Adds a filesystem relay service to GUI-side which currently reports an
exmaple fsentry for /stat and /readdir. Added filesystem services for
both sides. Moved filesystem implementations from src/modules to src/lib
since they're being called by services now rather than modules.
This functionality is effectively disabled unless
src/puter-js/src/services/Filesystem.js is updated to use init_app_fs_()
when the environment is 'app'.
Adds XDIncomingService to manage messages from another window. IPC now
registers with XDIncoming and reports whether it was handled, that way
messages intended to reach GUI can be reliably ignored.
On GUI-side, XDIncomingService will be used by FSRelayService (not yet
implemented) to handle requests for filesystem operations from another
window.
On App-side, XDIncomingService might be used by the PostMessageFS
client-side filesystem implementation terminal to listen for filesystem
events that were relayed by GUI (after a permission check) from their
original websocket source. Either that, or we'll open a filesystem
socket for each app using that app's token.
NoPuterYetService was added because IPC.js is loaded before
globalThis.puter exists, and was the easiest way to still allow IPC's
listener to be registered with XDIncomingService.
APIAccessService was added. This service currently holds auth_token and
api_token and does nothing else. FilesystemService listens to this to
maintain a websocket connection. APIAccessService will help to manage
the complexity of all further code dependent on being informed about
changes to the auth token or origin. Currently in puter.js these are
passed around to several modules which manage the same piece of state
information independantly.
mkdir operation was passing user instead if actor. I really think I
changed this already; getting a crazy deja-vu feeling about it. Maybe a
commit was lost somewhere along the road, but not sure.
Dispatching and listening to events is non-trivial. The apparent
triviality is in implementing a list of listeners and calling them. The
non-triviality is in the nature of what happens to a system when it has
multiple different interfaces to register listeners and publish events.
This commit adds TopicsFeature, which allows any class extending
AdvancedBase to declare topics. A topic is a simple pub/sub channel.
TopicsFeature will manage the state of listeners so the class doesn't
need to. A GC-friendly mechanism for detaching listeners is also provided.
The ServiceManager will replace manual management of services within
initgui, and will also be used within puter.js. Eventually Puter's
backend might use this instead of the existing Container class, although
this will be a large change that needs to be done incrementally.
The difference between ServiceManager and Container is the logic behind
when initialization occurs. Rather than have all services initialized at
once when Container's init() method is called, services are initialized
as soon as their dependencies have been initialized.
Adds CachedFilesystem layer to client filesystem chain. Currently only
stat is implemented. The stat implementation will hold onto an entry for
3 seconds as per hardcoded configuration. Eventually, once invalidation
via websockets is working, this TTL should be extended.
Support for trait method overrides will make it possible to use putility
traits with the decorator pattern while using a proxy class to avoid
redundant re-implementation of proxy methods.
Use of the pattern described above will occur in the implementation of
client-side filesystem caching.
- de-coupled xhr callback passing from the interface of the underlying
filesystem implementation.
- This makes the interface to delegate calls more suitable for use with
the decorator pattern.
- The decorator pattern will be used to manage the complexity of the
caching layer by separating the concerns of different caching methods.
This hides database entries with empty types from the list of suggested
apps for files that have no extension. These entries should also be
removed from the database and the bug causing them should be fixed in
dev center.
This could not be tested locally due to another but only affecting
self-hosted installations. This will be tested in staging in the
meantime while the other bug is being fixed.
Return to phoenix shell when a pty stream is closed by twisp.
Pipes in phoenix with commands from the emulator do not appear to be
working properly, but something in there is working.
This involves establishing the protocol through which phoenix instances
run a command on the emulator. The pty is able to communicate with the
terminal in both directions. This commit adds logs to be removed later.
There are a few things left that this commit does not address:
- handling close of delegate process
- handling sigint from phoenix to delegate process
- closing the connection to twisp
I'll elaborate here because the commit name is confusing. Any time an
app gets a connection to another app, either because it launched that
app (or was launched by that app) or requested a connection to that app,
the ID the app gets to represent the app it's communicating with is now
a pseudo app id rather than the app instance ID. This accomplishes two
things:
1. It's more secure. There are multiple places where GUI assumes that
knowing an app's instance ID means you can do things as that app.
2. Between the same two apps, there may now be more than one connection.
This is useful for situations like Phoenix shell talking to the
emulator in multiple separate instances to pipe commands. (this is
coming later)
Changing the email address on an account that did not already have a
confirmed email would result in the email_confirmed column never being
set. This commit fixes that.
Additionally, an SQL statement using the unconfirmed email to identify a
row was changed so it now uses the user ID instead, which is safer.
Resolves: 759
