r3.null creates a lot of new unexpected null cells likely by creating a null tile

[wenzeslaus]

When running r3.null I get a lot of NULL cells as a result regardless of input. When I show it in 3D as slices one part of the raster is missing. It looks like all values in one tile are NULLs (the tile is a NULL tile).

It seems that the computational region is important. I don't get the behavior with arbitrary region.

> g.region -3 n=220180 s=219900 w=638360 e=638720 t=153 b=106 res=2 res3=2

When replacing value by NULL, I get more NULL cells than expected:

> r3.mapcalc --o expr="test_1 = float(col() + row() + depth())"
> r3.univar map=test_1
total null and non-null cells: 604800
total null cells: 0
...
> r3.null map=test_1 setnull=10
> r3.univar map=test_1
total null and non-null cells: 604800
total null cells: 40356
...

When replacing NULLs by value, I get a lot of NULL values.

> r3.mapcalc --o expr="test_1 = float(col() + row() + depth())"
> r3.univar map=test_1
total null and non-null cells: 604800
total null cells: 0
...
> r3.null map=test_1 null=10
> r3.univar map=test_1
total null and non-null cells: 604800
total null cells: 40320
...

There is a lot of different Rast3d_*lock*() calls in the code but when I comment out the call of Rast3d_flush_tiles_in_cube(), I get the right results (there is Rast3d_flush_all_tiles() at the end). Documentation of Rast3d_flush_tiles_in_cube() says tiles which are not stored in the cache are written as NULL-tiles.

[wenzeslaus]

In 71421:

r3.null: remove code which causes nulls (maybe empty tiles) to be written to the map (tested by r71420, see #2992)

[wenzeslaus]

Added existing tests in r71418, tests for this ticket in r71420, removed the problematic code in r71421 from G7:r3.null. Tests now run.

This still points to a possible bug in the library.

[neteler]

Ticket retargeted after milestone closed

[martinl]

Any progress?

[neteler]

Ticket retargeted after milestone closed

[martinl]

Bump milestone to 7.4.4

[neteler]

Ticket retargeted after milestone closed