dREPL is a collection of fully featured language shells for Emacs. At the moment it supports the following interpreters:
- Python: requires IPython.
- Lua: requires luarepl and dkjson.
- Various SQL and NoSQL databases: based on usql, requires a Go compiler.
- Node.js: uses the built-in REPL library.
The following features are available, subject to variations across different REPLs (IPython supports all of them):
- Completion, including annotations and also on continuation lines
- Multi-line input editing
- Eldoc integration
- Normal pty interaction during code evaluation (e.g. debuggers)
- Graphics support via comint-mime
In fancier terms, dREPL can be described as a REPL protocol for the dumb terminal. One Elisp library defines the user interface and the client code; support for a new programming language requires only writing some backend code in the target language, plus a tiny bit of glue code in Elisp. If the target language provides a good embeddable REPL library, then the backend implementation is also reasonably straightforward.
To start a REPL, use one of the M-x drepl-* commands (making
sure first that you have the target language dependencies installed,
as described above). The rest should look familiar.
It is also possible to interact with a REPL from another buffer, say to evaluate a region of text. The relevant commands are the following:
drepl-associate: By default, dREPL tries to guess which REPL is the right one for any given buffer; an error is raised if there is no good guess. In this case, you can manually create an association with this command.drepl-pop-to-repl: Go to the REPL associated (implicitly or explicitly) to the current buffer.drepl-eval: Evaluate a string read from the minibuffer.drepl-eval-regionanddrepl-eval-buffer: Evaluate text of the current buffer.drepl-restart: Restart the interpreter. In IPython this is a soft reset; use a prefix argument to kill and start again the interpreter.
Documentation on a symbol in the REPL buffer, if available, can be
accessed with eldoc-doc-buffer.
This package extends Comint and so the communication between Emacs and the interpreter happens through a pseudoterminal. The conundrum is how to multiplex control messages and regular IO.
- From the subprocess to Emacs, control messages travel in JSON objects inside an OSC escape sequence (code 5161).
- From Emacs to the subprocess, control messages are passed as lines
of the form
ESC = <JSON object> LF. If the subprocess communicates over a PTY and the encoded message is too long to fit a line (this is an OS-dependent limit), then the message payload is split into fragments. All fragments except the last are transmitted asESC + <JSON fragment> LF.
At any given point in time, the subprocess expects either a framed
messages like this or regular IO. Emacs keeps track of the state of
the subprocess through status notifications as described below.
There are three types of message: requests, to which a response is expected, and notifications, to which no response is expected. A message contains the following fields:
op: The operation name. It must be present in every notification and request but is absent in response messages.id: A unique number which should be present in every request and repeated in the response message. It is absent in notification messages.- Further fields are parameters specific to each type of request, notification or response.
The following operations are defined:
The interpreter indicates whether or not it is ready to receive a framed operation message.
Parameters:
status: Eitherready(subprocess is expecting a framed message),rawio(IO, if it occurs, should not be framed) orbusy(no IO is allowed).
Note: The editor keeps track of the interpreter status and implicitly
switches to busy every time a request is sent. It is the
interpreter’s responsibility to notify about all other status changes.
Evaluate some code, blocking until the computation is complete.
Parameters:
code: The code to be evaluated
Result: The response contains no data (that is, it includes only the original request id). The REPL should evaluate the code and print the result.
Get completions at point.
Parameters:
code: A code snippet containing the completion point.pos: The offset (zero-based) from start ofcodeto the point of completion.
Response:
prefix(optional): The portion of code that is being completed.candidates(optional): A list of completion candidates, either strings or objects containing the following attributes:text: The completed text, including the existing prefix.annot: Annotation text to be displayed next to the candidate in the completion UI.
Check if a continuation line is needed.
Parameters:
code(string): A code snippet.
Result:
status: One ofcomplete(the code is valid),incomplete(the code is syntactically invalid, but may become so by adding more text) orinvalid(there is a syntax error in the existing portion of code).indent(optional): If present, this is the expected indentation of a continuation line, as a string.prompt: The prompt of a continuation line.
Obtain information on the symbol at point.
Parameters:
code: A code snippet.pos: An offset (zero-based) from start ofcodecontaining the symbol of interest.
Result: The response may be empty (no information on the symbol) or as follows.
name: The symbol name.type(optional): The symbol type or function signature.text(optional): Free-form documentation on the symbol.
Set configuration options. The parameters are arbitrary and interpreter-specific. The interpreter must send an empty response.
Indicates that the editor should send a setoptions request.
Typically emitted when the interpreter is initialized but before
printing the first prompt. Implicitly changes the tracked interpreter
state to ready.
This package is intended to do what the good old Comint does, but polishing some rough edges. For example, completion in Comint is spotty and one is able to edit only the last line of a multi-line input.