Updated Feb. 10, 2018
Just over a year ago I wrote a post about using Emacs is a C++ IDE. Over the past year many small improvements have led me to an entirely different configuration that I find to be much faster and easier to use. In this post I will show screencasts of the features I’m using and provide my entire init file at the end. There should be enough comments in the init file to make it clear what’s going on. If not leave a comment and I’ll try to explain.

Note: I use emacs-nox, i.e. a non-graphical Emacs. You may have to perform some minor changes to get things to work in the graphical Emacs. Edit: I’ve started using EIN to be able to edit Jupyter notebooks inside the GUI Emacs. While most of my work is still in the terminal, I now have more confidence that this configuration works well on both GUI and nox Emacs.

Overview of Changes

First I’ll just briefly outline what I changed and why.

  • Use Emacs server-client: faster startup when opening emacsclient than when opening Emacs
  • Better battery use: RTags is way too CPU intensive for what it offers
  • Better tags navigation: again, RTags uses too much CPU (really bad for a laptop), and it also doesn’t scale well to really large projects that have more than 100,000 lines of C++.
  • cmake-ide is clumsy: that’s just it. It requires a lot of configuration for each individual project, and depends too heavily on RTags
  • Autocompletion using company-mode with semantic, irony, and rtags is really slow, especially with large projects
  • Ivy turns out to be a faster, more lightweight alternative to Helm that suites my needs just fine.

The biggest reason that I continued investigating alternative configurations is that RTags was just way too CPU intensive. It would use up all 8 hyperthreads if I let it, and every time I changed something in a low-level header file it would re-parse the entire project. This was just way too expensive on a laptop. It might be okay if you have a powerful desktop and don’t care about battery life, but when you’re on the go it’s just not a realistic implementation.

Faster Startup - The Emacs Server and Client

One thing that can be annoying as an Emacs configuration grows in complexity is the increased startup time. There are a few ways of dealing with this. One that looks promising but I have not tried yet is [use-package][se-package]. The basic idea seems to be loading packages lazily so that the total load time is spread out rather than all at once.

I’ve opted for using the Emacs server-client approach. What this means is I start Emacs once at login using systemd, then connect to the running session using the emacsclient. The alias I use to connect to Emacs is alias ec="emacsclient -c". My systemd file is in ~/.config/systemd/user/emacsd.service and contains:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
[Unit]
Description=Emacs: the extensible, self-documenting text editor
Documentation=man:emacs(1) info:Emacs

