Conversion P to L wrong?

[CTPaHHuK-HEbA]

from PIL import Image

img = Image.open('p.png')
print('P->L: ', img.convert('L').getpixel((0, 0)))
print('P->RGB: ', img.convert('RGB').getpixel((0, 0)))
print('P->RGB->L: ', img.convert('RGB').convert('L').getpixel((0, 0)))

p.png:

Old
PIL: 6.2.2
P->L: 175
P->RGB: (177, 175, 175)
P->RGB->L: 175

7.0.0+
After commit #4320
PIL: 8.3.2
P->L: 175
P->RGB: (177, 175, 175)
P->RGB->L: 176

[radarhere]

This is because P->L uses the L macro, but RGB->L uses the L24 macro.

So this change would make P->L be 176 as well.

diff --git a/src/libImaging/Convert.c b/src/libImaging/Convert.c
index 9012cfcd7..cd6e536cb 100644
--- a/src/libImaging/Convert.c
+++ b/src/libImaging/Convert.c
@@ -1013,7 +1013,7 @@ p2l(UINT8 *out, const UINT8 *in, int xsize, const UINT8 *palette) {
     int x;
     /* FIXME: precalculate greyscale palette? */
     for (x = 0; x < xsize; x++) {
-        *out++ = L(&palette[in[x] * 4]) / 1000;
+        *out++ = L24(&palette[in[x] * 4]) >> 16;
     }
 }

But I'm by no means certain that this is the correct solution.

[radarhere]

I've created #5824 to fix this by applying rounding to palette conversion methods.