Fix special case of box size == 1 in jp2image.cpp

[pydera]

exiv2/src/jp2image.cpp

Line 781 in 05ec053

// FIXME. Special case. the real box size is given in another place.

[clanmills]

I've found a collection of JP2 test images here: https://github.com/openpreserve/jpylyzer-test-files_

I'm very pleased to say that both tvisitor and exiv2 read all the files without crashing.

I'm even more pleased to say that tvisitor is about 25 times faster than Exiv2, and reads every file without any error messages except is_codestream.jp2 which is currently unsupported as we have to wait for the Standard to be defined.

The code in src/bmffimage.cpp to respect box.length == 1 is list below. I know this code is used somewhere because I recall debugging it! Although I believe this code can be easily ported into jp2image.cpp, it's very desirable to have a test JP2 image that exercises the code when it has been added to jp2image.cpp.

    long BmffImage::boxHandler(std::ostream& out /* = std::cout*/ , Exiv2::PrintStructureOption option /* = kpsNone */,int depth /* =0 */)
    {
        long result  = (long)io_->size();
        long address = (long)io_->tell();
        // never visit a box twice!
        if ( depth == 0 ) visits_.clear();
        if (visits_.find(address) != visits_.end() || visits_.size() > visits_max_) {
            throw Error(kerCorruptedMetadata);
        }
        visits_.insert(address);

        bool bTrace    = option == kpsBasic || option == kpsRecursive ;

        BmffBoxHeader box = {0, 0};
        if (io_->read((byte*)&box, sizeof(box)) != sizeof(box))
            return result;

        box.length = getLong((byte*)&box.length, endian_);
        box.type = getLong((byte*)&box.type, endian_);
        bool bLF = true;

        if ( bTrace ) {
            bLF = true;
            out << indent(depth) << "Exiv2::BmffImage::boxHandler: " << toAscii(box.type)
                << Internal::stringFormat(" %8ld->%u ", address, box.length);
        }

        if (box.length == 1) {
            DataBuf data(8);
            io_->read(data.pData_, data.size_);
            result = (long) getULongLong(data.pData_, endian_);
            // sanity check
            if (result < 8 || result+address > (long)io_->size()) {
                result = (long)io_->size();
                box.length = result;
            } else {
                box.length = (long) (io_->size() - address);
            }
        }

        // read data in box and restore file position
        ....
    }

[clanmills]

More test images here: https://github.com/uclouvain/openjpeg-data

[clanmills]

The following files in my book use box.length == 1

svn://dev.exiv2.org/svn/team/book/files

2021-02-13-1929.heic
Canon.CR3
cr3.cr3
IMG1.HEIC
IMG_3578.HEIC
Photo-21-11-20.heic
Stonehenge.heic
VID_20180630_164713.mp4

[clanmills]

This issue and #1525 should be consider at the same time. I think the correct way forward is to remove jp2image.cpp from the code base and port jp2image::writeMetadata() into bmffimage.cpp.

We will only add writeMetadata() support for JP2 files. Other bmff files such as HEIC, AVIF, JPX and CR3 are read-only.

[piponazo]

I have been lately reading the JP2 standard and I would like to improve the support for such standard. I'll dig into this issue in the next days 😉