João Távora [Wed, 1 Jan 2020 22:05:29 +0000 (22:05 +0000)]
Avoid double shutdowns and simplify shutdown logic
* eglot.el (eglot-shutdown): Don't turn off eglot--managed-mode here.
(eglot--on-shutdown): Rather here, but without autoshutdown.
(eglot--managed-mode): Don't check eglot--shutdown-requested.
See also https://github.com/joaotavora/eglot/issues/386.
* eglot.el (eglot-code-actions): Replace tmm with completing-read
Copyright-paperwork-exempt: yes Co-authored-by: João Távora <joaotavora@gmail.com>
GitHub-reference: close https://github.com/joaotavora/eglot/issues/393
New eglot-confirm-server-initiated-edits defcustom
* eglot.el (eglot-confirm-server-initiated-edits): New defcustom.
Copyright-paperwork-exempt: yes Co-authored-by: João Távora <joaotavora@gmail.com>
GitHub-reference: close https://github.com/joaotavora/eglot/issues/382
Augusto Stoffel [Wed, 18 Dec 2019 12:44:28 +0000 (13:44 +0100)]
Add built-in support for tex and friends
plain-tex-mode and latex-mode are derived from tex-mode. Some other
TeX-related modes are not, so they require an explicit mention in
eglot-server-programs.
* README.md (Connecting to a server): Add Digestif to the list
* eglot.el (eglot-server-programs): Add Digestif for TeX-related modes
Copyright-paperwork-exempt: yes
GitHub-reference: close https://github.com/joaotavora/eglot/issues/379
João Távora [Wed, 20 Nov 2019 22:55:29 +0000 (22:55 +0000)]
Resolve compilation warnings
* eglot.el (company-tooltip-align-annotations): Forward declare.
(eglot--cached-server): Renamed from eglot--cached-current-server.
(eglot--managed-mode, eglot-current-server)
(eglot--current-server-or-lose)
(eglot--maybe-activate-editing-mode): use it.
(eglot-completion-at-point): Don't use insertTextFormat.
João Távora [Tue, 12 Nov 2019 08:51:09 +0000 (08:51 +0000)]
Let other imenu functions work if lsp server's doesn't
* eglot.el (eglot--stay-out-of-p): New helper.
(eglot--setq-saving): Use it.
(eglot--managed-mode): Use add-function :before-until for
imenu-create-index-function.
(eglot-imenu): Don't error. Fix indentation.
Felicián Németh [Tue, 12 Nov 2019 12:50:11 +0000 (13:50 +0100)]
Support serverinfo of lsp 3.15.0
Add support for serverInfo from the upcoming specification. This
changeset just stores the info sent by the server and slightly changes
a greeting message. But it opens up the possibility to identify
servers even when eglot uses a TCP connection and therefore makes
possible to implement server specific features (in eglot-x).
Old message:
```
Connected! Server `EGLOT (test-ccls/c++-mode)' now managing `c++-mode' buffers in project `test-ccls'.
```
New message:
```
Connected! Server `ccls' now managing `c++-mode' buffers in project `test-ccls'.
```
* eglot.el (eglot--lsp-interface-alist): Extend it with serverInfo.
(eglot-lsp-server): Add member variable server-info.
(eglot--connect): Store server-info and display server's name
when connected.
João Távora [Sat, 9 Nov 2019 22:58:08 +0000 (22:58 +0000)]
Provide stable eglot-current-server helper
It's better if eglot--current-server is removed, since it was being
abused by other packages, and has side effects. The only place where
it was really needed was eglot--maybe-activate-editing-mode, so the
find-and-cache logic has been moved there. All other places that can
handle a nil server now use eglot-current-server, the external
version.
* eglot.el (eglot-shutdown, eglot, eglot--read-server)
(eglot--mode-line-format): Use eglot-current-server.
(eglot--connect): Update comment.
(eglot--current-server): Remove.
(eglot-current-server): New helper.
(eglot--maybe-activate-editing-mode): find and cache the server here.
* eglot-tests.el (auto-detect-running-server)
(auto-shutdown, auto-reconnect, eglot-ensure)
(slow-async-connection): Use eglot-current-server.
