I’ve read a GeoTIFF file into a std::vector<std::uint8_t> which gets passed around as a const ref to various functions. One of those functions I need to open the image as a TIFF* so I can read some of the tags. As these files can be very large (10s or 100s of megabytes), I don’t want to copy the image in memory. I’ve found these answers which have similar approaches and definitely looked promising, but then TIFFStreamOpen fails using the following code:

class vectorbuf : public std::streambuf 
{
public:
    vectorbuf( std::vector<uint8_t>& v )
    {
        setg( (char*)v.data(), (char*)v.data(), (char*)(v.data() + v.size()) );
    }

    ~vectorbuf() {}
};

void GeotiffUtils::ProcessImage( const std::vector<std::uint8_t>& imgData )
{
    detail::vectorbuf strmBuf( const_cast<std::vector<std::uint8_t>&>(imgData) );
    std::istream memStrm( &strmBuf );

    TIFF* memTiff = TIFFStreamOpen( "MemTiff", &memStrm );

    std::uint16_t orientation = 1;
    auto success = TIFFGetField( memTiff, TIFFTAG_ORIENTATION, &orientation );

    //... (other processing)

    TIFFClose( memTiff );

    //... (even more processing)
}

I have been able to use TIFFStreamOpen by copying the data into a std::string and std::istringstream like so:

void GeotiffUtils::ProcessImage( const std::vector<std::uint8_t>& imgData )
{
    std::string imgString( imgData.begin(), imgData.end() );
    std::istringstream memStrm( imgString, std::ios::binary | std::ios::in );

    TIFF* memTiff = TIFFStreamOpen( "MemTiff", &memStrm);

    std::uint16_t orientation = 1;
    auto success = TIFFGetField( memTiff, TIFFTAG_ORIENTATION, &orientation );

    //... (other processing)

    TIFFClose( memTiff );

    //... (even more processing)
}

This is obviously less than ideal with large files.

Is there a reason why the copied data can be read successfully but not the original vector?