Introduces a new workflow that runs when a new build has been published
on GitHub Releases. It converts the release notes to Hangar Markdown and
sends it to Hangar along with the jar-file.
Note: The workflow currently relies on the version string being appended
to the filename of the jar-file. Without it, the file reference in the
`curl` request that uploads the build would need to change.
The workflow references a new secret, `HANGAR_TOKEN`, which is just an
API key for the Hangar API. The token was created in the Hangar profile
settings (API Keys), and its only permission is `create_version`.
In order to properly upload a new build to Hangar, we need to construct
a somewhat complex JSON object. This is because the Hangar API allows
for publishing releases on multiple platforms and for multiple versions,
which makes the simple use case for MobArena's single file upload look a
bit overcomplicated. Unlike the CurseForge API, the Hangar API supports
"normal" platform version strings, so we don't need to map anything. It
also supports patch version wildcards, so we can get away with `1.18.x`,
`1.19.x`, etc. for each version supported. The API only uses the API key
for authentication, which means we need to grab a JWT and use that for
the actual upload request. Note that the `pluginDependencies` property
is currently required, but it can be left empty.
The workflow can be invoked directly via the `workflow_dispatch` event,
which might come in handy if something in the pipeline breaks.
The Hangar base URL and project slug are both hardcoded, and things
would probably be cleaner if they were made into variables, but we don't
need this workflow anywhere else, so it's fine for now.
Introduces a new workflow that runs when a new build has been published
on GitHub Releases. It converts the release notes to CurseForge HTML and
sends it to CurseForge along with the jar-file.
Note: The workflow currently relies on the version string being appended
to the filename of the jar-file. Without it, the file reference in the
`curl` request that uploads the build would need to change.
The workflow references a new secret, `CURSEFORGE_TOKEN`, which is just
an API key for the CurseForge API. The token was created on CurseForge
under profile settings (My API Tokens).
In order to properly upload a new build to CurseForge, we need a list of
"game version IDs", which isn't completely trivial. The API gives us a
means of looking up _all_ Minecraft game version IDs, but we then have
to manually filter out the ones that don't apply to Bukkit plugins, as
there are duplicate entries for each Minecraft version, and only some of
them work for Bukkit plugins (which turns out to be the ones with game
version type ID 1). The structure of the `metadata` field combined with
how incredibly difficult bash can be to work with has resulted in some
gnarly text processing trying to filter the JSON response and turning it
into a list for use in the `jq` template, but it gets the job done.
The CurseForge base URL and project ID are both hardcoded, and things
would probably be cleaner if they were made into variables, but we don't
need this workflow anywhere else, so it's fine for now.
The workflow can be invoked directly via the `workflow_dispatch` event,
which might come in handy if something in the pipeline breaks.
Lots of inspiration was found in the probably really great GitHub Action
`curseforge-upload` [1]. We could have probably just used that, but it's
nice to have full control of the process. At any rate, thanks to itsmeow
and NotMyFault for publishing their work.
---
[1] https://github.com/itsmeow/curseforge-upload
We don't need the parser exposed outside of the `things` package, and
none of the other thing parsers are public anyway. Coincidentally, this
fixes a warning about exposing InventoryThing outside of its visibility
scope, so yay.
This fixes a warning about exposing ItemStackThingParser outside of its
visibility scope, but really it's just a good little cleanup step, since
the constructor in question is never used for anything. We might want to
eventually expose the ItemStackThingParser and use it in more places in
the code base, but in that case, and in that case it would probably make
sense to re-introduce the constructor, but I'm calling YAGNI on this in
order to nuke a warning.
Okay, the reason the code included the unary plus was to more directly
represent the resulting expression, but I'm guessing the compiler isn't
going to respect that intent even if it could somehow understand it, so
it will probably remove the symbols and just parse it all the same.
Unlike with the unary plus, the unary minus can be "fixed" by wrapping
it in parentheses. The end result is of course the exact same, but the
intent is perhaps a bit clearer this way. We want to try to coerce the
compiler into creating an expression with "add a negative value", just
for the sake of "correctness" at the runtime evaluation level, but even
if that isn't what will actually happen, the explicit code is still a
bit easier to read. While unary plus is easy to disregard, "fixing" an
unnecessary unary minus would mean having to change the binary operator
before it, which muddies the intent of the expression.
