| Commit message (Collapse) | Author | Age |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* build(utils): Replace `async_std` with `tokio`
removing an entire async runtime from the dependency tree. `tokio` has
previously been required by `wasmtime` in any case, and the version of
async_std we used was ancient (1.3) and relied on unstable features.
* build(server): Remove `async_std` from the codebase
and replace it with `tokio` instead, which is already used by
`zellij-utils` and `wasmtime`.
* build(server): Remove dead dependency.
* fix: Don't keep an async runtime in the `zellij-utils` subcrate
but prefer to reuse an existing runtime from a caller instead. If no
such runtime exists, create an ephemeral runtime on the current thread
to run async tasks instead. Ideally, the functions needing the async
task would be async themselves, but these are called by the `Layout`
code based on a `ClientMsg`, and the code handling this currently lives
in `zellij-utils` and is sync itself. So, for the time being, there is
no way around this runtime requirement without a huge refactoring.
* refactor(server): Unclutter async tasks
in a few cases where either blocking code was called in an (unnecessary)
async task or an entire thread was spawned to block on the result of an
async task.
* docs(CHANGELOG): Mention PR 4672
* fix(server): Don't terminate when killing a panes children fails
but log the error and continue instead.
* feat: Add config option for number of async tasks to spawn
so users on machines with few resources can configure a task limit
without requiring recompilation.
|
| |
|
|
|
|
|
|
|
|
|
| |
* only show startup loading message if we've been loading for more than 400ms
* create relevant cache folders before attempting to write startup cache files
* fix flaky tests
* rustfmt
* add pr
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* xtask/run: Use varargs when run with `-data-dir`
Previously any additional arguments passed on the command line were
ignored. Now they are appended to `cargo run ...` as documented.
* server/os_i_o: Improve error message when IPC dies
and display the last send/recv error to the user instead of a generic
"Buffer full" message.
* server/lib: Log error in `send_to_client!`
so we will know when an error occured while trying to send a message to
the client. The most likely cause for this is that the client buffer
filled up and hence we cannot send any new messages. While we still
disconnect the client as before, now we also write a log message that
explains the situation.
* utils/channel: Apply rustfmt
* server/lib: Detect when client responds too slow
and log a message before disconnecting it.
* server/os_i_o: Add retry queue to client senders
that is dynamically allocated on-demand and stores `ServerToClientMsg`
in case the regular IPC channel is currently full. This acts as a
dynamic buffer to hold and buffer messages for a while until the client
hopefully catches up.
Also write a message to the log to indicate when the client is
recognized to be too slow in handling server messages.
* server: apply rustfmt
* utils/ipc: Add session name to "Disconnect" error
* utils/ipc: Fix error message indent
* server/os_i_o: Undo IPC channel extension
via `Vec` and drastically increase the IPC message queue size instead.
Measurements didn't discover a drastic increase in RAM caused by this,
and it is a much easier fix for the problem at hand.
* CHANGELOG: Add PR #2068
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* refactor(fakes): clean up add_terminal_input
* refactor(fakes): append whole buf to output_buffer in FakeStdoutWriter::write
* refactor(fakes): append whole buf to output_buffer in FakeInputOutput::write_to_tty_stdin
* fix(fakes): allow partial reads in read_from_tty_stdout
This patch fixes two bugs in read_from_tty_stdout:
* if there was a partial read (ie. `bytes.read_position` is not 0 but
less than `bytes.content.len()`), subsequent calls to would fill `buf`
starting at index `bytes.read_position` instead of 0, leaving range
0..`bytes.read_position` untouched.
* if `buf` was smaller than `bytes.content.len()`, a panic would occur.
* refactor(channels): use crossbeam instead of mpsc
This patch replaces mpsc with crossbeam channels because crossbeam
supports selecting on multiple channels which will be necessary in a
subsequent patch.
* refactor(threadbus): allow multiple receivers in Bus
This patch changes Bus to use multiple receivers. Method `recv` returns
data from all of them. This will be used in a subsequent patch for
receiving from bounded and unbounded queues at the same time.
* refactor(channels): remove SenderType enum
This enum has only one variant, so the entire enum can be replaced with
the innards of said variant.
* refactor(channels): remove Send+Sync trait implementations
The implementation of these traits is not necessary, as
SenderWithContext is automatically Send and Sync for every T and
ErrorContext that's Send and Sync.
|
| |
|
