SysFont on Windows clobbers base fonts with alternate styled versions

[rmccampbell]

Environment:

Platform:               Windows-11-10.0.22631-SP0
System:                 Windows
System Version:         10.0.22631
Processor:              Intel64 Family 6 Model 154 Stepping 3, GenuineIntel     SSE2: Yes       AVX2: Yes    NEON: No
Architecture:           Bits: 64bit     Linkage: WindowsPE

Python:                 CPython 3.12.5 (tags/v3.12.5:ff3bc82, Aug  6 2024, 20:45:27) [MSC v.1940 64 bit (AMD64)]
pygame version:         2.5.1
SDL versions:           Linked: 2.30.6  Compiled: 2.30.6
SDL Mixer versions:     Linked: 2.8.0   Compiled: 2.8.0
SDL Font versions:      Linked: 2.22.0  Compiled: 2.22.0
SDL Image versions:     Linked: 2.8.2   Compiled: 2.8.2
Freetype versions:      Linked: 2.11.1  Compiled: 2.11.1

Display Driver:         windows
Mixer Driver:           wasapi

Current behavior:

Fonts which have multiple variants (see mods) are normalized on Windows by stripping certain keywords, and if multiple variants have the same name after normalization, only the last version will be kept. This may clobber the base version of fonts. For example, on my computer requesting the font "Arial" results in the actual font "Arial Narrow". Interestingly, this does not strip the word "Regular", so you end up with some odd cases such as "lucidasans" mapping to Lucida Sans Unicode or "lucidafax" to Lucida Fax Demibold while "lucidasansregular" and "lucidafaxregular" map to the plain fonts.

Expected behavior:

1. The regular fonts should be preserved. I.e. "Arial" should be "Arial"
2. Alternate font style variants should be selectable individually. So "Arial Narrow" should give the narrow font.

To preserve some backwards compatibility, I think the best solution is to register fonts with multiple aliases: the full name which only strips bold/italic, and the stripped name which does the complete normalization. If there are collisions between a full name and a stripped name then the full name font should always be preferred. There also should probably be some special casing for "Regular".

Test code

import pygame as pg
pg.init()
s = pg.display.set_mode((640, 480))
s.fill((255, 255, 255))

f = pg.font.SysFont('Arial', 48)
img = f.render('Hello, World!', True, (0, 0, 0))
s.blit(img, (50, 50))

f2 = pg.font.Font('C:/Windows/Fonts/arial.ttf', 48)
img2 = f2.render('Hello, World!', True, (0, 0, 0))
s.blit(img2, (50, 150))

pg.display.flip()
input()

Screenshot of test code
(these should match)

[rmccampbell]

It looks like the same problem occurs for Unix as well, although it doesn't mess with the font name, because it simply ignores extra font styles and keeps the last one with the same base name.

Interestingly, it seems that some of the font entries (on Ubuntu) have two names and two corresponding style tags: one "simple" name like Lato with a "complex" style like Black or Black Italic, and a "complex" name like Lato Black with a corresponding "simple" style of Regular or Italic. For the more basic font variants they only have one name and style. In this case it seems like it would make more sense to use the second font name here to allow access to the more niche styles, and avoid all of them mapping to the same simple name, though there may still be some collisions.