This commit releases the BinaryOperation and UnaryOperation interfaces
of the `formula` package from their `java.util.function` supertypes and
redeclares the previously inherited functions directly in the operation
interfaces, but also reifies them by explicitly using primitive doubles
instead of generics and wrapper classes. Doing so does not change the
functionality or any other code at all, but it makes the interfaces much
"stronger", since they no longer need to consider `null` values, which
they didn't actually take into account anyway. This fixes a warning in
Visual Studio Code (not sure how to get the same warning in IntelliJ)
about the operator registrations in the default formula environment
factory method being unsafe.
This commit adds a second job to the build workflow that runs after the
build job has completed. The job creates a GitHub Releases _draft_ that
needs to be manually published in order to be publicly available.
The job runs if, and only if, the following conditions are met:
- The build job has completed _successfully_, i.e. if the Gradle `build`
task fails, the draft job doesn't run.
- The `push` event that triggered the workflow happened on the `master`
branch, i.e. releases will never be created from temporary branches.
- The commit message of the most recent commit in the `push` event that
triggered the workflow starts with `"Release "`, i.e. there must be a
commit that explicitly tries to "release" a build.
- The version string does _not_ end with `-SNAPSHOT`, i.e. development
builds will not be released.
All of these conditions act as safeguards, so we don't end up releasing
something we don't want to release, but they also prevent bloating the
Releases page with a bunch of useless drafts.
The job uses the `version` output variable from the build job that was
introduced in a recent commit to extract release notes using the script
that was also introduced recently, as well as for the name of the _tag_
to create when the release is published.
Note that the `GITHUB_TOKEN` environment variable is required to be set
when we want to use the GitHub CLI in a workflow [1]. The job also has
an explicit `contents: write` permission, which is required for creating
releases from GitHub Actions.
This commit makes the build workflow output the version string as found
in the `version` property in `build.gradle.kts`. The version string will
be necessary further down the pipeline when we need to extract release
notes and create tags.
There are many ways to extract the version string:
- Use `grep` to grab it from `build.gradle.kts` directly. This is pretty
brittle, since we don't really know for sure if the structure of the
file will change in the future.
- Create a Gradle task in `build.gradle.kts` that prints the version
string. This is probably the most direct approach we could take, but
such a hyper-specific task feels like a code smell. It also requires
running Gradle again, which is a bit slow.
- Use the built-in `properties` task in Gradle to print the `version`
property and `grep` it. We avoid changing `build.gradle.kts`, which is
a plus, but we still have to actually run Gradle.
- Parse the filename of the resulting jar-file in `build/libs/`, since
it now contains the version string. This is also brittle, because we
don't know if we're gonna continue to append the version string to the
jar-file, and depending so much on it being there is a little scary.
- Extract `plugin.yml` from the resulting jar-file and `grep` it. This
is perhaps a little crude, but it is much faster than running Gradle,
and as a bonus, we get a bit closer to "what's inside the jar-file",
which should give us a bit more confidence that any given release is
actually the version it claims to be.
It may seem like a small thing to invest so much text on, but from an
automation standpoint, it is much easier to be confident in automations
with predictable and robust mechanisms for deriving input variables.
Introduces a crude Python script that can extract release notes from the
changelog for a given version and convert it to one of three different
output formats:
- `github` for GitHub Releases. This is just the Markdown itself, but
with the very first line (the version) removed, because the version is
also the title of the release itself.
- `spigot` for Spigot Resources. This is the BBCode format used on the
forums and in the resource descriptions.
- `curse` for CurseForge File Uploads. Curse uses a so-called "WYSIWYG"
format that's really just HTML underneath.
The formats for Spigot and CurseForge are straightforward to convert to
as long as we only use simple text formatting, bullet lists, and links,
but that is really all the changelog should consist of anyway.
While this script already makes the release process quite a bit easier
on its own, the end goal is to _automate_ releases as much as possible,
and to do that, we need to be able to extract release notes, and we need
to be able to do it from GitHub Actions, which is quite a bit simpler if
we don't use third-party libraries.
Publishing releases on GitHub is almost trivial, while CurseForge is
pretty easy, and Hangar should be very doable as well. Spigot, on the
other hand, is stuck in the dark ages, so we must continue to upload
files manually there.
