* Fix tests failing due to missmatched Factorio detection
When the configured factorio directory is full of junk then the tests
use the first directory it finds for one of its tests, while the server
code uses a more elaborate logic to figure this out.
Turns out this is complicating things more than needed. Simply trying
each entry in the directory and seeing if it works when treated as a
directory always work, regardsless of symlinks, junctions,
reparse-points or whatever else filesystem magic there is.
* Skip tests requiring Factorio if it's not present
Prevent confusing test failures by skipping all the tests that require a
working Factorio installation to sucteed, and print a warning at the end
of the test suite if tests may have been skipped due to this.
* Handle ENOTDIR when determining Factorio version
If the path to the factorio install contains a component that is not a
file then ENOTDIR may be returned on POSIX like systems. Treat this the
same as ENOENT so that detection logic works when used on a file.
---------
Co-authored-by: Hornwitser <code@hornwitser.no>
Some mods, for example cubium, includes bogus folders like __MACOSX in
addition to the root folder where everything is located. This only
works by chance as Factorio will check for the folder of the first
entry in the zip file and use that.
Match Factorio's logic in findRoot so that these mods work with
Clusterio.
When the config system was ported to TypeScript the "instance" location
was accidentally added to the type of available locations for access
control. This has never been a valid location as it was deemed too
complicated to work with. The correct location for instances is
"controller" and "host" as these are the locations the configs are
handled.
If only the config of an instance is updated the dirty flag of the
instance is not set, which results in changes being discarded when the
controller shuts down afterwards.
Fix by always triggering the update logic when an instance field is
changed.
Provide ease of diagnosing credential issues by accurately logging if
the credentials were sourced from the instance config, the host config,
the controller config, or if no credentials were found.
Prevent browsers from automatically autofilling the password entry box
for the Factorio credentials by setting the autocomplete property to
new-password on the password input.
Add config entries to controller and host for the Factorio username and
token to use when talking to the Factorio APIs. If set an username and
token is not set in factorio.settings then these settings are used,
making it easy to configure one set of credentials for a cluster/host.
For running more public provide your own hardware clusters, there's an
option to disable controller sharing the credential with the hosts so
that each host have to set their own credentials.
Implement a credential option to config entries which is a list of
locations the config entry can be read from. When specified and the
option is a string field this gets rendered as a password field in the
config UI. If the credential is not readable from control the value of
the fields is cleared when the UI is refreshed.
Rewrite shutdown logic for Factorio servers to send a /quit command via
RCON for both Linux and Windows when stopping a server to avoid having
two confusingly intermixed code paths for Windows and Unix systems.
This makes shutdown on Windows tell clients the server shut down instead
of abruptly stoping to respond.
Newer versions of Ant Design requires either dataIndex or key to be
defined on columns for sorting of the tables to work. Add missing key
entries on columns.
Support the for (const [k, v] of datastore) syntax which is commonly
used with maps by adding a Symbol.iterator function returning the
appropriate iterator.
When passing methods around they lose their binding to this, which means
the fromJSON methods no longer has this be their class. Fix by binding
the methods to the class they belong to before passing them on to the
datastore.
When renaming a file over another file the operation is only atomic if
both the source file and the target file is in the same directory. Fix
safeOutputFile writing the temporary file to a different directory than
the target file due to the path separator not being included when
creating the temporary file path.
If a factorio dir is configured in config-host.json then use that in the
tests. This removes the requirement of having a factorio installation
in ./factorio or linking to one there when running the tests.
The host.controller_url config is required to end with a / for the url
joining logic to work. Fix code added #619 setting this to a value not
ending with a /.
The @swc/core module depends on optional dependencies being installed in
order to get the correct native binding's package installed. Remove
--no-optional flag passed to pnpm to install these.
The usage of shell: true means that no escaping is done on the arguments
passed to the commands invoked by the installer. This leads to the
installation failing if the user for example puts a space in the
username for the admin account.
Fix by escaping the arguments passed on command lines. This is
unreasonably complicated on Windows 😡.
Use the settings prototypes to coerce any incorrectly typed mod settings
to the correct type either by casting or by replacing the value with the
default value from the prototype.
This will fix cases where an update to a mod changes the type of ono of
its defined settings.
Due a bad order of operations when a host connection was closed the
event handler listening on "close" to remove the host from the
hostConnections mapping was added after the HostConnection's event
handler listening on "close" and invoking plugin hooks in response to
the host going offline. This in turn causes plugins that send
broadcasts in repsonse to thos hooks to throw an error due to the
broadcast sending code assuming connections that exist are valid target
to send messages to.
Fix by ordering the listener for the removal of the host connection
before the listener invoking plugin events.
Store and use a cache of the loaded ModInfo data to mod-info-cache.json
in the mods directory during startup. For simplicity this does not
cache mods added to the controller or host after startup, these get
added to the cache on the next startup.
Docement the adaptations I've made to the publishing procedure which
uses a less dangerous publish command and removes the dist folders
before publishing to avoid accidental upload of stale build files.
If a bad download or file transfer causes a mod in a host's mod folder
to become corrupt the attempt to load the mod will throw an error that's
not the file not existing. This in turn causes starting the instance to
always fail as the mod is never attempted to be re-downloaded.
Fix by ignoring errors attempting to load mods already present in the
mods folder that aren't in the mod store so that a new download is
attempted after that fails.
It's possible to set the .icons to an arbitrary key value mapping
instead of an array and the game will still accept it. Convert the
icons to an array if it's an object mapping to handle this case.