João Távora [Wed, 30 Oct 2019 01:24:10 +0000 (01:24 +0000)]
Use completionitem/resolve more abundantly
It was already used to resolve documentation bits of completions, but
it can also be useful to resolve snippet templates and such.
To resolve a completion, you need some part of a completion to start
with. If it has a :data field exists and the server supports
:resolveProvider, fetch the new object, otherwise use whatever we had
already.
* eglot.el (eglot-completion-at-point): Add another local
function for resolving completions.
GitHub-reference: per https://github.com/joaotavora/eglot/issues/50
João Távora [Mon, 21 Oct 2019 21:25:13 +0000 (22:25 +0100)]
Fix race condition when company-completing quickly
For some reason, probably related to the way that Eglot tries to
maintain the responsiveness of Company completion tooltips (see
below), the user's explicit input will sometimes be surprisingly
deleted by Company, leading to a horrible completion experience.
This is sometimes hard to reproduce, but appears to match this
description perfectly:
https://github.com/joaotavora/eglot/issues/319#issuecomment-542955432
Fortunately, Company has a good fix for this, which is to pass
`:company-require-match 'never` in the completion-at-point function.
This is the fix applied in this commit.
However, this line shouldn't be required since the default value for
`company-require-match` is `company-explicit-action-p`, presumably
meaning that the auto-deletion should never take place for characters
typed by the user.
This points to a bug in Company, or at least something which may have
been exacerbated by the way that Eglot aggressively fetches
completions from the server by passing :cancel-on-input to
`jsonrpc-request`, discarding out-of-date replies. Perhaps that
discarding step bears with it some side-effects that make the
`company-explicit-action-p` test return `nil` instead of the correct
`t`. Company interprets this as carte blanche to delete the last
inserted character.
* eglot.el (eglot-completion-at-point): Use :company-require-match
'never.
GitHub-reference: per https://github.com/joaotavora/eglot/issues/319
João Távora [Mon, 21 Oct 2019 15:07:38 +0000 (16:07 +0100)]
Unbreak m-x vc-revert, which reverts preserving modes
Unlike the normal revert-buffer command, vc-revert, doesn't re-apply
the major mode, meaning it was missing a didOpen to pair with the
didClose that is unconditionally sent on both commands.
Needed to use the dynamic variable revert-buffer-preserve-modes, and,
curiously, also forward-declare it to appease the byte compiler.
* eglot.el (eglot--managed-mode): Use after-revert-hook.
(revert-buffer-preserve-modes): Forward declare.
(eglot--after-revert-hook): Signal didOpen when preserving-modes.
(eglot--maybe-activate-editing-mode): Tweak comment.
João Távora [Sat, 19 Oct 2019 22:07:47 +0000 (23:07 +0100)]
Let user keep control of some variables during eglot sessions
* NEWS.md: Mention new variable eglot-stay-out-of
* eglot.el (eglot-stay-out-of): New variable.
(eglot--setq-saving): Use it.
(eglot--managed-mode): Use eglot--setq-saving for imenu. No need
to remove 'eglot-flymake-backend from diagnostic functions.
João Távora [Fri, 18 Oct 2019 15:43:50 +0000 (16:43 +0100)]
Don't immediately request completions in eglot-completion-at-point
Yet another adjustment to this function. According to the
documentation of completion-at-point-functions, we should strive to
make functions like eglot-completion-at-point "cheap to run".
Requesting completion from the server immediately after calling the
function goes against that. The reason we were doing it is that it
might have helped compute more accurate "bounds" for the return value
(START and END) from possible TextEdit completion items. But I've
decided it's not worth the effort, at least for now.
João Távora [Wed, 16 Oct 2019 13:16:52 +0000 (14:16 +0100)]
Play along with lsp's filtertext hacks
Reworked important parts of eglot-completion-at-point.
One of the tasks was to cleanup the nomenclature so it's easier to
spot how LSP and Emacs's views of completion techniques differ. When
reading this rather long function, remember an "item" is a plist
representing the LSP completionItem object, and "proxy" is a
propertized string that Emacs's frontends will use to represent that
completion. When the completion is close to done, the :exit-function
is called, to potentially rework the inserted text so that the final
result might be quite different from the proxy (it might be a snippet,
or even a suprising text edit).
