Release machine acceleration, dropped books and baseline defaults for both tracks
Build and smoke / changes (push) Waiting to run

This commit is contained in:
dnorberg committed 2026-11-18 16:43:39 -05:00
1 parent d199ac253e
commit 6630ed7ab1
72 files changed
+2390 -108

No files matched your search

+81
View File
@@ -0,0 +1,81 @@
# Accelerated server machines — issue #112
Introduced in development candidates **1.21.1.0.7** and **26.3.0.11**.
Player in-game checks passed; included in releases **1.21.1.0.12** and **26.3.0.15**.
## Behavior
Each authored Accelerated Page requests one extra tick per server-side ticking
block entity in that Age, in addition to its normal tick. One Page targets 2x
normal ticker calls; four target 5x, before budgets. Identical Pages stack.
The normal registered bound ticker is reused, including optional modded machines.
Its own state validity, chunk-ticking and NeoForge error-handling rules remain active.
Entities, scheduled block ticks and the global clock are not accelerated.
Extra work respects removed/rebound tickers, loaded chunks, normal block-ticking
range, world border and server tick freeze. No new chunk tickets are created.
Machines in normally ticking force-loaded Age chunks may run without a nearby
player. Random-block acceleration remains restricted to its existing player-nearby
chunks. The Overworld and Ages without Accelerated Pages receive no extra work.
The existing authored-plan/page-count save data is unchanged. Existing Accelerated
Ages gain machine ticking after installing this build; no new Age is required.
Page counts persist through the existing save/reload/reseed paths. Work cursors
are transient and reset on reload; they do not alter the saved Page count.
## Independent budgets
Both targets expose these settings under accelerated_age:
| Setting | Default | Meaning |
| --- | --- | --- |
| max_machine_candidates_per_tick | 1024 | Maximum registered ticker entries inspected in one Age tick, rotating across the list. |
| max_machine_ticks_per_tick | 4096 | Maximum extra machine ticker calls per Age tick. Zero disables only machines. |
| max_random_tick_attempts_per_tick | 16384 | Existing random block samples; zero disables only random-block acceleration. |
| max_loaded_chunks | 8 | Existing random-block chunk cap; not a machine chunk-loading setting. |
The selected machine window is snapshotted before executing tickers. Work is
distributed round-robin across eligible machines rather than letting the first
machine use every repeated Page. Removal/replacement/freeze is checked again
before each call. Large installations can hit the cap and run slower than the
requested multiplier. Both machine settings have an upper bound of 65536.
## Compatibility limits
Additional ticker calls also consume normal fuel, energy, inputs and durability.
The effect does not synthesize power or bypass recipe requirements. Machines that
compare global game time, use once-per-tick guards, async jobs or scheduled work
may remain limited to their ordinary rate. Transport may bottleneck independently.
No optional machine mod is required or hard-coded.
## Validation and manual checks
Run sequentially:
python3 scripts/regression_machine112.py mc1211
python3 scripts/regression_machine112.py mc263
These use new disposable worlds and remove them on pass or fail. A 60-second
stall stops the run and hands remaining checks to the player, without a retry.
Fixtures remain outside shipping artifacts.
In-game checks, on each target:
1. In Survival, compare fuelled furnaces in Ages with zero, one and four Pages.
Smelting should be about 10, 5 and 2 seconds per ordinary 200-tick recipe,
before machine budgets; use the same input/fuel and avoid other speed effects.
2. Repeat with a ticking modded machine (Create on 1.21.1 if available). Record
the machine type if it does not accelerate; game-time guards may be a limit.
3. Remove and replace a working machine, leave its chunks, return, then restart.
No ghost ticking or force-loading should occur. Existing Page counts remain.
4. Confirm an ordinary Overworld furnace stays at its normal rate and random-tick
crops still accelerate in an Accelerated Age.
5. With two players in the same Age, confirm the effect is not doubled by player
count. Test tick freeze; machines must stop along with the normal world.
Automated probes test registered vanilla furnace recipes, zero/one/repeated Pages,
freeze, removal/replacement, nonloaded/border guards, a custom registered ticker,
Overworld event exclusion, bounded fair work, and saved Page-count decoding.
They do not constitute a real modded-machine or two-client/restart pass.
### Automated result
Sequential probes passed: **25 checks on 1.21.1**, **26 checks on 26.3**. The freeze check initially exposed a cached runsNormally value immediately after setFrozen; the adapter now checks both the current freeze flag and the tick manager's run state. No test stalled. Both temporary worlds were removed. Player in-game checks subsequently passed; no independent real-mod-machine or two-client/restart result is claimed.
