paste the file.

A workaround could be to save the file into dxf format instead which are human readable so you can ignore some data to consider the files identical.

Answer 2

Nowhere does any sort of a hash enter the picture, so to speak. You don't care what's in the DWG file, you care that it looks the same!

In general this can't work, unless the file you're hashing is in the canonical form. That'd also mean that such a canonical form has to exist - not a given at all, I don't know offhand about whether there is a canonical form of DWG, or whether it's even an easy problem to create one. such that visually identical results would have same DWG form, In general, the answer to this is a solid no .

The summary of the problem is that files that appear visually identical when rendered at a particular resolution may have an extremely large number of possible permutations in their DWG representation - in practice you could consider it infinite.

So, you cannot hash the DWG file itself, not even if you had a DWG library that could read the primitives - you'd need to use a computational geometry library to attempt to process the data into some canonical form.

The best you can do with limited resources is to rasterize both DWGs, and then run a visual comparison between the bitmaps thus obtained, at a reasonable printout resolution (say 600-1200dpi). The comparison algorithm should be insensitive to common pitfalls such as different "hardness" of line edges, but it should be sensitive to changes that alter the visual meaning of the document - this may not be straightforward either, and you'd need to look for a library/package that can do such comparisons for you, specifically targeted for line art and not photographs (.,). The files can also have different absolute coordinates for the primitives (due to transformations yielding same output), and slightly different scaling.

Thus, in most likelihood, you'd need to first visually align the bitmaps (there are algorithms for this that can have eg translation, scaling and rotation as degrees of freedom), then re-rasterize the second image in the pair taking the alignment into account, and then run a visual comparison, and produce some sort of a "score". You could also highlight the differences and display them on screen.