According to the documentation, by not specifying anything in the `push`
event, GitHub Actions will run the workflow for both branches and tags,
but by specifying `branches` alone, it will _only_ run the workflow for
branches and _not_ tags [1].
We want to build on pushes to _all_ branches so we can give folks a jar
to try out for feature requests and bug fixes. We don't want to build on
tags, however, because they don't provide any value in the current build
pipeline. If the version was derived from Git tags, it could make sense
to build on tags, but that's not where we are.
As for the `**` glob pattern instead of just `*`, the documentation says
that the latter does not match forward slashes [2], and while our naming
convention for branches doesn't use slashes right now, there's no reason
that they shouldn't be viable later down the road.
---
[1] https://docs.github.com/en/actions/using-workflows/workflow-syntax-for-github-actions#onpushbranchestagsbranches-ignoretags-ignore
[2] https://docs.github.com/en/actions/using-workflows/workflow-syntax-for-github-actions#filter-pattern-cheat-sheet
Removes the "archive version" override for the `shadowJar` task in the
build file, resulting in a jar-file that contains the current version
number in the filename. It's mostly a convenience tradeoff; either we
can see the version directly in the filename and avoid assumptions, or
we can easily overwrite an existing jar-file with a new one.
Also updates the upload step of the build workflow to a glob pattern so
we grab the file regardless of the version.
This commit converts the project from Maven to Gradle by removing the
old Maven-related files such as `pom.xml` and the Maven Wrapper, and
replacing them with their Gradle counterparts (more or less).
Comparing the artifacts from Maven and Gradle indicates no significant
differences in the resulting jar-files, and a quick test of the plugin
shows that things are still working as expected.
Bits of `build.gradle.kts` may need a bit of a tune-up later down the
road, e.g. the test sources "hack" put in place. It may be cleaner to
omit this hack and just suck up having to repeat dependencies, but in
that case, it might be better to embrace the "libs" file instead of
having to repeat dependencies in full.
Note that this commit changes the caching mechanism used in the GitHub
Actions build workflow, opting for the one built into `setup-java`.
This setting allows server owners to allow arena monsters to teleport
around _while inside the region_. They still can't teleport out of the
region.
Taken at face value, this should just be the default behavior. However,
due to arena regions being boxes, any non-box shaped arena will need a
region that covers more than the physical arena structure, which means
mobs like Endermen will be able to teleport into possibly unreachable
areas of the physical structure. So we have to make do with a setting.
Closes#762
The idea behind the previous implementation worked, but it was a tad bit
confusing. This commit refactors the activation logic by simply removing
it entirely. The "activation" part of the logic is now derived from the
wave number (if 0, it means we haven't _really_ started yet), instead of
relying on the spawn thread to toggle the flag on and off. This kind of
dependency inversion (spawn thread -> listener, listener -> "phase") is
a pretty decent (albeit super tiny) step towards cleaning up the whole
session system, so I call that a victory in and of itself!
Using the new `mobarena.admin.errors` permission, this commit provides
server owners with a way to make the infamous Spigot health error much
more visible by sending the error message to any "admins" online when
the error occurs.
Closes#764
Introduces a new permission for "admins" that can be used to increase
visibility of errors caught by the plugin. Server owners may not want
_all_ online players to see these types of messages, so the permission
gives error handlers a way to filter the list of online players before
sending the error message.
Introduces the concept of a _saved item_; an in-game item that has been
captured in a YAML file via Bukkit's item serialization mechanism. These
items can be referenced in the config-file in all places that any other
normal item can be used, assuming the ThingManager is in charge of the
parsing. This should help bridge the gap between class chests and the
config-file by allowing any Bukkit serializable item stack to be stored
and referenced as if MobArena's item syntax directly supported it.
Three new setup commands are introduced to help manage the items, such
that they can be created, deleted, and loaded (for "editing" purposes).
The commands are somewhat rough around the edges and may need a little
bit of polish going forward.
Together with the new inventory referencing Things, this functionality
should help provide most of the flexibility people have been missing
from the item syntax for about a decade... Hell, it's about time.
Closes#212
Adds three new Thing types that can be used to reference items in chests
(or any block-based InventoryHolder):
- InventoryIndexThing looks up an item by index/slot in an inventory.
- InventoryGroupThing groups all non-null/non-air items in an inventory
into a ThingGroup.
