fix(wow-blp): palettized BLP2 colour map is BGRA, not RGBA — swap R↔B

[samwhosung]

Summary

The BLP2 colour-map for palettized textures (compression = 1) is stored BGRA per the format spec (byte 0 = B, byte 1 = G, byte 2 = R, byte 3 = A). Both the decoder in convert/raw1.rs (all four alpha_bits branches) and the encoder in convert/palette.rs treated the palette as RGBA, so every real Blizzard palettized atlas decoded with R and B channels swapped.

Round-trip tests pass at either swap polarity because the encoder + decoder are equally consistent with themselves — only reading spec-conforming Blizzard data exposes the bug. This is why it has lived in the crate's regression suite undetected.

Reproducer

Read World\NoDXT\Detail\WestfallDetailDoodads01.blp from a vanilla 1.12.1 (build 5875) texture.MPQ (BLP2 v1, compression=1, alpha_size=1, alpha_type=8, 256×256). The authored palette (256 BGRA DWORDs at offset 148) is wheat-field tan: average R=167 G=147 B=78, with 201 of 253 non-zero entries red-dominant and the most-saturated entries being yellows and oranges like (253, 246, 0) and (249, 215, 3).

Decoded via wow-blp 0.6.4, the output is saturated cool blue/teal: average (78, 147, 167) — the same palette with R and B swapped.

Downstream impact

Discovered while chasing a "Westfall clutter is blue" visual bug in a from-scratch 1.12.1 client (mindforge-client). After patching the crate via [patch.crates-io], the bug fully resolves. Across a full vanilla 1.12.1 ./WoW/Data MPQ chain, the palettized code path is hit by:

Shared clutter atlases (World\NoDXT\Detail\…) — 4 files affected:

• WestfallDetailDoodads01.blp (Westfall)
• DeadwindDetails01.blp (Deadwind Pass)
• ArathiHighlandsDetails.blp (Arathi Highlands)
• EasternPlaguelandsDetails_256.blp (Eastern Plaguelands)

The other 33 shared clutter atlases (Elwynn, Duskwood, Stranglethorn, Tirisfal, Barrens, Durotar, Mulgore, Burning Steppes, Felwood, Hyjal, Un'Goro, Tanaris, Silithus, Loch Modan, Redridge, Silverpine, …) ship as DXT (compression=2), which takes the separate dxtn.rs code path that handles channel order via squish. Hence "fine everywhere except Westfall" as a visible diagnosis.

Terrain TILESET\…\*.blp textures — 13 of 1050 affected:

• 11 in Tileset\EmeraldDream\… (Emerald Dream — not a player-accessible map in 1.12)
• 2 in Tileset\RedRidge\… (RedridgeRockBase_s.blp, RedridgeRockRoadBase_s.blp — both _s specular auxiliaries, not the diffuse colour)

The remaining 1037 terrain BLPs are DXT and were unaffected — the visible terrain colour of every accessible zone was correct before and after.

The fix

Decoder (convert/raw1.rs, applied in all four alpha_bits branches):

// before — R in the B slot, B in the R slot:
pixel.0[0] = (color & 0xFF) as u8;          // = byte 0 = B
pixel.0[1] = ((color >> 8) & 0xFF) as u8;   // = byte 1 = G
pixel.0[2] = ((color >> 16) & 0xFF) as u8;  // = byte 2 = R

// after — spec-conforming BGRA:
pixel.0[0] = ((color >> 16) & 0xFF) as u8;  // R from byte 2
pixel.0[1] = ((color >> 8) & 0xFF) as u8;   // G from byte 1
pixel.0[2] = (color & 0xFF) as u8;          // B from byte 0

Encoder (convert/palette.rs) fixed in matching direction so the crate's internal round-trip stays consistent. All existing tests pass.

A note on regression testing

This bug demonstrates a known weakness of round-trip-only regression tests for serialisation formats: the encoder + decoder were equally consistent at the wrong polarity, so the round-trip never caught it. A useful follow-up (out of scope for this PR) would be adding a small golden-vector test against a checked-in palette + indices fixture with hand-verified RGB expected output — that would catch the kind of "two-wrongs round-trip green" failure mode this PR fixes.

Closes / refs

Not aware of an existing issue tracking this; happy to open one and link if preferred.

[codecov]

Codecov Report

❌ Patch coverage is 20.45455% with 35 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
file-formats/graphics/wow-blp/src/convert/raw1.rs	0.00%	20 Missing ⚠️
file-formats/graphics/wow-m2/src/model.rs	42.85%	12 Missing ⚠️
...le-formats/graphics/wow-blp/src/convert/palette.rs	0.00%	3 Missing ⚠️

📢 Thoughts on this report? Let us know!

[danielsreichenbach]

I merged this together with an update to the actual comments referencing this to make it a bit more complete. Thanks very much for the submission.