revised/new patch for improve ratio calculation

[alexander-n8hgeg5e]

Here is the description about why a change can be considered:
description right at the beginning of this closed pull request

Now, I tested a function that actually works and made screenshots.
It works with upright and landscape ratios and
also with my "ratio-1.0-square" use case.
Ratios > 1 are "landscape" and ratios < 1 are "upright".

It works like this:
the code from below more readable
`
def is_more_golden_than( a, b , gs=1):
goldstandard = gs
ao=a
bo=b
# TODO: use math to refactor the next 6 lines
a = a / goldstandard
b = b / goldstandard
wa = a if a < 1 else -1/a
wb = b if b < 1 else -1/b
wa = abs(abs(wa)-1)
wb = abs(abs(wb)-1)
judgement = True if wa < wb else False
print(f'{ao:>5.3} vs {bo:<5.3} > {wa:+.3f}, {wb:+.3f} golden = {gs:0.2f} {judgement}' )
return judgement

is_more_golden_than(1.5 , 0.66666 , gs=1 )
is_more_golden_than(2.0 , 2.1 , gs=2 )
is_more_golden_than(1.9 , 2.0 , gs=2 )
is_more_golden_than(1.9 , 2.1 , gs=2 )
is_more_golden_than(0.25, 0.75 , gs=0.5)

output:
1.5 vs 0.667 > +0.333, +0.333 golden = 1.00 True
2.0 vs 2.1 > +0.000, +0.048 golden = 2.00 True
1.9 vs 2.0 > +0.050, +0.000 golden = 2.00 False
1.9 vs 2.1 > +0.050, +0.048 golden = 2.00 False
0.25 vs 0.75 > +0.500, +0.333 golden = 0.50 False
`
update here: pictures removed, newer pictures below

[alexander-n8hgeg5e]

For the failing test_ratiotile.py test,
maybe I should make a script that draws rectangles and creates images with the ratios and screen sizes from the test side-by-side with the expected ones.
Can I update pull requests if they are fixed or do I need to close it ?

[elParaguayo]

Just keep adding commits to your branch.

[alexander-n8hgeg5e]

I got the data out and generated some side by side pictures which are only 5k each.
update here: most of this comment removed, because I added a comment with a condensed and newer version of all this.

[alexander-n8hgeg5e]

I rebased and force pushed it to clean it up.
It needs probably more cleanup after the code analysis tools
inspected it.
update: now, they seem to pass
I can make a new set of the pictures.
Because it's probably good to see what the changed resize test does.

[alexander-n8hgeg5e]

This here is everything summarized and more organized.
This is a new set of pictures based on the rebased commits of the pull request.
For the pull request I had to adjust the "add_windows" test and the "resizing" test.
Below are links to 2 branches in my repo that explain the generation of the pictures and show the way they are derived from the pull request. For the pictures of the "add windows" test I had to remove a commit from the pull request, because, to make the test pass, the commit changes the values of two "layout-infos" and the original values are needed for the side-by-side view.

For the "add_windows" test:

branch that made the pictures for the "add_windows" test
The commit "output window data from test_add_windows" shows that the previous expected values remain in the code and are used for the generation of the side-by-side pictures.

side-by-side pictures for the "add_windows" test

For the resizing test:

branch that made the pictures for the resizing test

pictures for the "resizing" test

[elParaguayo]

Can you explain, in as few words as possible, what the issue is, what your PR does and what the images above show? I'm finding this quite hard to follow but I appreciate you've put a lot of effort in here.

[alexander-n8hgeg5e]

Can you explain, in as few words as possible, what the issue is, what your PR does and what the images above show? I'm finding this quite hard to follow but I appreciate you've put a lot of effort in here.

The issue is, the layout calculation does not make the windows as close to the desired ratio as possible.
This PR would improve it.
Why it works? I'm not sure I know it myself but it turned out it works.
The pictures of the "add_windows" test show that it did not change that much in this test.
The pictures of the other test (resizing) are intended to show that the test, which I completely changed, makes sense (hopefully) and are meant to read together with this commit (one picture is taken at every "assert sizes()" statement):
commit

The issue in detail:

If the desired ratio=1 and there are 3 windows on a screen that is 1.125 higher than wide,
then the result is 3 windows 1.125 tall and 0.333 wide. (screen width=1)
Then the ratio is 1.125/(1/3)=3.375
But it would be better to have 3 "landscape" windows because they would be 1.125/3 in height and 1 wide, which is better because if you see the ratio as longer_edge_of_window/shorter_edge_of_window then it is 1/(1.125/3)= 2.666 (closer to 1)
(note: the ratios in this description are inverse(1/x) to the ones in class ratioTile)

I know that this is not the most important issue and it requires some time to supervise something like this.

[github-actions]

This PR is stale because it has been open 90 days with no activity. Remove the status: stale label or comment, or this will be closed in 30 days.