- InventoryRangeThing groups all non-null/non-air items in a given range
of an inventory into a ThingGroup.
The new Thing types aim to bridge a gap between the class chests and the
rest of the Thing-based parts of the config-file. The goal is two-fold:
allow for more in-game configuration so access to the config-file isn't
_quite_ as crucial, and propagate the item-wise feature completeness of
class chests to other parts of the plugin.
While class chests are low configuration and a bit "all or nothing", the
inventory Thing types require manually punching in the coords for chests
and possibly indices/ranges for items. This means that the initial setup
could be a bit unwieldy, and highly volatile wave setups are definitely
not a good fit. If the wave setup is mostly pre-defined, it is fairly
easy to tweak upgrade waves and rewards in the same way class chests are
tweaked.
As for item-wise feature completeness, the inventory Thing types share
the same "if Bukkit can copy it, it will work" rule of thumb as class
chests do, which means items with metadata such as custom names, lore,
or even NBTs, should just work. This could remove the need to employ
other plugins.
By no means can this solution be considered "optimal", but it it _does_
enable some long-requested features.
Closes#456
Since Minecraft 1.20, players can edit signs by right-clicking on them,
and that poses a problem for the sign-centric portions of the plugin,
such as class selection signs and the various types of arena signs.
This commit refactors the PlayerInteractEvent handler in ArenaListener
in order to break open the possibility of handling non-lobby players as
well. We're a little more strict with lobby players, and we still want
to handle class sign clicks and iron block clicks here. For players who
aren't in the lobby, we're really just blocking the event according to
the regular protection rules (block is inside region, protect is on, and
arena is not in edit mode).
It also blanket cancels events in the HandlesSignClicks event handler,
because there is no meaningful way to edit an arena sign, since their
contents come from the template file and not from what is written on
them by the sign renderer.
Ideally, we'd refactor some of this event handler logic, because it's a
lot easier to take care of the individual responsibilities in separate
event handlers.
Fixes#765
It boggles the mind that this tiny little class has worked as intended
since 2011 (!!), and all of a sudden, signs no longer retain their text
in the repair procedure...
By calling the somewhat arbitrary `update()` method on the sign after
setting the contents, the sign appears to correctly update again.
Fixes#772
I had no idea this class existed, but it seems like it's actually never
been used for anything, since the commit that introduced it didn't even
use it either.
Introduces a couple of tests for the FormulaManager test suite in order
to cover all the methods the class exposes. This means it is no longer
necessary to suppress the "unused" warnings.
It's not clear why this variation in the auto-ready logic exists, and
the commit history doesn't seem to have any clues either. Perhaps the
actual readying up logic was incompatible with auto-ready at some point,
but at this point in time it doesn't seem like this is necessary at all,
and it appears to be causing a bug with the MobArenaStats extension.
By simply calling the player ready procedure regardless of the status of
the auto start timer, MobArena fires the arena player ready event that
MobArenaStats depends on for some of its pre-session bookkeeping. It
could be argued that MobArenaStats should be more robust, but we would
much rather fix the root problem than slack on the otherwise fairly
sound strictness of the MobArenaStats data model.
Fixes#746
Makes the boss entity the source of the obsidian bomb explosion, which
then makes the damage event listener handle the explosion damage as if
the boss is the damager, which means the `monster-infight` flag should
be respected.
Fixes#759
Makes the sheep entity the source of the explosion that's created when
it triggers close to a player. This, in turn, makes the damage event
listener handle the explosion damage as if the sheep is the damager,
which means it will respect the `monster-infight` flag.
Fixes#758
This new spawn reason was introduced somewhere between 1.18 and 1.18.1,
and unfortunately it is a breaking change, so we have to employ it for
MobArena to properly allow and register vexes (again...), but we also
have to maintain a variation of the old logic so we don't break support
for older server versions.
Fixes#719
This time around we're trying to keep the API version bumped for the
sake of not leaving the people who actually keep their servers updated
behind. It does present some potential backwards compatibility issues
when we do employ new API, but we'll just have to be mindful of _how_ we
go about doing so...
The previous name was really just a way to try to "prettify" the whole
thing, but the badge in the README is a little wonky compared to other
projects, so we're just gonna go with "build" for now for consistency.
When a player with pets dies in the arena, we want their pets to be
removed. One could probably argue that the pets _should_ be able to
stick around, but the original intent was for them to be removed
alongside their owner.
