r3.univar: Bring r3.univar implementation closer to r.univar

[echoix]

This isn't much, but I happened to see when looking why there were some differences in the strings used between r and r3 versions of some code, that the implementations were more diverging than only what is needed for handling the third dimension.

I had the two files compared, side by side, and with moved blocks detection enabled to see.
Some parts of r3 had more understandable variable names, whilst the r version had gui section tags that I copied over to the r3 version.

There was a static function "univar_stat_with_percentiles" in r.univar that I would've like to use as is in the r3 version, but since it is "static", it doesn't work to define it in another file. Is duplicating that function the correct approach?

I've read about the sizeof(type) vs sizeof expression, and kinda concluded that it doesn't really matter to parenthesize an expression. I verified a sample little program on Godbolt: https://godbolt.org/z/Wc31Pboqx

[echoix]

Solved conflicts

[petrasovaa]

Looking at it again, I would move the function to stats.c, no?

[echoix]

I tried, but a static function doesn't show up to other files. It was getting harder and I didn't feel comfortable (as I couldn't evaluate the impacts), of changing the header file or not being static.

You're free to adapt this as you'd like. The function in the 2D and 3D version is either exactly the same, or almost the same. I opened both in comparison mode (with moved code detection) to see what diverged.

[petrasovaa]

I think you should move this to stats.c without the "static", and add it to globals.h. I don't see problem with that.

[echoix]

And is globals.h becoming something part of the "public" api? I imagine it should be something available only for the implementation of the two files that use it.

[echoix]

Looking at it again, I would move the function to stats.c, no?

I think you should move this to stats.c without the "static", and add it to globals.h. I don't see problem with that.

@petrasovaa A bit like this? It should work?

[echoix]

The remaining big differences between r.univar vs r3.univar:

• r.univar has OpenMP support.
• r.univar has nprocs parameter
• r.univar has an "r" flag: "Use the native resolution and extent of the raster map, instead of the current region"
• r.univar uses kahan sum (kahan_sum defined in that file)
• r.univar seems to have a loop to count input rasters, but r3.univar handles only one (see standard options G_OPT_R_MAPS vs G_OPT_R3_MAP)
• r.univar has open_raster and process_raster functions, some of it seems related to having OpenMP code more localized in process_raster

[echoix]

I tried, but a static function doesn't show up to other files. It was getting harder and I didn't feel comfortable (as I couldn't evaluate the impacts), of changing the header file or not being static.

You're free to adapt this as you'd like. The function in the 2D and 3D version is either exactly the same, or almost the same. I opened both in comparison mode (with moved code detection) to see what diverged.

The function univar_stat_with_percentiles was exactly the same between implementations (in r3.univar, it was added in a previous version of this PR, not on main yet)

[nilason]

I've read about the sizeof(type) vs sizeof expression, and kinda concluded that it doesn't really matter to parenthesize an expression.

Parentheses are required with types, as in sizeof(int) or sizeof(char), otherwise (variables) they are not. I always use it for clarity.

And is globals.h becoming something part of the "public" api? I imagine it should be something available only for the implementation of the two files that use it.

The modals are all non-public (end up in executables only), no header files are distributed. So, in this case the globals.h is available for all source files (that include it) in the raster/r.univar directory. Note in Makefile:

r_univar_OBJS = r.univar_main.o sort.o stats.o
r3_univar_OBJS = r3.univar_main.o sort.o stats.o

the stats.c and sort.c file are compiled with both r.univar and r3.univar.

[echoix]

Thanks for the theory, I appreciate it!

Addressed and approved by another after that?