`run-at-time' keeps `timer-list' list sorted by inserting each
timer based on the timer value. This means that
`timer--time-less-p' needs is executed ~N*N/2 times for each N
pending messages. This means that jsonrpc becomes unusable
for connections that generate a lot messages at the same time.
* lisp/jsonrpc.el (Version): Bump to 1.0.25.
(jsonrpc--process-filter): Improve performance by activating
timers in a different order. (Bug#69241)
5bb5590dec Fix blunder in labeled_narrow_to_region
78ddb32fad Fix documentation of icon-elements
725a3f32f8 ; Fix typos in symbol names
6653ee66ca Improve two docstrings in ox-latex
7d869a0402 Doc fix in auth-source-read-char-choice
f149de223b Merge branch 'emacs-29' of git.savannah.gnu.org:/srv/git/...
1f97a87879 Fix info-xref-tests
51f391998b Add @kindex in manuals for existing keybindings on 'C-x x...
This will help Eglot and some other extensions connect to network
servers that are started with a call to a local program.
* lisp/jsonrpc.el (jsonrpc--process-sentinel): Also delete inferior.
(jsonrpc-process-connection): Add -autoport-inferior slot.
(initialize-instance jsonrpc-process-connection): Check
process-creating function arity. Use jsonrpc-forwarding-buffer
(jsonrpc-autoport-bootstrap): New helper.
(Version): Bump to 1.0.20.
Try to decouple receiving text and processing messages in the event
loop. This should allow for requests within requests in both Eglot
and the Dape extension (https://github.com/svaante/dape).
jsonrpc-connection-receive is now called from timers after the process
filter finished. Because of this, a detail is that any serialization
errors are now thrown from timers instead of the synchronous process
filter, and there's no good way to test this in ert, so a test has
been deleted.
* lisp/jsonrpc.el (jsonrpc--process-filter): Rework.
* test/lisp/jsonrpc-tests.el (json-el-cant-serialize-this): Delete test.
'jsonrpc-request' already has its "interrupt on user input" mechanism
base on 'sit-for'. If called from a situation that uses
while-no-input to do basically the same, that mechanism may become
confused, so it's important to prevent that interfence.
This was confirmed to be a problem when using the Corfu completion
front-end.
See also
https://github.com/joaotavora/eglot/discussions/1127#discussioncomment-7277567
Many thanks to JD Smith <jdtsmith@gmail.com> for helping me debug this
and pointing in the right direction.
* lisp/jsonrpc.el (Version): Bump to 1.0.18
(jsonrpc-request): Bind throw-on-input to nil
EIEIO is an innacurate emulation of CLOS in many aspects and one of
them in accessor definition. Before this commit
commit 6c47931a1a
Author: Brandon <brandon.irizarry@gmail.com>
Date: Sat Nov 4 17:11:32 2023 -0400
Make EIEIO ':accessor' behave like ':reader' when reading (bug#66938)
An :initform-less, non-:initarg'ed slot with be read using an
:accessor which would just return nil. This is EIEIO specific of
course, but it made for (my) sloppy programming in jsonrpc.el and
eglot.el.
Tightening up the rules a bit meant these things broke and now I'm
fixing them.
* lisp/jsonrpc.el (jsonrpc-connection): Add a bunch of :initform nil
(jsonrpc-process-connection): Add a bunch of :initform nil
* lisp/progmodes/eglot.el (eglot-lsp-server): Add a bunch of :initform nil
5ef7ff0573 ; Start a new ChangeLog.4 file.
11126c6d30 Fix 'C-h k' for "Paste from Kill Menu" in context menus
74ddfe811f ; * doc/misc/calc.texi (Rewrites Tutorial): Fix a typo (b...
08cda286c3 Improve the documentation of the XDS support
14d1c00e80 Allow reindentation of images inserted by 'mm-inline-image'
b63a9eda01 Fix "C-h k" and "C-h c" with Paste from Kill Menu
b36c21e27d Change cursor color on NS port when it matches the face b...
96714c106b Improve documentation of image-related commands
6a2863ca01 Fix handling of sliced images
5be79fd05a ; * etc/NEWS: Announce 'cyrillic-mongolian' IM.
ca1a0fda98 ; Fix last change.
ce63462dbd Add cyrillic-mongolian input method
5880179270 ; Minor addition to the Emacs FAQ
88847dee12 Jsonrpc: don't bind inhibit-read-only to t so early
cb8c87a423 Allow active region when IM is used
# Conflicts:
# etc/NEWS
Related to https://github.com/joaotavora/eglot/discussions/1202, where
because of this very wide binding to inhibit-read-only to t, Eglot was
managing to write into read-only buffers from the response handlers
that ran from within the stack of the jsonrpc.el process filter.
This is of course illegal and dangerous, but Eglot wasn't made aware
because of the binding.
* lisp/jsonrpc.el (jsonrpc--process-filter): Don't bind
inhibit-read-only so early.
Some extensions, notably ert.el, set `debug-on-error' to non-nil,
which makes it hard to test the behaviour catching of the Elisp
error when processing a request and replying to the endpoint with
an JSONRPC-error.
The previous workaround relied on requiring lisp/emacs-lisp/ert.el in
lisp/jsonrpc.el, which really doesn't make sense.
This is better. For the single test of that behaviour, set a new
variable, jsonrpc-inhibit-debug-on-error.
Not only is this cleaner, it allows us to use ert.el's useful
debug-on-error setting.
* lisp/jsonrpc.el (ert): Don't require it.
(jsonrpc-inhibit-debug-on-error): New variable.
(jsonrpc-connection-receive): Use it.
(Package-Requires): Bump to 1.0.17
* test/lisp/jsonrpc-tests.el (signals-an--32603-JSONRPC-error):
Bind jsonrpc-inhibit-debug-on-error.
jsonrpc.el may lose JSON-RPC messages because of recursive process
filters. The problem happens in jsonrpc.el's jsonrpc--process-filter.
The code of the process filter didn't expect to be called recursively
and fails in that case.
But that can happen if the three conditions are verified.
1. the client code invoked by its jsonrpc--connection-receive inside
the process filter callee immediately sends follow-up input to
process within the same Lisp stack. This is a common scenario,
especially during LSP initialiation sequence used by Eglot, a
jsonrpc.el client.
2. that follow-up message is large enough for process-send-string to
send the input in bunches (output from processes can arrive in
between bunches).
3. the process happens to have already some more output ready
The fix in this commit detects recursive invocations and immediately
re-schedules them as non-recursive calls to the
jsonrpc--process-filter (but started from timers).
* lisp/jsonrpc.el (jsonrpc--process-filter): Rework.
(Version): Bump to 1.0.16.
By default, entering view-mode echoes a usage message. This is
particularly helpful with non-nil view-read-only, to notify the user
that view-mode has been enabled. It is less useful and more spammy,
however, if view-mode is (possibly inadvertently) entered from some
non-interactive code running in the background, such as when a major
mode is enabled in a temporary buffer for text formatting
purposes (bug#44629).
* lisp/jsonrpc.el (jsonrpc-events-buffer, initialize-instance): Use
buffer-read-only in place of read-only-mode for non-interactive use.
* lisp/view.el (view-mode-enter): Inhibit help message if either
view-inhibit-help-message is non-nil, or view-mode-enter was called
from an interactive command. Suggested by João Távora
<joaotavora@gmail.com>.
While playing around with the timing in this function, I left
this change that could freeze the function on some platforms.
* lisp/jsonrpc.el (jsonrpc-request): Use accept-process-output.
(Version): Bump to 1.0.14
