MacSurf

Hardware-verified on G3/G4 OS 9: mactrove.com and Drupal-platinum-themed pages now render at correct viewport width instead of laying out ~2x too wide. The "loads very wide" symptom is closed.

Root cause (fixes194)

layout_minmax_block_inner had a long-standing branch for replaced elements (<img>, <embed>, <object>) that set:

min = max = content_get_width(block->object);  // natural pixel width

It used the natural pixel width of the source bitmap and ignored css_computed_width entirely. NetSurf's hints.c::css_hint_width correctly translates HTML <img width="N"> into CSS_PROP_WIDTH = N, and wtype/width/wunit were already in scope at line 996 — this one branch just discarded them.

After fixes176 added the flex intrinsic-main-size fallback (b->max_width), the natural image width cascaded up the box tree, blowing flex/grid ancestors past the viewport. For mactrove's logo (<img src="logo.png" width="400" height="106"> over a natural 1058×245 PNG), the trace was:

img w=1058 (natural)
  → .page__logo flex w=1058
    → page__header flex w≥1058
      → nested flex w=1788
        → outer block w=1818
          → root descendant_x1 = 2054   (viewport = 949)

The chrome read descendant_x1=2054 as scrollable content width → page laid out at ~2× viewport.

Fix: when the replaced element has an explicit CSS width (px or length, not percentage), use the resolved px as both min/max-content. If only height is set, scale by natural aspect ratio. Otherwise natural intrinsic remains the fallback.

Cascade of fixes shipped in this tag

This release rolls up the work from fixes187 → fixes194:

fixes191 — six-feature theme-completion sprint (a–f)

| Sub | Feature | Approach | Status |
|---|---|---|---|
| 191a | inset shorthand | cssh_css preprocessor expands TRBL longhands with !important propagated | FULL |
| 191b | background-size | New end-of-enum libcss property (0x08D); auto/cover/contain/<length>/2-value forms; consumer in html_redraw_background | PARTIAL — bitmap-only; gradient-bg tiling deferred |
| 191c | position: sticky | Consumer in html_redraw_box adjusts y_parent so painted y never goes above top: | PARTIAL — vertical only, no parent-bottom clamp |
| 191d | object-position | cssh_css drop; apply_object_fit already centres (matches default) | PARTIAL |
| 191e | pointer-events | cssh_css drop | PARTIAL — parse-safe, no hit-test |
| 191f | transition/animation/@keyframes + 14 CSS3 UI knobs | cssh_css modern_compat DROP_PROPS extended | QUICKDRAW_FALLBACK — final static values still apply |

Cleanup chain after the initial fixes191 ship

• fixes192 — CW8 strict-pointer + uninit warning cleanup
• fixes193 — match css__parse_unit_keyword signature exactly

PNG-fade saga preceding fixes191

• fixes187 — PNG mask threshold 128→8 + RGB premultiplied by alpha. Edges sharper, partial fade.
• fixes188 — Scaled-PNG composite path attempt with CopyBits readback + alpha-over composite. Hardware-rejected — transparency disappeared.
• fixes189 — Force ARGB high byte to 0xFF in buf→pm copy. Insufficient.
• fixes190 — Revert fixes188 composite branch. Single CopyMask path. Transparency restored; scaled icons return to fixes187 sharper-but-faded baseline.

Mac-side action

If updating from < fixes191, add to MacSurf.mcp:

• p_background_size.c
• s_background_size.c

Trap-zone discipline followed

• New libcss property (background_size) added at END of enum (0x08D, after ASPECT_RATIO)
• Storage is int32_t self-aligning at END of _i struct (fixes151b padding lesson)
• dispatch.c and s_dispatch.c updated in mirror (fixes116b rule)
• No mid-enum insert, no propstring shuffling

Retro68 syntax-clean on all touched files.

Tests added

Under tests/css/:

• inset.html — inset shorthand probes I1–I5
• background_size.html — BS1–BS6 (cover/contain/explicit/auto)
• sticky.html — sticky sidebar probe
• object_position.html — degraded-to-centre probes
• pointer_events.html — parse-safe probe
• transition_static.html — static-value preservation under dropped declarations