[Service]
Type=forking
ExecStart=/usr/bin/emacs --daemon
ExecStop=/usr/bin/emacsclient --eval "(progn (setq kill-emacs-hook nil) (kill-emacs))"
Restart=on-failure
Environment=DISPLAY=:%i
# Provide access to SSH
Environment=SSH_AUTH_SOCK=/run/user/1000/keyring/ssh
# Setup modules
Environment=PATH="/home/nils/Research/mosh/install/bin/:/home/nils/Research/spack/opt/spack/linux-antergosrolling-x86_64/gcc-7.1.1/catch-1.9.4-qgdxfjhwk4mqhelhn2ggxmvyxvx6xm3k/bin:/home/nils/Research/spack/opt/spack/linux-antergos17-x86_64/gcc-6.3.1/gsl-2.3-hsntrnj45jqjdmb6fq74tnxkmqxekts6/bin:/home/nils/Research/spack/opt/spack/linux-antergos17-x86_64/gcc-6.3.1/openblas-0.2.19-if6giqoypphsmgle4anxox5kmb23g3vj/bin:/home/nils/Research/install/bin:/home/nils/.gem/ruby/2.4.0/bin:/usr/local/sbin:/usr/local/bin:/usr/bin:/usr/lib/jvm/default/bin:/usr/bin/site_perl:/usr/bin/vendor_perl:/usr/bin/core_perl:/home/nils/Research/spack/bin:/opt/intel/bin:/opt/intel/vtune_amplifier_xe_2017.2.0.499904/bin64:/home/nils/Research/llvm/build/bin:/home/nils/Research/templight-tools/build/bin:/home/nils/Research/standardese/build/tool:$PATH"
Environment=CPATH="/home/nils/Research/spack/opt/spack/linux-antergos17-x86_64/gcc-6.3.1/brigand-master-pcaoburzhlfibhh4mvmcia2e6dkbry5x/include:/home/nils/Research/spack/opt/spack/linux-antergosrolling-x86_64/gcc-7.2.0/kvasir-mpl-develop-hr2lgqezjkiif3vbjsoxccorj2f5wppm/include:/home/nils/Research/spack/opt/spack/linux-antergosrolling-x86_64/gcc-7.1.1/catch-1.9.4-qgdxfjhwk4mqhelhn2ggxmvyxvx6xm3k/include:/home/nils/Research/spack/opt/spack/linux-antergos17-x86_64/gcc-6.3.1/yaml-cpp-master-gmc3k4n2vsvmqwjcyqrxwaz2h5fukxc2/include:/home/nils/Research/spack/opt/spack/linux-antergos17-x86_64/gcc-6.3.1/gsl-2.3-hsntrnj45jqjdmb6fq74tnxkmqxekts6/include:/home/nils/Research/spack/opt/spack/linux-antergos17-x86_64/gcc-6.3.1/openblas-0.2.19-if6giqoypphsmgle4anxox5kmb23g3vj/include:/home/nils/Research/spack/opt/spack/linux-antergos17-x86_64/gcc-6.3.1/libxsmm-1.8.1-mevsuzxzo47g33rihi4t7fdohp3ga7gk/include:/home/nils/Research/spack/opt/spack/linux-antergosrolling-x86_64/gcc-7.2.0/blaze-3.2-mnkglifq5s6ib5ltbvil4xt66fekqq2l/include:/home/nils/Research/install/include::/opt/petsc/linux-c-opt/include:$CPATH"
Environment=LD_LIBRARY_PATH="/home/nils/Research/spack/opt/spack/linux-antergos17-x86_64/gcc-6.3.1/yaml-cpp-master-gmc3k4n2vsvmqwjcyqrxwaz2h5fukxc2/lib:/home/nils/Research/spack/opt/spack/linux-antergos17-x86_64/gcc-6.3.1/gsl-2.3-hsntrnj45jqjdmb6fq74tnxkmqxekts6/lib:/home/nils/Research/spack/opt/spack/linux-antergos17-x86_64/gcc-6.3.1/openblas-0.2.19-if6giqoypphsmgle4anxox5kmb23g3vj/lib:/home/nils/Research/spack/opt/spack/linux-antergos17-x86_64/gcc-6.3.1/libxsmm-1.8.1-mevsuzxzo47g33rihi4t7fdohp3ga7gk/lib:/home/nils/Research/install/lib:::/opt/petsc/linux-c-opt/lib:$LD_LIBRARY_PATH"
Environment=LIBRARY_PATH="/home/nils/Research/spack/opt/spack/linux-antergos17-x86_64/gcc-6.3.1/yaml-cpp-master-gmc3k4n2vsvmqwjcyqrxwaz2h5fukxc2/lib:/home/nils/Research/spack/opt/spack/linux-antergos17-x86_64/gcc-6.3.1/gsl-2.3-hsntrnj45jqjdmb6fq74tnxkmqxekts6/lib:/home/nils/Research/spack/opt/spack/linux-antergos17-x86_64/gcc-6.3.1/openblas-0.2.19-if6giqoypphsmgle4anxox5kmb23g3vj/lib:/home/nils/Research/spack/opt/spack/linux-antergos17-x86_64/gcc-6.3.1/libxsmm-1.8.1-mevsuzxzo47g33rihi4t7fdohp3ga7gk/lib:/opt/petsc/linux-c-opt/lib:$LIBRARY_PATH"
Environment=CMAKE_PREFIX_PATH="/home/nils/Research/spack/opt/spack/linux-antergos17-x86_64/gcc-6.3.1/brigand-master-pcaoburzhlfibhh4mvmcia2e6dkbry5x:/home/nils/Research/spack/opt/spack/linux-antergosrolling-x86_64/gcc-7.2.0/kvasir-mpl-develop-hr2lgqezjkiif3vbjsoxccorj2f5wppm:/home/nils/Research/spack/opt/spack/linux-antergosrolling-x86_64/gcc-7.1.1/catch-1.9.4-qgdxfjhwk4mqhelhn2ggxmvyxvx6xm3k:/home/nils/Research/spack/opt/spack/linux-antergos17-x86_64/gcc-6.3.1/yaml-cpp-master-gmc3k4n2vsvmqwjcyqrxwaz2h5fukxc2:/home/nils/Research/spack/opt/spack/linux-antergos17-x86_64/gcc-6.3.1/gsl-2.3-hsntrnj45jqjdmb6fq74tnxkmqxekts6:/home/nils/Research/spack/opt/spack/linux-antergos17-x86_64/gcc-6.3.1/openblas-0.2.19-if6giqoypphsmgle4anxox5kmb23g3vj:/home/nils/Research/spack/opt/spack/linux-antergos17-x86_64/gcc-6.3.1/libxsmm-1.8.1-mevsuzxzo47g33rihi4t7fdohp3ga7gk:/home/nils/Research/spack/opt/spack/linux-antergosrolling-x86_64/gcc-7.2.0/blaze-3.2-mnkglifq5s6ib5ltbvil4xt66fekqq2l"
TimeoutStartSec=0

[Install]
WantedBy=default.target

The important part is the Environment=SSH_AUTH_SOCK on line – this ensures that you have access to your SSH agent from within the Emacs session. Without this line you will need to re-enter your SSH key’s password each time you use it. Lines 14-18 add to various PATHs that I need for my work. Once you’ve created the file run systemctl --user enable emacsd.service && systemctl --user start emacsd.service to start the backgrounded Emacs process and also set it to start at login. The Emacs daemon can be restarted using systemctl --user restart emacsd.service.