The most important change in this commit reworks the way the
completion "bounds" are calculated in the buffer. This is the region
that Emacs needs to know that is being targeted for the completion. A
server can specify this region by using textEdit-based completions all
consistently pointing to the same range. If it does so, Emacs will
use that region instead of its own understanding of symbol
boundaries (provided by thingatpt.el and syntax tables).
To implement server-side completion filtering, the server can also
provide a filterText "cookie" in each completion, which, when
prefix-matched to the intended region, selects or rejects the
completion. Given the feedback in
https://github.com/microsoft/language-server-protocol/issues/651, we
have no choice but to play along with that inneficient and grotesque
strategy to implement flex-style matching. Like ever in LSP, we do so
while being backward-compatible to all previously supported behaviour.
* eglot.el (eglot-completion-at-point): rework.
GitHub-reference: close https://github.com/joaotavora/eglot/issues/235
João Távora [Wed, 16 Oct 2019 08:49:09 +0000 (09:49 +0100)]
Always filter completions client-side by prefix
Prefix completion is all we get in LSP because there are some servers
that send *all* completions everytime. This is horrible, but it's the
currently defined behaviour. See
https://github.com/microsoft/language-server-protocol/issues/651.
* eglot.el (eglot-completion-at-point): Use all-completions.
GitHub-reference: per https://github.com/joaotavora/eglot/issues/319
As is well known, LSP's and Emacs's completion mechanics don't fit
very well together, mostly because Emacs expects completion to be a
something of a pure function of a string argument and LSP treats as a
function of a concrete buffer position.
A further complication arises because some completion frontends like
"bare" completion-at-point make Emacs modify the buffer's contents
during the completion process, while other (notably company-mode)
don't do that. Thus, 'eglot-completion-at-point' must take extra care
to answer to the questions listed in the "(elisp)Programmed
Completion" info node based on its (quite hacky) "completions" local
var and _not_ based on the intermediate buffer contents. That var is
also used to cache the last LSP response and allow the :exit-function
callback to retrieve much more than just the completion text in
In yet another related problem, :exit-function won't be called at all
with completion-at-point if the completion table doesn't answer
properly to test-completion. A previous use of
completion-table-dynamic was found to be unsuitable here: we must
answer all the requests separately.
João Távora [Sun, 6 Oct 2019 15:10:33 +0000 (16:10 +0100)]
Rework and correct major part of xref glue code
See comments of https://github.com/joaotavora/eglot/pull/314. Up to
now, xref-backend-indentifier-completion-table was a gross hack that
only worked sometimes. It relied on some fugly gymnastics to cache a
response from :textDocument/documentSymbol and somehow used that
information to build a completion table. But it doesn't work well.
Summarily, LSP doesn't lend itself well to the xref interface of
prompting for an arbitrary identifier and then go look for whichever
type of references of that identifier. All the LSP
:textDocument/{definition,references,implementation,...} methods
expect to know the exact context of the search the user is about to
perform, in the form of a document location. That conflicts with the
xref "arbitrary string" requirement.
Therefore, the slightly limited, but much more correct way, for Eglot
to function is to override the user's preference of
xref-prompt-for-identifier, temporarily setting it to nil in
eglot--managed-mode (ideally, though, xref-prompt-for-identifier
should be a function of the backend.)
Later on, a possibly better behaved identifier completion table can be
built on top of the :workspace/symbol LSP method.
* eglot.el (xref-backend-identifier-at-point): Rewrite.
(eglot--lsp-xrefs-for-method): New helper.
(eglot--lsp-xref-helper): Use eglot--lsp-xrefs-for-method.
(eglot--xref-definitions-method): Delete.
(eglot--lsp-xref-refs): New variable.
(xref-backend-references, xref-backend-definitions): Use
eglot--lsp-xrefs-for-method.
(eglot--managed-mode): Set xref-prompt-for-identifier
to nil.
(eglot--xref-reset-known-symbols, eglot--xref-known-symbols): Delete
(xref-backend-identifier-completion-table): Nullify.
(eglot-find-declaration, eglot-find-implementation)
(eglot-find-typeDefinition): Use eglot--lsp-xref-helper.
* eglot.el (eglot--xref-definitions-method): New variable.
