jp3d: BUG in tcd_init_encode

[ghost]

discussion here

[rouault]

Content of the discussion:

Hi,

Compressing with JP3D as follows

~/openjpeg/openjpeg-2.1.2/build/bin/opj_jp3d_compress -i tibia.bin -m tibia.img -t 128,128,128 -o tibia.jp3d


got segfault when trying to initialize tcd for the second tile.
This happens for any tile size I tried (32, 64, 128).
Note that with no tiles (i.e. ONE tile) works.

gdb tells me that the crash is in
#0  tcd_init_encode ()
#1  j3d_encode ()
#2  main ()

My naive! bug fix in jp3d.c:


    /* initialisation before tile encoding  */
        if (tileno == 0) {
            tcd_malloc_encode(tcd, volume, cp, j3d->curtileno);
        } else {
            tcd_free_encode(tcd);
            tcd_malloc_encode(tcd, volume, cp, j3d->curtileno);
            //this must be kaputt tcd_init_encode(tcd, volume, cp, j3d->curtileno);
        }


Now it works, but I am worry about the performance.

Could somebody fix this the right way?

Hi,

Compressing with JP3D as follows

~/openjpeg/openjpeg-2.1.2/build/bin/opj_jp3d_compress -i tibia.bin -m tibia.img -t 128,128,128 -o tibia.jp3d


got segfault when trying to initialize tcd for the second tile.
This happens for any tile size I tried (32, 64, 128).
Note that with no tiles (i.e. ONE tile) works.

gdb tells me that the crash is in
#0  tcd_init_encode ()
#1  j3d_encode ()
#2  main ()

My naive! bug fix in jp3d.c:


    /* initialisation before tile encoding  */
        if (tileno == 0) {
            tcd_malloc_encode(tcd, volume, cp, j3d->curtileno);
        } else {
            tcd_free_encode(tcd);
            tcd_malloc_encode(tcd, volume, cp, j3d->curtileno);
            //this must be kaputt tcd_init_encode(tcd, volume, cp, j3d->curtileno);
        }


Now it works, but I am worry about the performance.

Could somebody fix this the right way?

[ValZapod]

Second comment from there:

I start to suspect that the bug is the reallocation of tilec->resolutions inside tcd_init_encode which maybe it was thought to be reused (res_max etc).

it works if I comment out line tilec->resolutions = (opj_tcd_resolution_t *) opj_malloc(res_max * sizeof(opj_tcd_resolution_t));

If I am correct about this, then the bug other than leading to a crash, because band->precincts is NULL, because it's not in turn allocated leads to a memory leak (tilec->resolutions)

someone should fix this upstream