Fixes#721
This new setting allows changing the fuse time for auto-ignited TNT,
which is normally a hardcoded 80 ticks in Minecraft. Note the somewhat
weak safeguarding without any sort of error message - with great power
comes great responsibility...
Closes#715
Now that MobArena sets the TNTPrimed source on auto-ignited TNT, the
"planter" logic becomes _somewhat_ obsolete. This is due to the fact
that manually ignited TNT blocks produce a TNTPrimed entity with the
source property set to the player that ignited the block.
Because the "planter" logic in the BlockIgniteEvent handler didn't
actually work, this change shouldn't actually do anything in that
regard. That is, a TNT block ignited by a player would still have that
player as its source regardless of who the planter was, because the
procedure that would have otherwise set the planter of the TNTPrimed
entity never ran.
Turns out igniting TNT _doesn't_ fire a BlockIgniteEvent but rather an
ExplosionPrimeEvent, which means this code has never actually been of
any use. This code block was likely dead on arrival and has been dead
for almost 9 years. Wonderful...
The proper handling here would be to listen for the PlayerInteractEvent
and detect flint and steel interactions with TNT blocks. This is super
cumbersome, however, because the event handler would have to listen on
the MONITOR priority to ensure that nothing changes down the line. What
we are _actually_ need to do is remove the TNT block from the arena's
block set when it is ignited, but since this hasn't worked for almost a
decade, we're not really in a hurry to fix it now. It just makes for a
slightly slower (but negligible) cleanup procedure most of the time.
This property makes TNT explosions look more "real" to other plugins who
may be consuming events from MobArena's sessions. It also gives way to a
potential rework of the "planter" logic that currently makes use of the
Bukkit Metadata API.
Closes#718
Changes the behavior of the ArenaListener's teleport event handler such
that it _ignores_ teleport attempts instead of _rejecting_ them when the
player in question has the `mobarena.admin.teleport` permission. Because
the global listener's event handler only ever checks the permission if
at least one per-arena listener has _rejected_ the teleport attempt, and
none of them have explicitly _allowed_ it, the change means that it will
never check the permission, because its internal `allow` flag will never
change to `false`. Thus, the check can be safely removed from the global
listener's logic.
When the response is to ignore instead of reject, the message that would
have otherwise been sent to the player is skipped. This fixes#702.
It is perhaps tempting to move the permission check up into the section
of sanity checks in the global listener, but this is very specifically
avoided to allow MobArena to _uncancel_ teleport events that have been
cancelled by other plugins, but that MobArena might need to go through.
Please see afcc526a71 for more info.
This is technically not necessary when pets are simple entities like
wolves and zombies, because these types of pets will never target and
thus attack other players in the arena. However, projectile entities
such as Blazes and Ghasts may hit players in the line of fire, and so
any such damage should be cancelled.
Fixes#712
It turns out that, according to the Spigot API, flaming arrows actually
_change_ TNT blocks to air before/instead of igniting them. While this
is a little counter-intuitive, the fix seems to revolve around allowing
the change to happen if the changed block is TNT and the "changer" is an
arrow.
Allowing the change event to happen means "foreign" primed TNT entities
will spawn, which makes it necessary to clean them up during the session
cleanup phase in ArenaImpl.
Fixes#696
It's not clear if this change actually solves the underlying issue,
because it has not been tested in a controlled environment. However,
considering the EntityExplodeEvent constructor, which doesn't take a
cancel state value, and the `HIGHEST` priority setting of MobArena's
event handler, the issue is realistic, and this change very likely
solves the issue.
Fixes#704
The arena signs predate the use of arena slugs everywhere, so something
slipped through the cracks in this regard. Incidentally, the handler for
arena updates is one of the few classes in the signs package that has no
unit tests, probably due to it being "obvious implementation". Not so
obvious after all, it seems, so now we have a basic test for it.
Fixes#705
Introduces a new arena setting to force spectators to leave the arena
when the current session ends. If set to `true`, it "overrides" the
behavior of `spectate-on-death: true` when players respawn, such that
they only become spectators if there is an active arena session.
The respawn behavior technically means that it is possible for a player
to begin spectating "the next session" if a new session begins between
them dying and clicking Respawn on the death screen. Ideally, we would
want the player to auto-leave, because the session _they_ participated
in ended, but we don't have a concept of "previous sessions", so this
quirky behavior will have to do.
