image/gif: "not enough image data" on gif that works in browser

[skyqinsc]

There are images which browsers can render but which image/gif chokes on.

The following runs successfully for most gifs:

import (
    "fmt"
    "image/gif"
    "os"
)

const example = "sample.gif"

func main() {
    fmt.Println("attempting to read gif")

    file, err := os.Open(example)
    if err != nil {
        fmt.Println("Can't open this file:", err)
        return
    }

    img, err := gif.Decode(file)
    if err != nil {
        fmt.Println("Can't decode this as a gif:", err)
        return
    }

    _ = img.Bounds()

    fmt.Println("gif decoded")
}

But when trying to run it on a particular gif (below), this happens:

$ go run main.go

attempting to read gif
Can't decode this as a gif: gif: not enough image data

Here’s the offending gif

[ALTree]

While my browser does indeed display it, that image is definitely corrupted and/or badly formed; imagemagick says:

$ curl -s -o img.gif https://user-images.githubusercontent.com/12746710/81430635-5412ea80-9192-11ea-9a62-80208204bf45.gif
$ identify -verbose img.gif 
identify-im6.q16: corrupt image `img.gif' @ error/gif.c/DecodeImage/513.
identify-im6.q16: corrupt image `img.gif' @ error/gif.c/ReadGIFImage/1384.

and my system's image viewer also refuses to open it.

This is similar to #9856, when image/gif was fixed to handle bad gifs a little better.

cc @nigeltao

[as]

To read this image, there need to be two changes to the gif decoder. I have made these changes in a modified reader.go (see below), which fix the problem, but the solution needs a more-refined implementation that I don't have time to implement.

(1) Allow the lzw reader to encounter short reads in the stream.

go/src/image/gif/reader.go

Line 427 in 4f65fb3

return errNotEnough

Example:
https://github.com/as/goissues/blob/173a03c197057be2f0fd7a842ddb4542e7b2e1e3/38958/image_gif_reader.go.txt#L430

(2) Handle zero-bytes sandwiched inside parts of the image.

go/src/image/gif/reader.go

Lines 243 to 258 in 4f65fb3

	case sExtension:
	if err = d.readExtension(); err != nil {
	return err
	}
	case sImageDescriptor:
	if err = d.readImageDescriptor(keepAllFrames); err != nil {
	return err
	}
	case sTrailer:
	if len(d.image) == 0 {
	return fmt.Errorf("gif: missing image data")
	}
	return nil

Example:
https://github.com/as/goissues/blob/173a03c197057be2f0fd7a842ddb4542e7b2e1e3/38958/image_gif_reader.go.txt#L252-L255

After this, the author's test program correctly decodes the image.

cp /go/src/image/gif/reader.go /go/src/image/gif/reader.go.old
curl https://raw.githubusercontent.com/as/goissues/master/38958/image_gif_reader.go.txt > /go/src/image/gif/reader.go

[nigeltao]

Handle zero-bytes sandwiched inside parts of the image.

Coincidentally,
https://go-review.googlesource.com/c/go/+/232057
is in review. Does patching in that change fix it?

[as]

@nigeltao That patch is necessary to fix it, but alone isn't sufficient. The LZW reader case (1) also needs to be handled.

[skyqinsc]

This is another sample which browsers also can render but which image/gif chokes on, but something is different.
$ go run main.go

attempting to read gif
Can't decode this as a gif: gif: reading image data: lzw: invalid code

@as @nigeltao

[as]

The above image is even-more problematic. It contains a byte interpreted as an invalid LZW code (0x66) instead of (0x00) like the first image. To fix that, I had to elide almost all error checks and then ignore the 0x66 byte.

Perhaps other decoders silently ignore garbage input entirely and march forward hoping to eventually draw a valid-looking image. Which encoder produced these?

The image contains this metadata (which may give us a clue above the question above):

NETSCAPE2.0�����!��XMP DataXMP<?xpacket begin="" id="W5M0MpCehiHzreSzNTczkc9d"?> <x:xmpmeta xmlns:x="adobe:ns:meta/" x:xmptk="Adobe XMP Core 5.6-c142 79.160924, 2017/07/13-01:06:39        "> <rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"> <rdf:Description rdf:about="" xmlns:xmpMM="http://ns.adobe.com/xap/1.0/mm/" xmlns:stRef="http://ns.adobe.com/xap/1.0/sType/ResourceRef#" xmlns:xmp="http://ns.adobe.com/xap/1.0/" xmpMM:OriginalDocumentID="xmp.did:7fc53bc1-ea46-e144-975e-5ee1c3561563" xmpMM:DocumentID="xmp.did:8EB38125D86411E9B26CC13B008A9B86" xmpMM:InstanceID="xmp.iid:8EB38124D86411E9B26CC13B008A9B86" xmp:CreatorTool="Adobe Photoshop CC 2018 (Windows)"> <xmpMM:DerivedFrom stRef:instanceID="xmp.iid:60b609bd-629e-ce49-ade4-1fc8e36fac43" stRef:documentID="adobe:docid:photoshop:bea01de1-6c51-3446-9bed-3ffc92398997"/> </rdf:Description> </rdf:RDF> </x:xmpmeta> <?xpacket end="r"?>