2008.08.19 05:17 "[Tiff] Regarding DICONDE and its Specification", by Harsha

2008.08.22 13:11 "Re: [Tiff] creating sparse files......", by Toby Thain

On 22-Aug-08, at 5:56 AM, Rogier Wolff wrote:

> Hi,
> I'm stitching kind of large panoramas. This results in big

> intermediate files. On my last run, which took overnight to stitch, > I thought 42 Gb of free disk space would be enough. Wrong!

> I got over thrity files of over 1.2Gbytes, filling up the disk.

> It turns out that most of the files contain lots of zeroes. On Unix > this can be stored effciently by not issueing a "write" with a buffer

> full of zeroes, but by seeking over the area. The operating system > will act as if the area was filled with zeroes.

I'v modified "tif_unix.c" to read:

static int isallzero (tdata_t buf, tsize_t size)
        int i;
        for (i=0;i<size;i++)
                if (buf[i]) return 0;
        return 1;

This appears unnecessarily inefficient. It's going to be much cheaper to test whole longwords (or whatever data alignment will allow) than byte by byte. (I wonder if something clever can be done with MMX/SSE on newer chips?)

static tsize_t
_tiffWriteProc(thandle_t fd, tdata_t buf, tsize_t size)
        if (isallzero (buf, size))
                return (lseek ((int) fd, (off_t)size, SEEK_CUR));
                return ((tsize_t) write((int) fd, buf, (size_t) size));

This penalises all writes to optimise a very special case. Not saying it can't be rationalised, but pros and cons can be debated.

just my 2c

static toff_t
_tiffSeekProc(thandle_t fd, toff_t off, int whence)
        return ((toff_t) lseek((int) fd, (off_t) off, whence));

static int
_tiffCloseProc(thandle_t fd)
        toff_t curoffset;
        toff_t fsize;

        curoffset = _tiffSeekProc (fd, 0, SEEK_CUR);
        fsize     = _tiffSeekProc (fd, 0, SEEK_END);

        if (curoffset > fsize) {
                ftruncate((int)fd, curoffset);

        return (close((int) fd));

I haven't tested it yet. Maybe someone here has the same problem I have, and is willing to do a test-run. I'll make a proper patch later on. This is just to allow you guys know and to invite comments....

While writing this Email message, I've found a bug: If an area is written, and then later zeroed by the application program the new code won't work. I'll have to think about how to get this case right. In practise a modified libtiff should be easy to plug into "hugin" and the results should be visible, the bug-situation will probably not trigger.

(I used to be able to hit the "compress" checkbox inside hugin, and it would work. But it seems to have gone "dead" in the version currently installed on my machine.)