+74
View File
@@ -0,0 +1,74 @@
# Block-instability baselines
Issue #115 uses version-specific bundled averages when an individual Overworld
material has a measured count of zero. It does not create a reference dimension.
## Defaults
The defaults are averaged from the developer's existing Overworld profiles:
987 chunks for Minecraft 1.21.1 and 1,024 chunks for Minecraft 26.3.
Every sampled chunk contributes to the denominator, including chunks without
that block. Ore and deepslate variants retain separate averages.
The bundled JSON files are:
- 1.21.1: src/main/resources/assets/rivencraft/instability/default_block_baseline.json
- 26.3: ports/mc263/src/main/resources/assets/rivencraft/instability/default_block_baseline.json
Each records the source sample count and source-export SHA-256. They contain no
save path, chunk coordinates, or dependency on the developer's save. Rebuild an
average with scripts/generate_block_baseline.py SOURCE_JSON VERSION OUTPUT_JSON.
## Scoring
For each watched block:
1. A positive measured Overworld count uses its actual count divided by sampled chunks.
2. A modded ore with zero measured occurrences uses the combined measured
iron_ore and deepslate_iron_ore average. If that combined count is zero,
it uses the combined bundled iron average. This takes precedence over any
modded ore entry in the bundled JSON.
3. Other zero-count materials use their own bundled average, or zero if absent.
The combined bundled iron average is 89.96656534954407 blocks/chunk for 1.21.1
and 76.5361328125 blocks/chunk for 26.3. Installed mods remain optional:
unused resource IDs do not register blocks or make a mod required.
The existing readiness requirements remain: at least 401 sampled Age chunks
and 500 sampled Overworld chunks. The 5% excess allowance, material weights,
difficulty multipliers and score cap are unchanged. The baseline is selected
per material, so one absent ore cannot replace all measured Overworld results.
Storage blocks and fluids do not receive the modded-ore iron substitution.
They use their own measured/default counts.
Existing saved profiles remain usable. The abandoned reference-source flag is
no longer written or read. Existing Ages use this calculation without rebinding.
## Validation
The former reference-world generation probes stalled on both versions and were
stopped without retry. That implementation and its temporary fixture were
removed. Their logs remain under build/verification/baseline115/.
The replacement is checked without starting a server: regenerate/compare the
bundled averages and exercise the production shared Java policy with each
version's actual resource. Production builds check both Minecraft integrations.
Player checks:
- In a void/skyblock or flat Overworld, enter an Age with valuable generated
blocks. Allow the Age and Overworld profiles to reach the existing thresholds.
Confirm excess materials produce block instability rather than every missing
Overworld ore being compared against zero.
- A modded ore absent from the Overworld should use iron, even if the modded
ore has an entry in the bundled defaults.
- In a normal Overworld, measured positive counts should still take priority.
- Reopen the save and verify profiles and scores persist normally.
Inside the Age, use /rivencraft age instability on 1.21.1 or
/rivencraft_port instability status on 26.3 to inspect readiness and scoring;
instability scores remain absent from the book GUI. The player reported in-game acceptance on both versions before publication.
This is manual acceptance, not a passing result from the abandoned reference-world probes.
Offline verification: both source-average audits passed; the production shared
Java policy passed 14 checks per version. Both production builds passed.
+126
View File
@@ -0,0 +1,126 @@
# Open-book drops — issue #114
Initial conversion slice: local **1.21.1.0.9**, followed by the interaction candidates below.
The existing 26.3 conversion was already player-validated and remains unchanged
in **26.3.0.12**. That initial conversion slice changed only 1.21.1; the interaction follow-up below changes both versions.
## Central conversion
Both targets listen for ordinary ItemEntity joins on the server at LOWEST
priority. Linking and Descriptive Books convert regardless of binding state.
Pages, Notebooks, folders, portfolios and sealed collections retain ordinary
item behavior. Existing DroppedBookEntity instances are skipped.
A queued server task waits until the original join callback completes. It
rechecks the level identity, entity registration/removal and item type, then
creates a custom book. The source is discarded only if addFreshEntity accepts
the replacement. A canceled source cannot create a ghost replacement; canceled
replacement spawning leaves the source intact. Another join/reload may retry.
There is no per-tick scan, force-loading, or conversion recursion.
This covers Q/Ctrl-Q and inventory tosses, container/loot/dispenser-style item
spawns, stand/receptacle contents, full-inventory deliveries, death inventory
drops, and Disarm. Existing ordinary book entities loaded from disk take the
same join path. Books created directly by handheld travel already use the
custom type and retain their existing departure placement.
