fix(physics): traverse prepared indoor portal topology
This commit is contained in:
parent
b12d94047c
commit
4d095be286
9 changed files with 400 additions and 5 deletions
|
|
@ -343,6 +343,15 @@ GPU/frame-time and ordinary-process memory comparison remains a physical-
|
|||
display measurement, not a Slice-I correctness blocker. Evidence:
|
||||
[`../research/2026-07-25-slice-i7-closeout.md`](../research/2026-07-25-slice-i7-closeout.md).
|
||||
|
||||
**Corrective gate 2026-07-26.** Later connected house/dungeon testing exposed
|
||||
one I7 consumer omission: `CEnvCell::find_transit_cells` still required the
|
||||
deleted parsed portal-polygon dictionary, even though the prepared topology
|
||||
and polygon table retained the exact same plane. The production consumer now
|
||||
uses that direct prepared index. Synthetic graph-free tests and the captured
|
||||
Holtburg cottage camera path prove identical graph/prepared results; no
|
||||
package rebake or parsed-graph restoration is needed. Evidence:
|
||||
[`../research/2026-07-26-prepared-indoor-transit-regression.md`](../research/2026-07-26-prepared-indoor-transit-regression.md).
|
||||
|
||||
## 4. Commit order
|
||||
|
||||
Each unit is independently buildable and reversible:
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue