SpiralGenesis

Upgrading from 1.0.0-alpha.1. data.yml gains an installed-at key, a centre on each spiral plot's record as it is written (a point set with /sgen setspawn has none, and a record without one reads as centre 0), and an optional per-record placement-owed. Once a spiral centre has been placed it also gains a centres table and an active-centre key; until then neither is written. A file written by 1.0.0-alpha.1 loads as it is, with all of its spiral plots on centre 0 and installed-at taken from its earliest assigned-date, or the current time if it has none. Going back to 1.0.0-alpha.1 still loads the file, but a plot that version records writes no centre key and reads back as centre 0, so one it allocates on another centre is not protected against later centres. A server whose origin.world does not name a loaded world stops allocating until the name is corrected, and the owner of a spawn claim created under ADMIN_CLAIM before this release cannot /trust anyone on it until /sgen protect repairs it. The entries below have the detail.

Added

• /sgen release-all confirm releases the spawn claims, for uninstalling. Under the default protection.claim-as: ADMIN_CLAIM players cannot abandon their spawn claims, so removing the plugin meant a /sgen reassign <player> release for each of them, which also moved them. The new command releases the claim around every stored player's current plot, online or not, through the same check reassign ... release uses, a few per tick, and reports how many were released, not ours, unclaimed or unreachable, listing each claim left standing in the console. It keeps every spawn record, is refused while protection is inactive (including on Folia) and under PLAYER_CLAIM, stops if a reload switches either mid-run, and without confirm only says what it would do. The admin guide has a new "Uninstalling" section covering it, the claims it cannot reach, and why respawn points are left as they are.

Changed

• origin.world is matched exactly, and no other world is ever substituted for it. Previously a name that matched no loaded world fell back to whichever world the server loaded first, silently, so a typo carved spiral plots into a lobby or the Nether and overwrote the respawn points of everyone it allocated. The plugin now binds nothing, reports the configured name and the loaded worlds at SEVERE, and tells /sgen reload what it bound. On upgrade, a server whose origin.world does not name a loaded world stops allocating entirely rather than allocating into the wrong one; correct the name and run /sgen reload. A world created after the plugin enables is still picked up on the first join that needs it.
• The Hangar resource page is synced on release. After the version upload succeeds, the release workflow writes docs/modrinth-description.md, without its generated comment, to the Hangar resource page, then reads the page back and fails the job unless Hangar holds exactly that text. The version stays published when only the sync fails; RELEASE_PROCESS.md has the command to run the sync on its own. The Hangar API key now needs edit_page in addition to create_version.
• Hangar versions declare their optional dependencies. Floodgate, GriefPrevention and AuthMeReloaded are listed as optional Paper dependencies, matching softdepend and the Modrinth upload. AuthMeReloaded is not on Hangar and is declared as an external link.
• Registry tokens are scoped to the steps that use them. MODRINTH_TOKEN and HANGAR_API_TOKEN are no longer in the release job's environment, so the build and the tests run without either.

Fixed

• A player can share their own spawn plot again. Under the default protection.claim-as: ADMIN_CLAIM the spawn claim granted the player ClaimPermission.Build and nothing else, and Build does not imply Manage in GriefPrevention's permission model - the two are separate grants and isGrantedBy connects neither to the other. /trust checks for Manage, so the player could build on the plot the plugin had made for them but could not invite anybody onto it, and because administrative claims are administered through griefprevention.adminclaims, no ordinary player could work around it. The claim now carries both grants. Resizing, subdividing and deleting still belong to the server, because GriefPrevention refuses to delegate ClaimPermission.Edit on an administrative claim at all; PLAYER_CLAIM remains the setting for a plot the player owns outright. No default changed and no configuration key was added or renamed. Claims created before this fix carry Build alone and are repaired in place by /sgen protect, which under ADMIN_CLAIM now adds the missing grant to a claim it still recognises as one of its own and says so in the skip reason.
• Allocation no longer places a player outside the world border. Candidate scoring checked terrain only, so once the spiral grew past the border a player was teleported outside it and had their respawn point forced onto the same spot, leaving them taking border damage with every respawn returning them to it. Candidates outside the border are now rejected before their chunk is even requested, so a cell lying wholly outside is skipped without generating terrain nobody may stand on, and the spiral advances. A scan that spends max-scan-attempts without reaching inside the border fails with a console message naming the border instead of placing the player: they stay where they are, and the operator is told to widen the border or move origin.x/origin.z. Allocations after that are refused on the spot, without claiming a spiral index, so a player rejoining cannot walk the spiral outward a scan at a time with no plot to show for it. The refusal is held against the border's position and size rather than as a latch, so widening or moving the border resumes allocation with nothing for an operator to reset. The console is told once, in plain text and without a stack trace: by the scan that gives up, not by the refusals that follow or by other scans in flight that give up against the same border. It is told again when a changed border is exhausted again, and when the border is put back where a scan already gave up, by the first join refused on its return. /sgen reassign that finds no plot tells the operator why in chat. /sgen simulate is outside all of this in both directions: it scans from the origin rather than from where the live spiral has reached, so it still runs and reports after a live allocation has given up, and a run of its own can never refuse a joining player.
• A broken bed on Folia no longer costs a player their plot. Sleeping in a bed replaces the respawn point allocation set, and on Paper a later respawn with that bed gone is sent back to the plot by PlayerRespawnEvent. Folia never fires that event for a death respawn, so the player landed at world spawn instead. When the server finds the bed or anchor gone during the respawn and clears the respawn point (PlayerSetSpawnEvent with cause PLAYER_RESPAWN), the plot is now stored in its place and, on Folia, the player is moved there as soon as they are placed, after the plot passes a fresh safety check; with doImmediateRespawn as well. The respawn itself still lands at world spawn for that one moment. When the point the server rejected is the plot itself, it is restored as the respawn point all the same; a flooded or otherwise unsafe plot fails the safety check, so the player stays at world spawn while the existing repair runs. A player left with no respawn point at all has the plot restored when they die. A bed or anchor that still works, and a point forced elsewhere such as by /spawnpoint, is left alone on both platforms.
• Building on your own spawn point no longer moves it. The death-time re-check failed a plot whenever the block at the player's feet or head could not be walked through, so a chest, crafting table, bed, door or slab placed on the landing spot made the plot "unsafe", and the repair rewrote the stored spawn to another point in the cell, away from what had been built. The re-check now fails a plot only for what can hurt a player: a missing floor, or water, lava, powder snow, underwater plants, cactus, magma or a campfire at the feet, head or underfoot. An obstruction is not one of them. Instead the player respawns at the first clear position above the plot, on top of the build: on Paper that position is the respawn location itself, since Paper 1.21.11 and later place a respawning player exactly where they are sent, inside any block there; on Folia, which declines such a point and places the player at world spawn, the player is moved there once they are placed. The stored spawn is not changed by the lift. If the column has no clear position below the build limit, or the first one sits on something that hurts, the player is held at world spawn. On Paper the client still shows the vanilla "no respawn block" message on such a death; the placement is unaffected. Ice no longer fails the re-check either, so an ice floor or ice road through the spawn is kept; allocation still rejects ice as a surface for new plots.
• A save that keeps failing no longer floods the console. The background flush retries data.yml every five seconds, and each failure logged a SEVERE stack trace, so a full disk or read-only mount filled the console for as long as it lasted. The first failure is still logged in full; the retries after it are logged at FINE, and one line is logged when a save succeeds again, with the number of failed attempts. Removing a leftover data.yml.tmp at startup and on /sgen reload now also waits for a save in progress instead of deleting the scratch file out from under it.
• Repairing a plot no longer takes away a working bed or anchor. When the death-time re-check found a plot unsafe and the repair moved it within the cell, the player's respawn point was forced onto the repaired point unconditionally, so a player with a bed, a charged anchor or a point set elsewhere by /spawnpoint lost it the next time their plot was repaired, and a player who had already respawned at their bed was teleported to the plot. The repaired point is still recorded as the plot, but the respawn point now follows it only when it is unset or is the old plot, matched on its block column in the plot's world so a player lifted on top of a build still counts. Anything else is left alone, and so is the player. Whether the bed still works is not checked: one that has stopped working is handled when the server clears the point on respawn.