Closes#682
By turning things into their own thing pickers, we can avoid creating a
bunch of SingleThingPicker wrappers, which are now redundant. Because
the semantics of using things and thing pickers are quite different, it
does perhaps make it necessary to be a bit more careful about picking
the right type.
We're good with on-demand imports for the static Hamcrest and Mockito
functions in test code, because the arrange and assert steps of unit
tests all make use of them, so it's known where they come from.
This commit makes the ThingGroupPicker return different values depending
on the size of the picked result set. If the set is empty, the picker
returns null, emulating a NothingPicker. If the set is a singleton, the
element itself is returned, emulating the SingleThingPicker. Finally, if
the set contains more than one element, a ThingGroup containing those
elements is returned.
This commit filters the result list of a ThingGroupPicker by a non-null
predicate to avoid null values in the resulting ThingGroup instance.
Since null values represent `nothing`, and we don't usually announce it
when players earn a `nothing` reward, it makes sense that they wouldn't
bubble up and somehow "manifest" in groups of things either.
Fixes#691
This commit introduces a new type variant, `angry-bee`, which is a bee
whose anger level is maximized upon spawning, much in the same vein as
its angry wolf cousin.
Note that bees are not available prior to Minecraft 1.15, so a sentinel
`null` name is used in the registration to prevent warnings from being
logged on server versions that don't have a concept of bees.
Closes#584
Because this bullet covers both the commit with the version bump and the
following compatibility commits, it didn't really fit properly into any
one of those commits, so now it gets its own!
Instead of relying on the evolving Material enum values, we can use the
new BlockData API as advised by some of the folks "in the know". It is
unclear how much of a performance impact the the "block data gathering"
and `instanceof` checks incur, but this is a pretty secluded place in
the code base, so probably nothing to worry about.
An alternative solution could have been to check if the _name_ of the
Material equals "SIGN" or "WALL_SIGN", or ends with "_SIGN". That should
cover all cases in a sorta kinda safe manner, but it isn't as resilient
as the BlockData/BlockState hierarchies.
We could also employ the new Materials utility class and enumerate all
sign types by name and just check for membership of the resulting set,
but this creates another brittle crash point.
Introduces the Materials utility class, which works much like the static
MACreature registration process, but for certain material types. Instead
of storing everything in a stringly typed map, certain Material values
are stored as constants.
Right now it's just the `OAK_SIGN`/`SIGN` pair for the autogeneration in
MAUtils, and chances are we can throw it out at some point in the near
future, but at least now there's room for more materials, should the
need arise.
Changes the fallback default wave in the wave parser to be consistent
with the default wave of the default config-file. Not a huge deal, but
it does feel a bit more neat.
Replaces the `zombie-pigmen` entry in the `spec1` wave of the default
configuration with `slimes-big`. This is due to the pig zombie rework
into piglins, which makes the custom name a little flaky. If we ever
decide to ditch the custom names somehow, this will probably make such a
transition a little easier.
While `slimes-big` is also a custom name, so are `powered-creepers` and
`angry-wolves`, but at least they are "API stable" in comparison, which
is really what this change is about.
It's K&R style braces in all modern portions of the code base, and
IntelliJ keeps complaining about final values, so let's just make it
happy.
The `plural` field should go away at some point, but for now it's fine
to just null it out, since the legacy `register()` method is only ever
used by the deprecated-for-removal constructors, both of which actually
set it to something non-null.
The new constructor and registration method pair is what other plugins
should be using. They provide a slightly cleaner interface, and we don't
need so many different ways to do the same thing.
This commit changes the way MACreature initializes its internal map of
available monster types. Specifically, it replaces all static references
to EntityType enum values with a series of registration helper methods
that try to resolve EntityType enum values from a given set of strings,
stopping on the first match. If nothing matches, no registration, but a
warning is logged unless a specific sentinel `null` value is passed as
an argument, in which case the key is just silently skipped (not in use
as of this commit, but it will be necessary for modern types like bees
from 1.15+).