## State and lifetime
fromDrop uses Minecraft's native restoreFrom NBT path, copies the full stack and
gives the replacement a fresh UUID while the source is still registered. On 1.21.1 the live transform/motion is copied explicitly after NBT loading, which otherwise clamps very high velocities.
Position, facing, velocity, pickup delay, owner/thrower, persistent data and
serializable entity metadata are retained. Arbitrary tosses do not get the
handheld departure constructor's extra height or zero velocity.
The replacement adopts open-book lifetime (unlimited ordinary ItemEntity age)
and the existing slow book decay and rain/water damage. The original ordinary
despawn age is intentionally not retained. Normal book entity save/load retains
its own decay clock thereafter.
The 1.21.1 custom copy() override remains intact so client pickup particles keep
the custom renderer type.
## Validation
Run only the changed target:
python3 scripts/regression_drop114.py mc1211
A disposable server probe exercises native metadata/component copying, bound and
unwritten variants, direct joins and NBT-loaded item joins, actual player toss,
container drop, stand break/full inventory retrieval, station break, Disarm,
inventory death-path drops, already-custom skip, non-book items, cancellation and
retry, custom NBT round trip and pickup-copy entity type. Temporary worlds and
fixture outputs are removed on pass or fail. Stop and hand off a stalled probe.
Player checks on 1.21.1:
- Q/Ctrl-Q and inventory/cursor tosses for Linking/Descriptive Books, bound and
unwritten. Verify open-book orientation, throw motion, pickup delay, name,
destination and modifiers.
- Break a loaded stand; expect one lectern and exactly one open book. Shift-remove
with a full inventory. Repeat with a receptacle and Link Modifier station.
- Test death and Disarm drops and handheld travel. Full-inventory Return Link
delivery can be checked if automatic Return Links are explicitly enabled.
- Leave/reload the chunk and restart with books on the ground. Pick them up and
verify no client pickup-particle crash. Pages/collections stay ordinary items.
An NBT round trip is not a full server restart, and FakePlayer drop paths do not
replace real GUI, death, portal or client rendering acceptance.
## Automated result
The isolated 1.21.1 server probe passed **62 checks**. High-velocity native NBT loading was accounted for by explicitly retaining live transform/motion. No test stalled; the disposable world and fixture build outputs were deleted. The production build passes. No new 26.3 code or tests were run for this root-only change. Player acceptance was subsequently reported; included in 1.21.1.0.12.
## Legacy-style interaction follow-up
Candidates **1.21.1.0.10** and **26.3.0.13** add:
- Walking over a custom dropped book never picks it up.
- Right-click opens the existing Linking/Descriptive Book GUI. Clicking its
destination panel uses normal linking rules and leaves the same book in place.
- Shift-right-click explicitly attempts normal item pickup, respecting ownership,
pickup delay, inventory capacity and NeoForge pickup hooks. A full inventory
leaves the book in the world. This uses inventory pickup rather than Legacy's
special empty-hand assignment.
- Ground books use the lectern-style view: rotating destination previews are
supported, **interactive exploration is disabled**. GUI input closes if the
entity disappears or moves out of reach.
- Unwritten dropped books retain their naming/binding screens. The server
resolves and updates the actual nearby entity, never a held item.
- ID + UUID + expected full stack + alive/range checks reject stale, replaced,
picked-up or remote source actions. Normal travel rules and 26.3 catalog
acknowledgement remain in effect. Preview leases on 26.3 also validate the
dropped entity and distinguish it from hands/stands.
Reviewed local Legacy EntityLinkbook interaction bytecode. Legacy uses sneaking
and an empty hand for direct hand pickup, otherwise opens BOOK_ENTITY. Rivencraft
keeps the requested gesture but uses existing pickup permissions/hooks.
New dropped-action/open-screen payloads reuse existing screens and binding/travel
helpers. 1.21.1 networking protocol is 20; 26.3 book-network registration is 2.
The 26.3 preview request adds an optional dropped-entity UUID with a default
constructor for existing hand/stand callers. Update clients and servers together.
Sequential isolated interaction probes exercise no-contact pickup, normal open,
explicit pickup with owner/delay/full-inventory checks, stale/forged/remote source
rejection, actual binding helper dispatch, teleport request dispatch without
duplicating/removing the ground book, and packet codec round trips. The fixture
captures teleport requests because FakePlayer's connection intentionally ignores
teleports. It does not establish real client rendering or dimension transition.
The 26.3 fixture acknowledges its catalog explicitly; there is no production bypass.
Player checks on both versions:
1. Drop bound Linking and Descriptive Books. Walk over each: no pickup.
2. Right-click: normal book GUI, rotating preview when its optional renderer is
available, and no usable interactive-view control. Click the destination panel
to travel. Return and verify the same book remains.