(xref-backend-definitions): Use it.
(eglot-find-declaration, eglot-find-implementation,
eglot-find-typeDefinition): New functions.
* README.md (Language features): Add new capabilities.
* eglot.el (eglot-client-capabilities): Add new capabilities.
(eglot-ignored-server-capabilites): Add new capability.
João Távora [Mon, 30 Sep 2019 16:06:48 +0000 (18:06 +0200)]
Unbreak elm language server which does use :triggercharacters
Only query completionProvider -> triggerCharacter information if the
server has provided it. Elm's, and probaly other's, do not provide
it, which doesn't mean they don't support completion.
* eglot.el (eglot-completion-at-point): Check that completion
capability is a list before treating it like one.
A capability of "null" is downright invalid, and must NOT be mistaken
for a value of "{}" (which indicates the presence of the capability)
or "False" (which indicates its asence). See
https://github.com/microsoft/language-server-protocol/issues/830#issuecomment-537849292
for a clarification from the LSP maintainer.
Don't send dummy json object in "initialized" notification ()
Eglot uses a JSON object { __dummy__ : true } as a placeholder instead
of the empty object {}. It does this out of necessity, since encoding
an empty object can't currently be easily using the current jsonrpc.el
library. However, this also causes the parameter to be actually sent
to the server.
Since the JSON-RPC specification states "The names MUST match exactly,
including case, to the method's expected parameters" this is
non-conforming to the protocol.
The LSP specification does not seem to indicate how servers should
handle method calls with parameters they do not support. As such,
ignoring the parameter, or reporting an error, or crashing all seem to
be "valid" behaviors as far as the specification is concerned.
We can avoid this by using an empty hash table instead of a dummy
parameter. Currently, an empty hash table is the only Emacs Lisp
object which jsonrpc.el serializes to an empty JSON object in
jsonrpc--json-encode.
* eglot.el (eglot--connect): Use make-hash-table instead of dummy
object.
Also use signature label offsets for parameter info
According to the LSP specification, a parameter of a callable-signature
has a label and a optional doc-commet. The label of a parameter
information is either a string or an inclusive start and exclusive end
offsets within its containing signature label.
Previously, this was only taken in account for highlighting the
parameter in the definition signature.
* eglot.el (eglot--sig-info): Handle signature label offsets
when printing the signature parameter information.
Some language servers may specify null for some capabilities in the list
of server capabilities. This does not conform to the specification, but
treating it as false is more reasonable than treating it as true.
A current example is the PHP language server. which specifies null for
every capability it does not handle, like documentHighlightProvider.
This would cause Eglot to send constant textDocument/documentHighlight
requests, which all timed out.
* eglot.el (eglot--server-capable): Change the handling of null values
for capabilities to treat them as false instead of true.
Change the default of eglot-move-to-column-function
Previous default (move-to-column) works on visual columns, the LSP
specification and the new default (eglot-move-to-column) use "real"
columns. Fixes https://github.com/joaotavora/eglot/issues/293 and https://github.com/joaotavora/eglot/issues/297.
* eglot.el (eglot-move-to-column): New function.
(eglot-move-to-column-function): Use it as default.
David Florness [Mon, 19 Aug 2019 00:09:13 +0000 (18:09 -0600)]
Require array package to use current-line ()
The jsonrpc package (one of eglot's dependencies) recently
updated and removed the line requiring the array package. Since
current-line is provided by array and is used by eglot, require
array explicitly.
Jürgen Hötzel [Mon, 12 Aug 2019 20:13:47 +0000 (22:13 +0200)]
Expand directory watcher globs containing ** ()
Previously, if the server requested a glob pattern like foo/**/*
to be watched, we would just error. Now we watch foo/bar/ and
foo/baz/ as if the server had requested those two watchers
instead of just the one with the **.
As a limitation, the implementation of file-expand-wildcards
doesn't fully handle ** globstars (** matches at most one path
segment).
* eglot.el (eglot-register-capability workspace/didChangeWatchedFiles):
Use file-expand-wildcards to make a ** glob into multiple **-less
globs.