While this is arguably a huge step back in terms of type safety, it does
make the code base more resilient to API version bumps, and it allows
compiling against modern API versions and still run on older (1.13-1.16)
servers. It _does_ mean that changes to the EntityType enum (such as
PIG_ZOMBIE to ZOMBIFIED_PIGLIN) must be discovered manually, which makes
maintenance a little more difficult. The warning in the server log does
make it fairly easy to fire up an early build of a new server version
and check if everything initializes correctly.
Spigot has decided to discontinue distribution of the Bukkit artifacts,
which means Bukkit "doesn't exist" after Minecraft 1.15. This commit
takes the plunge on that change and moves to the Spigot API rather than
the Bukkit API. Goodbye, Bukkit. Hello, Spigot.
The version is also bumped to 1.17 to give way to new features and bug
fixes otherwise blocked by staying on 1.13. Unfortunately, this brings
with it a swath of potential issues with version compatibility going
forward, because backwards compatibility in the API sometimes takes a
back seat, such as with the Material and EntityType enums. This commit
specifically changes some references that break compatibility with older
server versions, but compatibility with those is reinstated later.
Finally, the Spigot repository is changed to match the consensus about
which path on the Spigot Nexus instance to use. This seems to be what
everyone else is doing, so `bandwagon.jumpOn()`.
This commit introduces two reload events, MobArenaPreReloadEvent and
MobArenaReloadEvent. Both events are fired when MobArena reloads, the
former right before the reload, and the latter right after.
The reason for two events is to allow plugins to reload either entirely
or partially along with MobArena, if it makes sense for them to do so,
_when_ it makes sense for them to do so. Some plugins need to reload and
re-register themselves with MobArena before the config-file itself is
loaded (e.g. ThingParsers), while others either require MobArena to be
fully loaded or are more decoupled and thus don't depend on anything in
MobArena's innards to function. The "pre" event is for the former, while
the other event is for the latter.
As for naming, the choice of "Pre" and no prefix was made for the sake
of consistency with the event names in the Bukkit API, which has just
one example of such a pair (PlayerPreLoginEvent and PlayerLoginEvent).
Some event naming conventions (e.g. in the .NET world) seem to favor
present and past tense (reloading, reloaded), but this would be wildly
inconsistent with the rest of the event names, so it might be better to
just stay consistent. Names may change before an actual release, but for
now, this is what we're rolling with.
Closes#677
This commit introduces a new `register()` method on the CommandHandler
class to allow registering pre-instantiated subcommands. This means that
subcommands are no longer restricted in terms of instantiation, so they
can have their dependencies injected at creation, rather than having to
resort to Singleton anti-pattern means.
Also refactors the existing internal MobArena command registration to
use the new method to "drink our own champagne" and to reuse the code.
Fixes#675
This commit changes the pattern of the player list command to one that
isn't quite as greedy. This change is enough to fix issue #676 and thus
allow MobArenaStats to take control of its own command.
The command framework could definitely do with a bit of a rework away
from pattern matching towards aliases, which would prevent similar
issues from cropping up down the line. For now, this is good enough.
Fixes#676
These should have been included in their respective commits, but they
didn't seem particularly interesting for end users at the time. Might as
well include them, though.
This commit introduces a Github Actions workflow in the form of the file
`.github/workflows/build.yml`, replacing Travis CI and its `.travis.yml`
file.
The new workflow does a little bit more than Travis CI: it uploads an
artifact after a successful build. Due to a known issue with Github's
package UI, the MobArena.jar file is dynamically zipped on download, so
we get MobArena.jar.zip containing just MobArena.jar. It's redundant,
but there doesn't seem to be a simple way around it at the time of this
commit, so we'll leave it be.
For now, the workflow runs on every push to every branch. I'm probably
going to regret that, but we'll leave that as it is for now as well.
This commit re-frames the formula concept used by the wave growth, swarm
amount, and boss health wave configuration properties. It fundamentally
changes how these values are calculated, from a static, compile-time set
of enum values and hardcoded expressions, to a powerful math expression
feature that supports constants, variables, operators, and functions.
In part to remain backwards compatible with existing MobArena setups,
and in part for a better user experience, the old enum-based expressions
are relocated into a new file, `formulas.yml`, as _macros_. The file is
written to the plugin folder if missing, and it contains a formula for
each of the legacy values for each of the enums. Additionally, it has a
global section with some predefined macros for inspiration's sake. The
goal of this file is to allow people to define new formulas and reuse
them in their wave configurations instead of having to duplicate the
same formulas again and again.
