flow-website/content/docs/architecture/inner_data_exchange.smd
2025-10-31 21:41:48 +01:00

122 lines
No EOL
5.5 KiB
Text

---
.title = "Command arguments and message passing",
.date = @date("2025-10-27T00:00:00"),
.author = "CJ van den Berg",
.layout = "tutorial.shtml",
.draft = false,
.custom = {
.githubedit = "docs/architecture/inner_data_exchange.smd",
.codepath ="src/tui/editor.zig",
},
---
Flow uses actor model to offer an snappy experience when working with
projects that have tens of thousands of source files, also features
async communication with the threads that are working in independent
tasks that contribute to have vcs, lsp and tree-sitter integration,
apart from the directory introspection to make available what is
expected from an IDE. The command arguments travel to the target
command and are en/decoded powered by
[cbor](https://github.com/neurocyte/cbor), the same as the parameters
that are sent from one thread to another. The process management is
provided by [thespian](https://github.com/neurocyte/thespian).
This chapter describes the mechanisms that flow has to pass arguments
between components.
[]($section.id('libraries'))
## Library usage
* The Thespian library sends and receives `thespian.message` values,
which are strongly typed, but schema free CBOR encoded structures.
It supports spawning, linking, killing, etc., of lightweight processes
(aka the "Actor Model" with "Green Threads") and provides async file
and network IO and child process management.
* The CBOR library encodes decodes CBOR structured data to/from Zig
variables
* Encoding happens via the `cbor.write*` functions. These are wrapped
by `command.fmt` and `thespian.message.fmt` which provide fast
allocation free encoding to a thread local buffer. Note that the CBOR
data encoded via the `*.fmt` functions will only survive until another
message is encoded and must be copied somewhere for more permanent
storage, or possibly sent somewhere via thespian.
* Decoding happens via the `cbor.match`, `cbor.extract` and
`cbor.extractAlloc` group of functions. `cbor.extract` functions do not
allocate and cannot be used to extract some types that require allocation.
`cbor.extractAlloc` functions _do_ allocate and can extract arrays and
structures that require allocation. Both `cbor.extract` and
`cbor.extractAlloc` produce strings that **reference** the original CBOR
data buffer. `thespian.message.match` and `thespian.extract` functions
are fairly simple wrappers.
The most basic example on deserialization of an integer value can is shown
in [commands](/docs/architecture/command#command_arguments).
Cbor features en/decoding arrays, json and compounds of basic types and the
only requirement is to decode in the same order as encoding the data, more
samples on using cbor can be seen in
[cbor tests](https://github.com/neurocyte/cbor/blob/master/test/tests.zig).
For example, when interacting with the clipboard, the messages sent are
multiple chunks of information,
[]($section.id('scoping'))
## Buffer scoping
CBOR structures are mostly stored in a way that avoids allocation entirely.
This is really fast, but requires that you always know where the CBOR data
you are working with is stored.
* Received messages are read directly from the thespian process (actor)
receive buffer and remain in scope only for the duration of an actor's
receive method call
* `thespian.message.fmt` encoded messages are stored in the thread local
`thespian.message_buffer` and remain in scope only until the next
`thespian.message.fmt` call on the same thread
* `thespian.exit_message` encoded message are stored in the thread local
`thespian.error_message_buffer` and remain in scope only until the next
`thespian.exit_message` call on the same thread
* `command.fmt` encoded messages are stored in the thread local
`command.context_buffer` and remain in scope only until the next
`command.fmt` call on the same thread
All of this implies several things worth keeping in mind:
* `thespian.pid.send` will encode it's parameters to
`thespian.message_buffer` and then send them to the destination actor's
receive buffer. This will invalidate the contents of
`thespian.message_buffer` and destroy any message previously encoded with
`thespian.message.fmt` (on the same thread).
* Calling `command.fmt` inside a command that uses `command.Context.args`
will possibly invalidate the command's own arguments. I say _possibly_
because the `ctx.arg` may come from somewhere else entirely, like the
actor's receive buffer if the command was called remotely, or some other
explicitly allocated buffer.
* Use `*.fmtbuf` to encode to different buffer if there may be scoping
issues. You can allocate and scope this buffer any way you want.
* Calling `thespian.exit_message` while propagating an error up the stack
that was previously created with `thespian.exit_message` will overwrite the
original error
* Don't ever try to free a CBOR buffer unless you know exactly where it came
from.
* Strings extracted from CBOR buffers are **references** into the original
CBOR data and will be invalidated implicitly when the CBOR buffer they came
from is invalidated/overwritten.
[]($section.id('next'))
## Next steps
Serialization and deserialization occurs almost everywhere
* [Editor](/docs/architecture/editor)
* [Commands](/docs/architecture/command)
* [Minimodes](/docs/architecture/minimode)
* [Architecture](/docs/architecture)
[]($section.id('more'))
## More information
* [Deepwiki on cbor](https://deepwiki.com/neurocyte/cbor)
* [Samples of cbor usage](https://github.com/neurocyte/cbor/blob/master/test/tests.zig)
* [Deepwiki on thespian](https://deepwiki.com/neurocyte/thespian)