Note that this doesn't preserve the order the game reads the keys in
because JavaScript reorders numeric key indexes on objects differently
to how the game orders them.
Implement the shorthand notations that Factorio recognises for colours
where individual components may be left out or the colour specified as
an array of values instead of an object.
Build and minify the Web UI using swc instead of babel, tsc and terser.
This results in a significant speedup of the build process as it skips
checking types and uses native code to do the minification.
To prevent accidental type errors being introduced the TypeScript
compiler is still used to perform type checking on the Web UI files.
Due to peer dependency resolving shenanigans the swc/core package had to
be added to the workspace, controller and web_ui to avoid pnpm creating
a duplicate install of webpack that break the build.
Add back unqualified tsconfig.json files to each package that references
both the Node.js build and the browser build. This makes tsc --build
compile both builds when the variant to do is not specified.
Distinguish the Node.js build in the same way the browser build is with
the .node suffix to the tsconfig files it uses. This means
tsconfig.node outputs to dist/node and tsconfig.browser outputs to
dist/browser.
Use a separate folder for the TypeScript build for Node.js in the dist
folder. This makes each type of build consistently have its own output
folder instead of being interleaved with the Node.js build.
Use dist/browser as the output of the browser oriented TypeScript build
of packages instead of dist/web/build. This distinguishes it better
from the Web UI bundle which is put together with webpack.
Rename the config files for the browser build of modules to better
communicate that it's purpose is to compile code for a browser
environment rather than the Web UI in particular. This also aligns with
the "browser" option in package.json, the existing browser.ts entrypoint
in lib, and how the browser environment is typically referred to.
When the level of a technology changes either by setting level
directly or by setting researched to true, the saved progress of the
research is reset only if the technology is not the current research.
If during the initial synchronisation with the controller an instance
has progressed on a multi level research, it's the current research, and
the controller's level of that research is higher, then the level is set
to the higher level and the game keeps the progress. This in turn
results in all of the progress made on the previous level being added to
the higher level.
Fix by setting the progress to zero when changing the level or
researched status of a technology that's the active research.
Throw an error if an attempt is made to check for a permission that does
not exist in the Web UI to prevent future bad permission checks going
unnoticed.
Fix attempts to check for permissions that don't exist in the Web UI
that has gone unnoticed because the check still succeeds when the
account has the cluster admin permission.
Follow the same pattern as all the other subscribable lists by having a
separate .list and .subscribe permissions for saves. This also fixes
the save list not being visible due to a typo in the permission in the
Web UI.
Without a unique per page key set react will try to reuse the existing
state and components of the previous page and update them to match the
new page. This causes unexpected behaviour as components get turned
into into a Frankenstein abomination of the previous page and the
current one fused into one.
Fix by setting a key on the top level fragment of the page which in turn
forces React to re-render the whole page whenever the user switches to
another page.
If the value in a subscribed list is updated while the updatedAtMs
timestamp is kept the same then the caching logic of the subscription
system will cause updates to be silently dropped. Log a warning if a
subscriber catches this happening to help tracking down such cases.
Always send the instance details update when an instance is reassigned,
not just when it is unassigned. The idea for not sending an update when
assigned to a host was that the host would send an update on the status
that in turn updates the instance details. However this is not actually
the case; when a stopped instance is reassigned its status does not
change so the update is ignored.
Set the save's updatedAtMs timestamp when clearing the save list of a
reassigned or deleted instance. This fixes the Web UI not updating the
saves list to remove the saves when an instance is reassigned.
If a link is closed while the prepareDisconnect action is in progress
the link will still try to signal the disconnect ready over the socket
when the prepareDisconnect action completes, which throws due to the
socket no longer being present. This error is not caught due to the
catch happening before the finally on the promise and triggers an
unhandled rejection.
Fix by rearranging the finally to come before the catch on the promise
so that errors are always handled, and checking that the link still has
a session before attempting to send the disconnect ready signal.
When using .finally on a promise a new promise is created which calls
the handler passed to it when the the original promise is resolved or
rejected, and resolves or rejects with same value as the original
promise. This means if the rejection of the new promise that was
created by finally is not handled then an unhandled promise rejection
will be raised.
Fix by making sure the promise rejection from the finally promise
created in subspace_storage's handleUpdateStorageEvent is properly
handled.
When messages were forwarded they were sent using the low level
connector .send method which accepts a message and sends it over the
link. This method sends the message as is with no modification which
means that at no point during the forwarding logic was the message
sequence number that originated from Link A renumbered to the correct
sequence number to send out on Link B which it was forwarded on.
This in turn caused the logic which drops acked messages from the send
buffer to break which in turn meant the resume could cause duplication
of message and/or messages to be lost. With enough duplicated messages
the resume of one connection could cause other connections to drop and
also initiate resume.
Fix by renumbering the sequence on forwarded messages. In the future
the sequences sent and received should also be checked to make sure this
kind of problem does not happen again, this is not done here because
partially updating a running cluster with sequence problems to one that
checks and enforces sequences numbers will be problematic.
When host and instance id in Address was reordered in 1e66f11 the
toString method Address was updated to account for this. Fix toString
showing host addresses as being instance addresses and vice versa.
When a plugin broadcasts an event to instances there may be hosts and
instances in the cluster which does not have the plugin loaded. This
would cause an error to be logged for every instance which does not have
the plugin loaded.
Filter out broadcast events sent to hosts and instances such that
instances and hosts that does not have the event do not received them.