Parts of the system are extensible. It is possible for other plugins to
register additional constants, variables, operators, and functions.
Closes#460Closes#461
This commit constitutes a major rewrite of how arena signs are stored
and loaded. It fixes an issue where an unloaded or missing world would
cause MobArena to throw errors if there were any arena signs recorded in
said world.
The solution is to load signs of a given world when it is available,
rather than loading all signs at once indiscriminately. At startup,
signs for all _currently available_ worlds are loaded. This fixes the
errors. When a world loads and a WorldLoadEvent is fired, signs in that
world are loaded. This ensures that all valid signs in existing worlds
will _eventually_ load. To keep things tidy, a WorldUnloadEvent will
unload all signs for the unloaded world.
Bukkit's own YAML deserialization implementation doesn't re-throw all
deserialization errors, which means we can't actually catch the problem
of missing worlds without doing an awkward "scan" of the deserialized
objects. This prompted a rewrite of the serialization and data storage
into a custom CSV-like format which is both simpler and also provides a
lot more control over the process. Instead of relying on world _names_,
the new format uses world _UUIDs_. While the rest of the plugin won't
necessarily adapt well to a world being renamed, the signs data store
should be resilient enough to handle it.
Most of the actual sign rendering code and almost all of the template
code is intact, but quite a lot of other classes have been rewritten or
replaced. Some of the rewrites weren't strictly necessary, but because
the components were already fairly small, rewrites were much faster and
a lot less awkward than attempting to adapt existing code.
Fixes#645
This file has an interesting history with content that goes all the way
back to MobArena's dark days of both Ant and Eclipse. This commit cleans
up the file a bit, removing a lot of unnecessary bits and pieces, and
making it a little more tightly bound to the current project setup.
This is a bit awkward to deal with, because the actual fix exists only
on the 1.8 branch, but we release from the main branch, so we want the
changelog entry to come from there.
Rearranges and reformats the changelog entry for the upcoming version to
better fit with the Keep a Changelog guidelines. We're already grouping
stuff decently, but we can do better with section headers. In an entry
like this one with lots of changes, it just makes the reading experience
much better, and we do want people to read the changelog :)
This is a very simple change that could potentially save a lot of
headaches when people look elsewhere in their config-file and see a
class name like "Big Gary" and wonder why putting that value in the
`default-class` property won't work.
This is really just an internal change, since the `arenaName()` method
is likely not used anywhere. If it is, however, it shouldn't be depended
on as an identifier anyway, so I'm confident that removing it won't mess
up anything that doesn't need a rework anyway.
This just removes the internal use of "lowercase name" in Upgrade Waves
and _should_ have no effect on the actual config-file and the parsing
done there.
Introduces support for multi-word arena names in the two commands. The
approach is to simply join the arguments by spaces. Because of the new
slug-based lookups, multi-word names are fairly straightforward.
This commit is a minor refactor of how arenas are resolved by name,
similar to the previous arena class resolution refactoring.
The difference this time around is that there is an ArenaMaster method
that does most of the work for a lot of different areas of the plugin,
`getArenaWithName(String)`. This method is called from well over 30
different places in the code base, making it a cornerstone of all arena
resolution. Luckily, it is called primarily from places that shouldn't
care _how_ an arena is resolved.
Affected by this commit are all commands that resolve arenas by name,
including all those not listed in the diff, because of the change to
`getArenaWithName(String)` on ArenaMaster.
Commands that can tab complete arena name arguments now complete slugs
instead of config names.
Going forward, it should be possible to show "pretty" arena names by
simply using a pretty name in the config-file instead of slug-like names
now that slugs are on the horizon.
Conversely to how the "lowercase name" of ArenaClass was deprecated in
favor of the slug, the "arena name" method here is deprecated in favor
of the "config name" of arenas. Instead of trying to pretty up an arena
name (which is doomed to fail), we just use the "config name" for pretty
printing instead, and start using slugs elsewhere.
This should give way to a much better experience with multi-word arena
names as well.
Swaps the weird "camel casing" approach in the autogenerate command out
with simply using the "config name" of the arena classes when creating
class selection signs.
This fixes the breaking change to how classes are resolved in a previous
commit, but only in future arena generation procedures. Arena generated
before this change may still contain broken class signs.
Andreas Troelsen