• A plot left outside a shrunken world border no longer passes its re-check. The re-check on death and respawn looked at terrain only, so after an operator drew the border in past an existing plot, its owner kept respawning outside the border and taking border damage. A stored point outside the border now fails the re-check, even when its chunk is not loaded. The in-cell repair then looks for a point inside the border in the same cell. When the whole cell is outside, it finds none and the player respawns at the main world's spawn, which is the plot world's only while the plot world is the main world; Folia always uses the overworld. On Folia, which accepts a forced respawn point without looking at the border, the plot is taken off the player's respawn point at death, before the respawn can use it; a bed, anchor or point set elsewhere is left alone. No new spiral index is claimed and the stored plot is not rewritten. Once the border takes the plot back in, the player's next death restores it as their respawn point and re-checks it. Separately, on Folia a player whose repair finds no safe point while they are still on the death screen now respawns at the overworld's spawn instead of back on the unsafe plot.
• The respawn listener no longer keeps an entry for every player who has left. Each respawn that fired PlayerRespawnEvent without its point failing left the player in a set that only their next death cleared, so one entry stayed behind per player who quit before dying again. The entry is now dropped on quit.
• A step, slab or trapdoor under your spawn point no longer moves it. The death-time re-check treated anything a player can pass through as a missing floor, so a staircase dug down from the spawn point or an open trapdoor over it failed, and the repair moved the plot away from the build. The floor is now whatever has a collision shape under the centre of the column: slabs, stairs, closed trapdoors, carpet, two or more layers of snow and a closed fence gate all count, and air, fluids, plants, a single snow layer, a door, an open trapdoor or an open fence gate do not. When the block under the spawn is not a floor, a step of one block down onto a floor is accepted too. Anything deeper still fails, so a spawn over a real drop is still repaired. The floor that is used, and the block stepped into, get the same checks as before for water, lava, magma, cactus and campfires. Allocation of new plots is unchanged.
• A player who cannot be allocated because no world is bound stays held until one is. The log said such a player would be retried once origin.world named a loaded world, but the hold lasted one action: their next step retried while the world was still missing, and afterwards nothing retried them for the rest of the session. They now stay held across every action and are allocated as soon as a world is bound, by /sgen reload or by a world that loads late, without having to act again. The hold is logged once per player at WARNING instead of at SEVERE on every retry, and the missing world is still reported at SEVERE once per configured name, now also when two threads retry the bind together.
• An unreadable data.yml no longer restarts the spiral from index 0. A file that did not parse loaded as an empty one, indistinguishable from a fresh install, so the counter reset and every returning player was allocated a new plot in cells other players already held; the next save then wrote that empty state over the original. Such a file is now reported once at SEVERE and copied aside as data.yml.broken-<timestamp>, and storage is marked failed: nothing is saved by the flush, /sgen reload or shutdown, nobody is allocated, and no respawn point is changed. Joining players wait in the same hold as when no world is bound, operators with spiralgenesis.admin are told in chat when they join, and the commands that read or write player records are refused. A /sgen reload that reads the repaired file clears the state and allocates everyone held; one that still cannot read it reports again without copying the same file twice. A missing or empty file is still a fresh install.
• /sgen simulate no longer discards its report when a sample fails. One sample that found no plot inside the world border, or failed for any other reason, aborted the whole run with "Simulation failed" and none of the samples already taken. A sample that exhausts against the border is now counted and the run carries on: the report shows how many samples exhausted, the first one that did and the spiral index it scanned from, and the console summary line gains exhausted=. Any other failure ends the run at that sample and the report for the samples before it is still delivered, with the failure reported alongside it and logged in full.