3. Shift-right-click to collect, including a full inventory and a fresh toss's
pickup delay. Confirm contents/title/modifiers remain intact.
4. Try unwritten books, input names/bind and reopen. Check changing/removing the
ground book while its GUI is open closes it safely.
5. Check reload/restart, ordinary held/stand book GUIs and optional live previews.
### Follow-up automated result
Interaction probes passed sequentially: **21 checks on 1.21.1, 23 on 26.3**. No stall occurred; temporary worlds were deleted. Dedicated-server startup caught and corrected eager client-screen class loading on 1.21.1; client handling now uses a deferred lambda. The fixtures capture teleport dispatch and simulate the 26.3 catalog acknowledgement; the player subsequently reported in-game acceptance on both versions. No independent automated full client/restart matrix is claimed. Logs remain in build/verification/drop_interactions.
The complete interaction update is included in releases **1.21.1.0.12** and **26.3.0.15**.
+7
View File
@@ -19,3 +19,10 @@ former behavior or loader version.
audits are intentionally retained for source attribution and migration
history; these three entry points replace the former practice of treating
every dated task note as a current installation guide.
## Current cross-version updates
- [Accelerated machines](ACCELERATED_MACHINES.md)
- [Writing Desk display model](WRITING_DESK_MODEL.md)
- [Dropped books](DROPPED_BOOKS.md)
- [Void/skyblock block baselines](BLOCK_BASELINE.md)
+54
View File
@@ -0,0 +1,54 @@
# Writing Desk display model — issue #113
Introduced in candidates **1.21.1.0.8** and **26.3.0.12**.
Player in-game checks passed (2026-11-18); included in **1.21.1.0.12** and **26.3.0.15**.
## References and placement
Reviewed dev images/current 26.3.png and the local Legacy ModelWritingDesk /
RenderWritingDesk bytecode. Legacy uses dedicated paper geometry with small yaw
offsets and separate selected-item rendering. Reuse the imported Legacy desk
atlas, including its white paper patch; no new artwork or texture edits.
The existing two block JSON models put the tabletop at Y=12/16. Their red mat
crosses the controller/extension seam. Earlier attachments used unrelated heights:
paper at 0.27 (below the middle shelf at 0.375), selected page at 0.572 (below the tabletop at 0.75).
WritingDeskDisplayModel is shared by both clients:
- Selected page: desk-local (0.5, 0.752, 0), scale 0.7, centered on the seam/mat.
The renderer origin is the controller center (0.5, 0, 0.5).
- Paper stack: local center X=1.15, Z=-0.1, sitting on the right middle shelf at 0.376.
Its sheets are 0.5 x 0.625 blocks, 1/32 high with alternating +/-2-degree yaw.
This is the extension/right half, shifted toward the desk's front.
- One visible layer for 1–16 paper, two for 17–32, three for 33–48 and four for
49–64. Zero paper emits no mesh. Each layer has closed, shaded white geometry.
- Both attachments share the desk's existing facing transform. Geometry stays
within their respective supporting surfaces and does not overlap the selected page.
- The 26.3 renderer captures count/page during extraction and submits immutable
precomputed paper geometry; it does not read the block entity during submission.
The desk's base geometry, non-occluding registration, GUI, storage/copy costs and
item textures are unchanged. Dynamic paper is model geometry rather than repeated
vanilla Paper item sprites. Symbol Pages retain their item renderer so changing
the selected symbol displays its actual artwork.
## Validation
Both Gradle builds pass. A disposable Java geometry audit checked 126 faces over
counts 0, 1, 15, 16, 17, 32, 33, 48, 49 and 64, including outward winding, support-surface
clearance/bounds, separation from the selected page and four facing transforms.
No game server was started for this client-model change.
This does not replace GPU screenshots.
Player checks on both targets:
1. Place four desks, one per horizontal facing. Compare with neighboring solid
blocks and inspect from above, below and each side.
2. Check neither attachment, page only, paper only and both.
3. Try 1, 16, 17, 32, 33, 48, 49 and 64 paper; verify visible height transitions
and no clipping/floating geometry. Paper should be white and on the right shelf,
slightly forward; the selected page should sit over the red mat.
4. Change/remove selected pages and paper, leave/reload the chunk and restart.
Displays must update; neighboring block faces must remain visible correctly.
5. Provide comparison screenshots of all facings for final visual acceptance.
Player acceptance was reported after testing these candidates. The checklist above records the suggested visual matrix; no automated GPU screenshot or independent full restart matrix is claimed.