(master) xnest: replace global arrays by devPrivate's

[metux]

instead of having several global arrays for storing additional
per-screen private data, use devPrivate infrastructure.

that way also decoupling that data from sequential screen numbers.

Signed-off-by: Enrico Weigelt, metux IT consult info@metux.net

[metux]

🤖 Automated review — generated by Starfleet ship Enterprise (model: heavy-model) on behalf of @metux. Not a human review.

Verdict: pass — the blocking bug is fixed; CI fully green including the Xnest runtime smoke test.

What changed since the previous review

• xnestScreenPrivateKeyRec is now registered (dixRegisterPrivateKey(&…, PRIVATE_SCREEN, 0),
  hw/xnest/Screen.c, in xnestOpenScreen).
• The priv is allocated before any use (calloc + dixSetPrivate at the top of xnestOpenScreen) and
  freed in xnestCloseScreen.
• xnestScreen() and xnestSaveScreen() now NULL-guard the priv.
• Dead xnestSetScreenPrivate() removed; shared layout moved to xnest-screen_priv.h with a
  static inline getter; new eventmask.c/xnest-eventmask.h wired into meson.build.

Verification

• CI: 20 pass / 0 fail, including xserver-build-ubuntu (builds -Dxnest=true -Dwerror=true
  and runs the X-NAMESPACE/XNS smoke test that actually starts Xnest, build-xserver.yml:150-185)
  — so startup-time registration/alloc is exercised at runtime, not just compiled.
• Pattern matches the canonical PRIVATE_SCREEN idiom (register size 0 + manual dixSetPrivate,
  as in dbe/randr/picture).
• No _X_EXPORT/struct-layout change → no driver-ABI impact (abi-changes-check passes).

Advisory (non-blocking)

1. eventmask.c (new): xnestUpdateEventMask() guards if (walkScreen) but dereferences
   screenPriv->defaultWindow without checking screenPriv — inconsistent with the guarded
   xnestScreen()/xnestSaveScreen(). In the normal lifecycle the priv is always set, but a
   teardown/regen window would NULL-deref. Suggest if (walkScreen && screenPriv).
2. Header hygiene: xnest-screen_priv.h uses DevPrivateKeyRec, dixLookupPrivate() and
   ->devPrivates but only includes screenint.h; a standalone include fails (verified). It
   compiles only because each consumer .c happens to include privates.h/scrnintstr.h first.
   Add #include "privates.h" (+ scrnintstr.h) to make the header self-contained.
3. Teardown: xnestCloseScreen() frees the priv but doesn't clear it (dixSetPrivate(..., NULL)) —
   any late callback after close would use a dangling pointer. Minor for the current single-shot
   lifecycle.

Rules

• Backport: N/A — refactor, targets master.
• Driver-ABI: no impact.

[metux]

🤖 Automated review — generated by Starfleet ship Enterprise (model: heavy-model) on behalf of @metux. Not a human review.

Verdict: pass (review update) — all prior advisories addressed, CI re-running but code-correct.

Update since previous review

• eventmask NULL-guard fixed: xnestUpdateEventMask() now dereferences screenPriv only
  if (screenPriv) (was: if (walkScreen)).
• Header is now self-contained: xnest-screen_priv.h includes include/privates.h +
  include/scrnintstr.h, so DevPrivateKeyRec, dixLookupPrivate(), ->devPrivates resolve
  regardless of include order (verified).
• Bonus cleanup: the duplicated per-screen eventmask loops in Keyboard.c/Pointer.c now
  call the shared xnestUpdateEventMask() helper (eventmask.c).
• Teardown (free() in xnestCloseScreen) is a non-issue — the ScreenRec is torn down right
  after close, so no stale-pointer window.

Still-valid review points (from previous round)

• Key registered (dixRegisterPrivateKey, PRIVATE_SCREEN, 0) + priv calloc'd/dixSetPrivate'd
  at the top of xnestOpenScreen(), before any consumer runs; freed in xnestCloseScreen().
• xnestScreen() and xnestSaveScreen() NULL-guard the priv.
• Remaining unguarded derefs (Color/Cursor/GCOps/Pixmap, xnestWindowParent macro) are only
  reachable on live, fully-open xnest screens where the priv is guaranteed to be set — consistent
  with the documented lifecycle.
• Canonical PRIVATE_SCREEN idiom; no _X_EXPORT/struct-layout change → no driver-ABI impact.

CI

Re-running on this revision (previous revision: 20 pass / 0 fail incl. the ubuntu lane that builds
-Dxnest=true -Dwerror=true and starts Xnest in the X-NAMESPACE smoke test).

• Backport: N/A — refactor, targets master.
• Driver-ABI: no impact.