Jürgen Hötzel [Thu, 18 Jul 2019 19:36:56 +0000 (21:36 +0200)]
Fix invalid guess for php language server ()
* eglot.el (eglot-server-programs): Change the position of the php
language server, otherwise it will always be hidden by the c-mode
server (php-mode is derived from c-mode).
João Távora [Thu, 9 May 2019 10:38:58 +0000 (11:38 +0100)]
Work around a bug in emacs's change detection
When using capitalize-word, or any case-fiddling function,
before-change-functions will record e.g. the whole word's start and
end, even though only the first character has changed. Not only is
this longer than needed but also conflicts with what we get in
after-change-functions, which records just the one-char-long change.
Also, if the word didn't need any fiddling at all then
before-change-function will run but after-change-functions won't: an
"orphan" before-change will erroneously be sent to the server.
* eglot.el (eglot--after-change): Detect problematic case and fix
change description.
(eglot--before-change): Store markers of changed region.
(eglot--signal-textDocument/didChange): Weed out orphan changes.
João Távora [Sun, 6 Jan 2019 22:14:27 +0000 (22:14 +0000)]
Rename new defcustoms with friendlier names
* eglot.el (eglot-doc-too-large-for-echo-area): Rename from
eglot-eldoc-extra-buffer-if-too-large.
(eglot-put-doc-in-help-buffer): Rename from
eglot-eldoc-extra-buffer.
(eglot-auto-display-help-buffer): Rename from
eglot-auto-display-eldoc-extra-buffer.
(eglot--eldoc-message): Use new variable names.
João Távora [Sun, 6 Jan 2019 12:55:21 +0000 (12:55 +0000)]
Fix test failure introduced by previous commit
Remove the hack of unsetting eldoc-last-message in
eglot--eldoc-message. This allows any subsequent eglot-eldoc-function
calls (prompted by simple cursor movement) to return it immediately,
thus refreshing the help buffer with the same contents. For this to
work, we also have to set eglot--eldoc-hint globally in
eglot-eldoc-function.
An alternative to making the test pass would be to keep the hack of
unsetting eldoc-last-message only in the case that we actually get to
display the help buffer. This would actually be more efficient, but
potentially more hacky.
The bottom line here is that eldoc doesn't have a good API to deal
with asynchronous docstring fetching. See this thread:
João Távora [Sat, 5 Jan 2019 22:47:43 +0000 (22:47 +0000)]
Show large docs in help buffer instead of echo are by default
* eglot.el (eglot--managed-mode): Add and remove from eglot--eldoc-message
(eglot--eldoc-hint, eglot--help-buffer): New helpers.
(eglot-eldoc-extra-buffer)
(eglot-auto-display-eldoc-extra-buffer): New defcustoms.
(eglot--eldoc-message): New helper.
(eglot-eldoc-function): Set eglot--eldoc-hint.
(eglot-help-at-point): Use new helpers.
(eglot-eldoc-extra-buffer-if-too-large): New predicate.
GitHub-reference: per https://github.com/joaotavora/eglot/issues/198
João Távora [Sat, 5 Jan 2019 13:32:13 +0000 (13:32 +0000)]
Handle (un)registercapability requests via generic functions
* eglot.el (Version): Bump to 1.4
(eglot-register-capability, eglot-unregister-capability): New
generic functions.
(eglot--register-unregister): Call eglot-register-capability,
eglot-unregister-capability.
(eglot-register-capability s (eql
workspace/didChangeWatchedFiles)): Rename from
eglot--register-workspace/didChangeWatchedFiles.
(eglot-unregister-capability s (eql
workspace/didChangeWatchedFiles)): Rename from
eglot--unregister-workspace/didChangeWatchedFiles.
João Távora [Wed, 2 Jan 2019 17:12:36 +0000 (17:12 +0000)]
Run connection hooks with proper dir-locals
eglot-connect-hook and eglot-server-initialized-hook must run in a
buffer with properly setup directory-local variables for the project.
This is crucial for things like eglot-signal-didChangeConfiguration,
which needs a properly setup value of eglot-workspace-configuration to
succeed.
I could have chosen any of the buffers where Eglot is activating
itself, but the approach using
hack-dir-local-variables-non-file-buffer seems more correct, despite
the name.
* eglot.el (eglot--connect): Run connection hooks with proper
dir-locals.
projects / emacs.git / log