Failed to load saved sub-level: Plot already exists at 0, 36 #1626

Activity

1. wannholtemil-png

   added

   type: crashSomething crashes my gameSomething crashes my game

   on Oct 12, 2026
2. anenigmabeyondwords commented on Oct 13, 2026

   

   anenigmabeyondwords

   on Oct 13, 2026

   More actions

   We hit the same Plot already exists on Sable 2.0.6 (NeoForge 21.1.248, MC 1.21.1, Create Aeronautics 1.3.2) and traced one cause that produces it. (We run one local bytecode patch, the #1627 split fix in SubLevelHeatMapManager; every storage and plot class is stock.)

   What we see

   • 688 saved overworld sub-levels sit in 654 plots; 10 plots are claimed by 2 to 10 saved records each. No UUID appears twice.
   • Each of those 10 plots contains exactly one sub-level built only from blocks of a mod we later removed (Domum Ornamentum), and every such sub-level is in one of those plots.
   • Every boot logs Recoverable errors when loading plot section ... Unknown registry key ... using default, then Failed to load sub-level, invalid plot bounds: EMPTY for those records. IllegalArgumentException: Plot already exists at x, z follows whenever two of the other claimants load together, and one of them never loads.

   Cause

   1. When a record loads with no non-air blocks, fullyLoad calls removeSubLevel(subLevel, REMOVED) (

      sable/common/src/main/java/dev/ryanhcode/sable/sublevel/storage/serialization/SubLevelSerializer.java

      Lines 166 to 170 in ad8b4d3

      	if (plot.getBoundingBox() == BoundingBox3i.EMPTY || plot.getBoundingBox().volume() <= 0) {
      	Sable.LOGGER.error("Failed to load sub-level, invalid plot bounds: {}", plot.getBoundingBox() == BoundingBox3i.EMPTY ? "EMPTY" : plot.getBoundingBox());
      	plotContainer.removeSubLevel(subLevel, SubLevelRemovalReason.REMOVED);
      	return null;
      	}

      ).
   2. REMOVED clears the plot's occupancy bit and marks it dirty (

      sable/common/src/main/java/dev/ryanhcode/sable/api/sublevel/SubLevelContainer.java

      Lines 495 to 497 in ad8b4d3

      	if (reason == SubLevelRemovalReason.REMOVED) {
      	this.getOccupancy().clear(index);
      	}

      ). The record itself is never deleted: lastSerializationPointer is still null at that point (it is only set after fullyLoad succeeds,

      sable/common/src/main/java/dev/ryanhcode/sable/sublevel/storage/holding/SubLevelHoldingChunkMap.java

      Lines 564 to 578 in ad8b4d3

      	public void loadHoldingSubLevel(final HoldingSubLevel holdingSubLevel) {
      	final ServerSubLevel subLevel = SubLevelSerializer.fullyLoad(this.level, holdingSubLevel.data());
      	if (subLevel != null) {
      	subLevel.setLastSerializationPointer(holdingSubLevel.pointer());
      	} else {
      	final MinecraftServer server = this.level.getServer();
      	if (server instanceof final SableToastableServer toastable) {
      	toastable.sable$reportSubLevelLoadFailure(holdingSubLevel.pointer());
      	}
      	Sable.LOGGER.info("Failed to load holding sub-level {} with pointer {}. This is a problem.", holdingSubLevel, holdingSubLevel.pointer());
      	}
      	this.allHoldingSubLevels.remove(holdingSubLevel.data().uuid());
      	}

      ), so queueDeletion does nothing.
   3. The record keeps its plot on disk while the occupancy file says the plot is free, and the next allocateNewSubLevel (first free plot,

      sable/common/src/main/java/dev/ryanhcode/sable/api/sublevel/SubLevelContainer.java

      Lines 180 to 189 in ad8b4d3

      	private Vector2i getFirstEmptyPlot() {
      	for (int x = 0; x < (1 << this.logSideLength); x++) {
      	for (int z = 0; z < (1 << this.logSideLength); z++) {
      	if (!this.occupancy.get(this.getIndex(x, z))) {
      	return new Vector2i(x, z);
      	}
      	}
      	}
      	return null;
      	}

      ) hands it out.
   4. On every later restart the record loads again. allocateSubLevel only rejects a plot that has a loaded sub-level in it and never checks occupancy (

      sable/common/src/main/java/dev/ryanhcode/sable/api/sublevel/SubLevelContainer.java

      Lines 247 to 250 in ad8b4d3

      	public SubLevel allocateSubLevel(final UUID uuid, final int x, final int z, final Pose3d pose) {
      	if (this.getLocalPlot(x, z) != null) {
      	throw new IllegalArgumentException("Plot already exists at " + x + ", " + z);
      	}

      ), so the record takes the plot from an unloaded owner, fails again, and frees it again.
   5. Each cycle adds one more claimant; our worst plot has 10.

   Repro

   1. Assemble a one-block sub-level from a mod block near spawn, stop, remove the mod, start. The log shows invalid plot bounds: EMPTY.
   2. Assemble a vanilla sub-level C: it gets the same plot. Let C unload and restart.
   3. Assemble D: it also gets that plot. Load C and D together: Plot already exists.

   Possible fix

   • On a failed load, release the sub-level without clearing occupancy (or also delete the record through the holding pointer), so the disk and the occupancy data agree.
   • Longer term, store the owning UUID per plot so a load can tell its own plot from another sub-level's and move the newcomer to a free plot instead of refusing to load.

   Unchanged on main as of d501fb3. For the original report: it may be worth checking the log for invalid plot bounds: EMPTY earlier in the same boot.