new colors branch with clean history

[pmocz]

meant to replace: #786

Todo:

• update changelog
• move rest of custom things out of run_stars_extras.f90
• add new colors options to rsp tests, which used to use old colors?
• add table caching (for future PR)

[warrickball]

Do we still use this variable at all, and it really is now scalar instead of array? Or should we just delete it entirely?

Suggested change

	character(len=strlen) :: color_file_names !(max_num_color_files)
	integer :: color_num_colors !(max_num_bcs_per_file)

[warrickball]

If we're no longer using anything from colors_def, can we not just delete this?

Suggested change

use colors_def !, only: max_num_color_files, max_num_bcs_per_file

If we are using anything, the only clause is useful for tracing where a subroutine etc. came from (because Fortran doesn't have namespacing like other languages).

[warrickball]

This (and the other loops) look like an elaborate way of writing

Suggested change

	! Function values at corners
	DO iz = 1, 2
	DO iy = 1, 2
	DO ix = 1, 2
	! Function values
	IF (ix == 1) THEN
	IF (iy == 1) THEN
	IF (iz == 1) THEN
	sum = sum + h00_x * h00_y * h00_z * values(ix,iy,iz)
	ELSE
	sum = sum + h00_x * h00_y * h01_z * values(ix,iy,iz)
	END IF
	ELSE
	IF (iz == 1) THEN
	sum = sum + h00_x * h01_y * h00_z * values(ix,iy,iz)
	ELSE
	sum = sum + h00_x * h01_y * h01_z * values(ix,iy,iz)
	END IF
	END IF
	ELSE
	IF (iy == 1) THEN
	IF (iz == 1) THEN
	sum = sum + h01_x * h00_y * h00_z * values(ix,iy,iz)
	ELSE
	sum = sum + h01_x * h00_y * h01_z * values(ix,iy,iz)
	END IF
	ELSE
	IF (iz == 1) THEN
	sum = sum + h01_x * h01_y * h00_z * values(ix,iy,iz)
	ELSE
	sum = sum + h01_x * h01_y * h01_z * values(ix,iy,iz)
	END IF
	END IF
	END IF
	END DO
	END DO
	END DO
	! Function values at corners
	sum = sum + h00_x * h00_y * h00_z * values(1,1,1)
	sum = sum + h00_x * h00_y * h01_z * values(1,1,2)
	sum = sum + h00_x * h01_y * h00_z * values(1,2,1)
	sum = sum + h00_x * h01_y * h01_z * values(1,2,2)
	sum = sum + h01_x * h00_y * h00_z * values(2,1,1)
	sum = sum + h01_x * h00_y * h01_z * values(2,1,2)
	sum = sum + h01_x * h01_y * h00_z * values(2,2,1)
	sum = sum + h01_x * h01_y * h01_z * values(2,2,2)

I think you can rather loop over this (and possibly the next blocks too) if h??_? become an array h(2,2,3).

In Fortran, you can declare different array bounds if you want e.g.

real(dp) :: my_array(0:3)

if you want e.g. lower bound 0 and upper bound 3.

[warrickball]

Ah, to loop, values and d{x,y,z}_values would also have to become an array, I think of shape (4,2,2,2).

[warrickball]

Can we use the Fortran instrinsic MINLOC here? Something like

Suggested change

	min_dist = ABS(x_val - x_grid(1))
	i_x = 1
	DO i = 2, SIZE(x_grid)
	dist = ABS(x_val - x_grid(i))
	IF (dist < min_dist) THEN
	min_dist = dist
	i_x = i
	END IF
	END DO
	i_x = MINLOC(ABS(x_val - x_grid))

[nialljmiller]

Yep, I have changed this. Thank you

[warrickball]

For bit-for-bit reproducibility, use pow or powN from our math_lib, or just write the multiplications out explicitly. E.g.

Suggested change

	h = 2.0_dp*t**3 - 3.0_dp*t**2 + 1.0_dp
	h = 2.0_dp*t*t*t - 3.0_dp*t*t + 1.0_dp

[nialljmiller]

Thank you I did not know that

[rhdtownsend]

Many of the DEALLOCATE statements at the end of routines are superfluous -- allocatable arrays that are allocated inside a routine are automatically deallocated on exit (unless they are dummy arguments).

[pmocz]

This function is already defined above in the class? Is this re-definition needed/used?

[nialljmiller]

This was a mistake as I was working to remove pandas dependency. I have changed this now.