r.terraflow should support large file dimensions by default

[mankoff]

Currently r.terraflow is limited to rasters with width or height < SHRT_MAX (2^{15, 32678). I imagine this default made sense years ago when this was coded, but now on even a decent laptop r.watershed works with rasters that large. I suggest r.terraflow remove this limit so that users with large rasters do not need to recompile it. Patch attached and inline below:}

Index: raster/r.terraflow/types.h
===================================================================
--- raster/r.terraflow/types.h	(revision 73638)
+++ raster/r.terraflow/types.h	(working copy)
@@ -26,8 +26,8 @@
 /* input parameters type */
 /* ------------------------------------------------------------ */
 
-typedef short dimension_type; /* represent dimension of the grid */
-static const dimension_type dimension_type_max=SHRT_MAX;
+typedef long dimension_type; /* represent dimension of the grid */
+static const dimension_type dimension_type_max=LONG_MAX;
 
 typedef short direction_type;  /* represent the direction of a cell */
 static const direction_type DIRECTION_UNDEF=-1;

[mmetz]

Replying to mankoff:

Currently r.terraflow is limited to rasters with width or height < SHRT_MAX (32678). I imagine this default made sense years ago when this was coded, but now on even a decent laptop r.watershed works with rasters that large.

Makes sense. Note that the concept of r.terraflow is to avoid seek'ing on slow spinning harddrives. With SSD's, there is no measurable speed penalty when seek'ing, and r.terraflow looses this advantage over e.g. r.watershed.

I suggest r.terraflow remove this limit so that users with large rasters do not need to recompile it. Patch attached and inline below:

Index: raster/r.terraflow/types.h
===================================================================
--- raster/r.terraflow/types.h	(revision 73638)
+++ raster/r.terraflow/types.h	(working copy)
@@ -26,8 +26,8 @@
 /* input parameters type */
 /* ------------------------------------------------------------ */
 
-typedef short dimension_type; /* represent dimension of the grid */
-static const dimension_type dimension_type_max=SHRT_MAX;
+typedef long dimension_type; /* represent dimension of the grid */
+static const dimension_type dimension_type_max=LONG_MAX;
 
 typedef short direction_type;  /* represent the direction of a cell */
 static const direction_type DIRECTION_UNDEF=-1;

GRASS uses int as type for rows and columns which should be more than sufficient. long, if it is a 64 bit integer, is too much, and long is sometimes 32 bit, sometimes 64 bit, therefore I suggest to change short to int.

[mankoff]

Replacing patch with INT instead of LONG

[mankoff]

When I run it with type int it crashes with:

------------------------------
COMPUTING FLOW DIRECTIONS
classifying nodata (inner & boundary)
finding nodata
r.terraflow: 3scan.h:140: void scan3line(FUN&, AMI_STREAM<T>*, AMI_STREAM<T>*, AMI_STREAM<T>*, BASETYPE, dimension_type) [with T = float; BASETYPE = float; FUN = detectEdgeNodata; dimension_type = int]: Assertion `ae == AMI_ERROR_END_OF_STREAM' failed.
Aborted (core dumped)

This does not occur if I use long.

[mankoff]

Upgrading milestone so that this ticket is migrated to GitHub

[neteler]

Ticket retargeted after milestone closed

[neteler]

Ticket retargeted after milestone closed

[neteler]

Ticket retargeted after milestone closed