Gotcha

Message: 
    Assert.Throws() Failure
    Expected: typeof(SixLabors.ImageSharp.InvalidImageContentException)
    Actual:   typeof(System.IndexOutOfRangeException): Index was outside the bounds of the array.
    ---- System.IndexOutOfRangeException : Index was outside the bounds of the array.

  Stack Trace: 
    HuffmanTable.ctor(ReadOnlySpan`1 codeLengths, ReadOnlySpan`1 values) line 58
    JpegDecoderCore.ProcessDefineHuffmanTablesMarker(BufferedReadStream stream, Int32 remaining) line 1098
    JpegDecoderCore.ParseStream(BufferedReadStream stream, HuffmanScanDecoder scanDecoder, CancellationToken cancellationToken) line 358
    ...

One of the invalid test images was going out of bounds during new HuffmanTable ctor. The only reason it wasn't spotted before is that exposed code in Configure() was called after checks for invalid image which led to these crashes.

Ah nice, well done finding that. So now I guess we have to find a way to sanitise cheaply and then try to move the table building to the constructor.

It's actually a bug, there's a possibility for malformed image go out of bounds without any exceptions in current master. I found a way to check that without a noticable slowdown. I'd also create an example image for a test suite.

Codecov Report

Merging #1926 (544c557) into master (ce34e84) will decrease coverage by 0%.
The diff coverage is 100%.

@@          Coverage Diff           @@
##           master   #1926   +/-   ##
======================================
- Coverage      88%     88%   -1%     
======================================
  Files         966     966           
  Lines       51350   51320   -30     
  Branches     6401    6397    -4     
======================================
- Hits        45218   45188   -30     
  Misses       5074    5074           
  Partials     1058    1058           

Flags with carried forward coverage won't be shown. Click here to find out more.

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ce34e84...544c557. Read the comment docs.

High hopes it would pass all the tests, it is ready to be reviewed:

• Rewritten part of the huffman table building routine, it now requires 256 bytes less memory workspace
• No stackalloc calls
• Rewritten DHT marker parsing code, now it fetches a big single buffer for binary parsing and table building routine
• Added sanity checks for huffman table input, now it should be impossible to break decoder with malformed DHT marker

There's no need for a new test image, image from #839 has malformed DHT marker which goes out-of-bounds during huffman table for unsafe pointers (which is unspottable without wrapping them with Span). I've left a note just in case but this PR should nullify this error.

this is my favourite
On top of a critical security breach fix and extra checks we are not actually slower:

MASTER
// Decoding
// Elapsed: 4643ms across 400 iterations
// Average: 11,6075ms

PR
// Decoding
// Elapsed: 4634ms across 400 iterations
// Average: 11,585ms