• A respawn lifted above an open trapdoor or a door checks the drop beneath it. When a build over the spawn point lifts the player onto a block that leaves the column centre clear, the player falls through it onto the build below, and only the cells at the lifted position were checked. The fall is now followed to the first floor, and the cells passed and that floor get the same checks for water, lava, magma, cactus and campfires. A drop onto or through one of them, one that falls past the step below the spawn point, or one longer than 3 blocks, the most a fall takes without damage, holds the player at world spawn as a hazard on top of the build does.
• A held player's bind retry can no longer undo a /sgen reload that bound a world. On Folia the reload and a held player's retry run on different threads, and a retry that had read the old origin.world could clear the manager just after the reload bound the new one. The log then said the world was bound while held players stayed held until they acted again. Binding is now serialised, so the reload's bind stands and releases everyone held.
• A spawn that storage refused to record no longer moves anyone. A /sgen reload that failed to read data.yml could land after a first allocation or /sgen reassign had checked storage and before it wrote the new spawn. The write was refused, but the player was still teleported, their respawn point set and the new plot claimed, and reassign's release removed the old claim, all for a record that did not exist. Everything after the write is now gated on the write itself. A refused first allocation leaves the player where they are, logs one line and holds them until storage reads again, and the line says they left instead if they disconnected before they could be held; a refused reassign changes nothing and tells the operator so. The refused plot's index is recorded against nobody, so it is never shared. /sgen setspawn and the in-cell repair are gated the same way.
• A scan in flight across a failed /sgen reload no longer shares its plot. An allocation reserves its spiral index when it starts and records the player when it finishes. If a reload failed to read data.yml in between, and a later reload read back a file saved before the reservation, the counter was restored below the scan's index: the scan recorded its player there and the next player was handed the same plot. A reservation made between a reload's save and its load could be handed out twice the same way. A load now never restores the counter below an index already reserved or recorded in this run, except one whose write was refused, which is handed out again.
• A player who disconnects during their terrain scan keeps the plot it finds. The scan reserves a spiral index when it starts, and a player who left before the result was applied had it dropped: the index was recorded against nobody and they were allocated a second one on their next join, leaving a permanent gap in the spiral. The plot is now recorded against them anyway, and they are placed on it when they return - respawn point, teleport and claim - without another index being reserved. Placement happens when a new player would have been allocated: under the default FIRST_ACTION, on their first uncancelled action after they join, and at once under ON_JOIN and for Bedrock players. The same holds for a player who rejoined while the scan was still running, once it finishes. A write refused because data.yml could not be read records nothing, as for a connected player. The pending placement is saved with the record as an optional placement-owed key in data.yml, so it survives a restart, and is removed once the player is placed; a file written before the key existed loads with nobody owed a placement.
• Installing on a server people already play on no longer moves them. Every player without a record was treated as new, so each one who had played there before the install was allocated a plot on their next visit: their bed or respawn anchor was overwritten and they were teleported away from their base, with the first of them landing on whatever already stood around the spiral origin. A player who played on the server before the plugin was installed is now left alone: no index is reserved, nothing is set, moved or claimed, and they are not gated. The console says so once per player per run, and /sgen reassign gives them a plot when an operator wants them to have one; /sgen allocate refuses them and says the same. "Before" is the server's first-played time against a new installed-at key in data.yml, not whether they have played before, so a player who first joined after the install and left before being placed is still allocated. A fresh install records the current time. A file written by an earlier version records its earliest assigned-date, since that version allocated the first player to join on their first action, or the current time if it assigns nobody. Every record write rewrites its assigned-date, so that can be later than the real install, which leans toward leaving players alone. A player with a record is never skipped, so one owed a placement is still placed.
• A respawn point forced elsewhere is kept on Paper 1.21 and later. The respawn handler left a respawn alone only when Paper flagged it as a bed or anchor spawn. Paper 1.21.11 and later flag neither for a forced point, so a working point set by /spawnpoint, EssentialsX or Multiverse was replaced with the plot for every player who had one, and so was a location another plugin chose for the respawn, such as EssentialsX respawn-at-home. Paper 1.20.4 flags every working point as a bed, the plot included, so the respawn-time re-check of the plot never ran there. The handler now acts only on a respawn headed for the plot's block column, or one whose point failed and fell back to world spawn, and leaves a working bed, anchor or forced point and a location another plugin set alone on every version.
