Get "Invalid PNG data." exception during Identify phase of loading PNG image when image contains sBIT chunk

[flew2bits]

Prerequisites

• I have written a descriptive issue title
• I have verified that I am running the latest version of ImageSharp
• I have verified if the problem exist in both DEBUG and RELEASE mode
• I have searched open and closed issues to ensure it has not already been reported

ImageSharp version

v2.0.0

Other ImageSharp packages and versions

ImageSharp.Drawing v1.0.0beta14

Environment (Operating system, version and so on)

Windows 10, Rider

.NET Framework version

6.0.1

Description

When loading an image that contains an sBIT chunk, the Identify phase throws an exception stating that the data is invalid. The switch statement in the identify function (around line 250 of PngDecoderCore.cs) does not include the PngChunkType.SignificantBits as a case, nor does it have a default clause. Adding a default clause as follows fixes the issue:

default: this.SkipChunkDataAndCrc(chunk); break;

I may try to submit a PR if I can figure out how to do so.

Unfortunately, I don't have permission to upload the image that is causing this problem.

Steps to Reproduce

Image.Load(stream) from a stream that has a PNG with a sBIT chunk.

Images

No response

[brianpopow]

Unfortunately, I don't have permission to upload the image that is causing this problem.

You cannot attach the image here? Just drag and drop it in the text box should work.

[flew2bits]

It's not that I'm unable to attach the image -- it's a copyright image that I don't have permission to upload here.

[brianpopow]

It's not that I'm unable to attach the image -- it's a copyright image that I don't have permission to upload here.

Ok, I misunderstood that. Then we need to find another image with the same chunk for a unit test.

Thanks for opening a PR for this.

[flew2bits]

It looks like I'll need to figure out why the three resolution tests are failing.

[flew2bits]

It looks like there's some other reason my file isn't working right. Several of the test files also have sBIT chunks and they work fine. However, Windows and Chrome can both open my file just fine, and a file analyzer I used on it didn't find any problems, so I'll see if I can find an alternate solution.

[brianpopow]

It looks like there's some other reason my file isn't working right. Several of the test files also have sBIT chunks and they work fine. However, Windows and Chrome can both open my file just fine, and a file analyzer I used on it didn't find any problems, so I'll see if I can find an alternate solution.

Maybe it has todo with chunk ordering? It could be that a chunk is in an unexpected position of the stream.

[flew2bits]

I can't provide the actual image file, but below is the first 48 bytes of the file -- all that follows is a series of IDAT chunks and an IEND chunk. The identify method starts at byte 8, so it skips the magic. When it reads the IHDR chunk, the stream position is 33, which is the start of the sBIT chunk. However, when it's reading the sBIT chunk, the position should be 48, but it's 41, which is the start of the data for the chunk -- 08 08 08.

In the other files with sBIT chunks, it works correctly; e.g. the ratio-1x4.png file in the test images.

I'm going to continue investigating this to see if I can find a solution.

89 50 4E 47 0D 0A 1A 0A PNG Magic

00 00 00 0D 49 48 44 52 IHDR
00 00 01 AD 00 00 02 11
08 02 00 00 00 76 D4 75
F3

00 00 00 03 73 42 49 54 sBIT
08 08 08 DB E1 4F E0

[flew2bits]

The PR should fix the problem at this time.

[flew2bits]

I have been unable to create a new file that exhibits this same issue. I opened the errant file in PS and blurred it so I could share it and replaced the chunks before the IDAT chunks with the data from above, and it loads just fine.

I've also been unable to reproduce the problem in a minimal application using the errant PNG. I still believe there's a bug somewhere in ImageSharp, but I can't pinpoint it. I'll try to find some time this week to trace through things a little more closely, but I'm going to have to set this aside for now.

[brianpopow]

You could try out what libpng and see what it reports. It comes with pngtest util which reports errors.