emacs-ycmd is a client for ycmd, the code completion system. It takes care of managing a ycmd server and fetching completions from that server.
emacs-ycmd comprises a core set of functionality for communicating with ycmd as well as integration with the Emacs completion framework company-mode.
ycmd package itself doesn't provide a real UI for selecting and inserting completions into your files. For that you need to use
company-ycmd or another "completion framework".
First make sure that
ycmd is installed on your system. See the ycmd instructions for more details.
ycmd-mode in all supported modes, add the following to your emacs config:
(require 'ycmd) (add-hook 'after-init-hook #'global-ycmd-mode)
ycmd-mode to a specific supported mode:
(require 'ycmd) (add-hook 'c++-mode-hook 'ycmd-mode)
Use the variable
ycmd-server-command to specify how to run the server. It will typically be something like:
(set-variable 'ycmd-server-command '("python" "/path/to/ycmd/package"))
NB: We do not do filename expansion on the elements of
a result, paths using "~" to represent the home directory will not work
properly; you need to expand them yourself.
If you've got a global ycmd configuration, specify that in your
emacs configuration by setting
(set-variable 'ycmd-global-config "/path/to/global_config.py")
Spacemacs users: Note that if you don't set
will set it for you. This is not always what you want! See the spacemacs ycmd
documentation for more info.
If you've got project-specific ycmd configurations (almost certainly
.ycm_extra_conf.py), and if you want them automatically
loaded by ycmd as needed (which you probably do), then you can
whitelist them by adding entries to
example, this will allow automatic loading of all
files anywhere under
(set-variable 'ycmd-extra-conf-whitelist '("~/my_projects/*"))
Alternatively, you can set
ycmd-extra-conf-handler to control how
ycmd.el deals with non-whitelisted extra configs. By default this is
'ask, meaning it will ask the user each time one is encountered. The
other options are
'ignore, in which case the extra config will be
'load, in which case the extra config will be loaded.
Now a ycmd server will be automatically launched whenever it's needed. Generally, this means whenever you visit a file with a supported major mode. You should not normally need to manually start or stop a ycmd server.
With a server running, you can now get completions for a point in a
ycmd-get-completions. This doesn't actually insert the
completions; it just fetches them from the server. It's not even an
interactive function, so you can't really call it while editing. If
you just want to see the possible completions at a point, you can try
ycmd-display-completions which will dump a raw completion struct
into a buffer. This is more of a debugging tool than anything.
It is recommended to use
company-mode for completion, however there is basic support for Emacs' built-in completion mechanism.
(defun ycmd-setup-completion-at-point-function () "Setup `completion-at-point-functions' for `ycmd-mode'." (add-hook 'completion-at-point-functions #'ycmd-complete-at-point nil :local)) (add-hook 'ycmd-mode #'ycmd-setup-completion-at-point-function)
More likely, you'll want to use a completion framework like
company-mode to manage the completions for you. Here's how to do
(require 'company-ycmd) (company-ycmd-setup)
After this you can use your standard
company-mode keybindings to do
IMPORTANT: Unbuffered output on Windows
There have been some reports that
ycmd.el doesn't work on Windows when Python's output is buffered. See, for example, issue #104. This is because we rely on the ycmd server printing out its host and port information in a timely (i.e. unbuffered) manner. We will almost certainly update the defaults for
ycmd.el to force unbuffered output.
In any event, if you are facing problems with ycmd not starting and/or hanging Emacs on Windows, try adding
-u to your
ycmd-server-command. For example:
(set-variable 'ycmd-server-command '("c:/path/to/python.exe" "-u" "c:/path/to/ycmd"))
flycheck-ycmd.el allows you to use
ycmd as a backend for
flycheck. With this enabled, whenever
ycmd parses a file the
results will be passed to
flycheck for display. This is a really
nice way to get quick feedback on problems in your code.
The simple way to enable
flycheck integration is to use
(require 'flycheck-ycmd) (flycheck-ycmd-setup)
This will make sure that
flycheck sees the parse results, and that
flycheck-ycmd backend is enabled.
If for some reason you want to do this manually, the instructions are like this:
(require 'flycheck-ycmd) ;; Make sure the flycheck cache sees the parse results (add-hook 'ycmd-file-parse-result-hook 'flycheck-ycmd--cache-parse-results) ;; Add the ycmd checker to the list of available checkers (add-to-list 'flycheck-checkers 'ycmd)
Disabling ycmd-based flycheck for specific modes
If you use
flycheck-ycmd-setup or otherwise put
ycmd at the front of
flycheck-checkers, flycheck will use the ycmd checker for every buffer in
ycmd-mode. This may not be what you want. For example, even though ycmd
supports completion (and, thus, flycheck) for Python, you may wish to use
pyflakes for flychecking Python code.
To disable ycmd-based flychecking for specific modes, you can modify
flycheck-disabled-checkers list in your mode hook. For example:
(add-hook 'python-mode-hook (lambda () (add-to-list 'flycheck-disabled-checkers 'ycmd)))
With this, the ycmd checker will be ignored in
flycheck-disabled-checkers is buffer-local, the ycmd-based checker
will still be available for other modes.
Making flycheck and company work together
In some cases you may see that
flycheck interfere with one another. You can end up with strange completion artifacts in your buffers. This mostly seems to happen when you run emacs in "terminal mode", i.e. with
The short answer for how to deal with this is:
(setq flycheck-indication-mode nil)
The slightly longer and probably better answer is:
(when (not (display-graphic-p)) (setq flycheck-indication-mode nil))
ycmd-eldoc adds eldoc support for
(require 'ycmd-eldoc) (add-hook 'ycmd-mode-hook 'ycmd-eldoc-setup)
Note: eldoc messages will only be shown for functions which are retrieved via semantic completion.
emacs-ycmd reports found errors through emacs buttons; to integrate those with
next-error prepend something like
(require 'ycmd-next-error) before require'ing ycmd (after adding the
contrib directory to your
In some common configurations
emacs-ycmd can produce lots of messages, and some people find these noisy and distracting. If you're seeing a lot of messages like
Contacting host: 127.0.0.1:NNNNN and you'd like to quiet them, set
nil. This can effect non-ycmd-related buffers, so consider using buffer-local settings if this worries you.
You might also see a flurry of messages like this:
REQUEST [error] Error (error) while connecting to http://127.0.0.1:38987/completions. REQUEST [error] Error (error) while connecting to http://127.0.0.1:38987/event_notification. [26 times]
These almost never indicate something you need to be concerned about. To quiet them, you can set
See issue #173 for the initial discussion of this topic.
emacs-ycmd comes with a number of tests that you can run. This is mostly
useful for developers. They are built with
ert, so you can run them using any
ert provides. For example:
(require 'ycmd-test) (ert-run-tests-interactively "ycmd-test")
It is also possible to run the tests on the command-line with the Makefile provided in this repository. Before running test, you need to install the Cask in order to be able to install the package dependencies.
You can do this by running
The other thing that is required is to have the
ycmd folder right next to
To run the tests:
It is also possible to have the
ycmd server at a different location. In that
case the path needs to be passed to the
make command explicitly:
make YCMDPATH='/path/to/ycmd/ycmd' test
Make sure that you provide the path to the ycmd module and not the ycmd root directory.