• The pre-install log line warns that /sgen reassign replaces a bed. The line that recommends reassigning a player from before the install now says that it replaces their bed or respawn anchor with the new plot, as /sgen setspawn also does. The admin guide has a new section on installing onto a server people already play on: writing config.yml before the first start, choosing an origin away from existing builds and claims, and what happens to the players already there. The README quick start said the first player gets plot #1; the first plot is #0,0.
• SpiralGenesis loads after Multiverse-Core. Multiverse-Core creates its worlds in its own startup, and nothing ordered the two, so an origin.world that Multiverse loads could still be missing when SpiralGenesis bound it: the log said no spawn would be allocated, and the world was only bound on a later join. Multiverse-Core is now a soft dependency, and an optional dependency on Hangar and Modrinth. The error for a world that is not loaded now says allocation starts once it is, and that a wrong name needs correcting and /sgen reload, rather than that no spawn will be allocated. The README's respawn line and the admin guide's respawn section now name every respawn that outranks the plot, a point forced elsewhere and a location another plugin sets included, and the guide and the respawn re-check no longer say the plugin has no claim or protection system.
• Moving the spiral centre or changing cell-size no longer puts new plots on top of existing ones. A plot's cell was the configured origin plus its index's grid position times the configured cell size, with one running index, so after /sgen setcenter or an edit to origin or cell-size the next indices landed in cells other players already held. Each geometry is now its own spiral centre, recorded in a centres table in data.yml with its own counter, and every record carries the centre its index is on. Plots are named #centre,index in commands and logs, and /sgen info shows the centre's origin and cell size. Before a cell is used it is tested against the whole cell of every plot on another centre, every point set with /sgen setspawn, and every cell another scan is still searching; one that overlaps is skipped and logged, and does not count toward max-scan-attempts. Returning to an earlier geometry resumes its centre's counter. A file written by an earlier version loads with all of its plots on centre 0, recorded at the origin and cell-size configured when it is first loaded; a centre moved under an earlier alpha is not detected. current-spiral-index remains the active centre's counter, so an earlier version still loads the file. A plot that earlier version records writes no centre key and reads back as centre 0, so one it allocates on another centre is not protected against later centres. With an even cell-size the in-cell search no longer reaches the first column of the neighbouring cell, so every candidate stays inside its own cell.
• Repairing a plot searches the plot's own cell. The in-cell repair rebuilt the cell from the configured origin and cell-size, so after /sgen setcenter or a reload with a new origin or cell size it searched the cell at the same index of the new spiral, often another player's, and a point set with /sgen setspawn had the origin cell searched for it. A repair now searches the record's index on its own centre, at the origin and cell size recorded for that centre, and reserves nothing. A setspawn point that fails its re-check is not searched around: it is kept, and the player is sent to world spawn with a warning. A scan also keeps the centre its first cell was on for every later cell, so a setcenter during a scan applies from the next scan, and a scan that gives up against the world border records the spiral it walked rather than the one configured when it gave up, so the moved spiral is still scanned on the next join.
• Allocation no longer places new players inside existing claims. Candidates were judged on terrain alone, so on a populated world a new player could be placed, and their respawn point forced, inside somebody's base; the spawn claim was then refused and they lived there without build rights. Wherever GriefPrevention is installed, whether or not protection.enabled is on, a candidate whose protection.size square would overlap any existing claim is now rejected as CLAIMED, before its chunk is loaded: players' own claims, administrative claims and spawn claims left behind by /sgen reassign without release alike. A cell whose every candidate is claimed is skipped and logged, and does not count toward max-scan-attempts. Repairs and /sgen setspawn do not check, and there is no check without GriefPrevention or on Folia. Unclaimed builds are not detected. A spawn claim refused for an overlap on first allocation or /sgen reassign, which now means the claim appeared after the plot was chosen, is logged at warning; the player still keeps the plot.