PBM decoder robustness improvements and BufferedReadStream observability #2551
Conversation
| Assert.Equal(default, info.Size); | ||
| Configuration.Default.ImageFormatsManager.TryFindFormatByFileExtension("pbm", out IImageFormat format); | ||
| Assert.Equal(format!, info.Metadata.DecodedImageFormat); | ||
| Assert.Throws<InvalidImageContentException>(() => Image.Identify(bytes)); |
There was a problem hiding this comment.
It doesn't really matter but I don't think decoding makes much sense when the EOF is right in the header.
There was a problem hiding this comment.
Yeah, I'm happy with that.
| stream.SkipWhitespaceAndComments(); | ||
| int height = stream.ReadDecimal(); | ||
| stream.SkipWhitespaceAndComments(); | ||
| if (!stream.SkipWhitespaceAndComments() || |
There was a problem hiding this comment.
Only thing I would suggest is to change the method names to use Try prefix to match convention.
There was a problem hiding this comment.
I've pushed those changes.
There was a problem hiding this comment.
Ha, didn't realize you'd explicitly removed them! I can revert at a later time if you feel strongly about it.
There was a problem hiding this comment.
I really prefer it to be the other way.
By convention, TryDoSomething(...) == false should indicate that the operation failed as a whole. This is not the case with TryReadDecimal(out value) == false when it happens at the very las digit in the file: the decoding of the digit was succesful (we should threat the result as valid!) just the file reached an EOF.
I like following such conventions to the letter, because ambiguity can lead to a misunderstandings and programming mistakes.
There was a problem hiding this comment.
I’m not precious about it (and happy to revert) but there’s something a little clunky about the method there. We’re really returning two things, the decimal and the stream state. I’ve seen something in the runtime I’m sure that handles it better (Maybe inside guid parsing code)
There was a problem hiding this comment.
I considered returning a (bool, int) tuple, but that would lead to much more complicated code on the callsites, since we couldn't do !stream.SkipWhitespaceAndComments() || .... Defining the names without the Try prefix + documenting the behavior was the best idea I was able to come up with.
| byte value = (byte)stream.ReadDecimal(); | ||
| stream.SkipWhitespaceAndComments(); | ||
| rowSpan[x] = new L8(value); | ||
| stream.ReadDecimal(out int value); |
There was a problem hiding this comment.
Shouldn't we test here?
There was a problem hiding this comment.
I guess it doesn't matter since we're simply assigning 0.
There was a problem hiding this comment.
No, when it's the last digit of the file the value is valid! For simplicity, I'm ignoring the retval here and letting the SkipWhitespaceAndComments call below to detect EOF.
| @@ -71,7 +71,11 @@ private static void ProcessGrayscale<TPixel>(Configuration configuration, Buffer | |||
|
|
|||
| for (int y = 0; y < height; y++) | |||
| { | |||
| stream.Read(rowSpan); | |||
| if (stream.Read(rowSpan) == 0) | |||
There was a problem hiding this comment.
This is lame actually. I should have sliced down rowSpan with the result of stream.Read(), for the pixel conversion done later, or used the condition stream.Read(rowSpan) < rowSpan.Length (the latter isstricter, but simpler).
This has little practical relevance (BuffferedReadStream is hitting EOF twice & FromL8Bytes is converting some memory garbage for broken files), so no need to fix it now, but the code looks stupid :)
There was a problem hiding this comment.
We can improve it with follow up.
Prerequisites
Description
Add additonal checks to handle corrupt files better in the PBM decoder.
Also extend
BufferedReadStreamto monitor the number of times it has been made to hit EOF by read calls. This improves the testability of decoder behavior. The perf impact of theBufferedReadStreamchange is within the margin of error.