Ivy and Swiper

I’ve switched to using Ivy for fuzzy matching instead of Helm. Ivy is quite a bit smaller than Helm because it only serves as the underlying completion engine that other packages plug in to. The important thing to note is that I also use ripgrep for faster searching. If you do not or cannot install ripgrep then just remove any lines referring to ripgrep or rg from my init file. In the screencast below I show the use of Ivy’s fuzzy matching when opening a file, as well as when using project-find-file. For ease-of-use I’ve mapped C-x M-f to project-find-file. Because of this I haven’t really found a strong reason to use projectile.


Ivy Example

Another useful plugin that builds on top of Ivy is swiper, which replaces the standard search in Emacs. Rather than having to manually jump through the file, swiper shows all the matches in an extended minibuffer. This means that forward and reverse search are effectively the same and both can be remapped to swiper.


Swiper Example

Similar to RTags, there are CTags, and GTags. These allow you to rapidly navigate the code base by jumping to the definition of a class or function. For example, say I want to see the definition of a function that’s used in the code I’m looking at. Well, I can jump to the definition by pressing M-.. Once I’ve understood what I needed to from the function definition I can jump back to where I was previously by pressing M-,. Similarly, M-t shows all the occurrences of the word-at-point in the tags database. Hopefully this makes the use case for tags quite clear. What I wanted was a fast-parsing, and accurate tag implementation. After trying around five different ones I finally settled on Universal-CTags as the generator for the tags database. The first step is installing Universal-CTags on your system. You should verify you have Universal-CTags installed and not some other implementation by running ctags --version.

For navigating the tags inside Emacs I use counsel-etags, which works really well despite being quite young. My configuration triggers a rebuild of the tags database every time you save a file in the project and 3 minutes have elapsed since the last time a rebuild was triggered. This ensures navigation to new classes and functions is possible. Note that as of this writing I override the provided auto-update function with one that is capable of using wildcards in both file names and directory names. By default only file names support wild cards. Below is a demo of the source code navigation.


CTags Example

Pretty Code With ClangFormat

If you aren’t using ClangFormat you should very seriously consider it. Here’s my simple Emacs config for it:

1
2
3
4
5
;; clang-format can be triggered using C-c C-f
;; Create clang-format file using google style
;; clang-format -style=google -dump-config > .clang-format
(require 'clang-format)
(global-set-key (kbd "C-c C-f") 'clang-format-region)

Syntax Highlighting for C++11 and Beyond

The builtin syntax highlighting doesn’t do a very good job when it comes to modern C++. Luckily there’s a remedy: modern-cpp-font-lock. Configure using:

1
2
(require 'modern-cpp-font-lock)
(modern-c++-font-lock-global-mode t)

YouCompleteMe - Fast Code Completion

The two most important parts of code completion speed and accuracy. If either of these is not outstanding then the code completion will not be useful to a seasoned developer. A third criteria that is important for me is that the code completion engine uses less CPU resources than compiling the code would, otherwise I might as well just use the compiler to give me suggestions. I also want to be able to use code completion when I’m traveling and low CPU usage means better battery life. The best solution that I’ve found is the YouCompleteMeDaemon with emacs-ycmd. ycmd uses libclang for finding completions and so is accurate and fast, even with large projects. Additionally, it only parses the file you currently have open, unlike some completion frameworks that would re-parse the entire project.

First you must install the ycmd server, for which instructions are available here. Next install the emacs-ycmd package in Emacs, which hooks into company and flycheck to provide completions and on-the-fly syntax checking. In the screencast below I show the use of the code completion and syntax checking features together to output the size of a container to standard out. The error message that appears at the end in the minibuffer tells us that the <iostream> header was not included.


ycmd Example

In my init file I have separated the setup of ycmd, company, and flycheck into three different sections for easier maintenance in the future. With ycmd, company, and flycheck I get really fast and really accurate completions with my CPU idling most of the time, so I also get great battery life. Finally, mission accomplished!

YouCompleteMe Tips

I use YCMD for a fairly complex C++14 project that uses many new language features and has a lot of class and function templates. To get really good code completion I’ve developed a python script for YCMD that very aggressively finds compilation flags for header files. The reason such a script is necessary at all is because a compile_commands.json file does not contain any compilation flags for header files, since they aren’t compiled. I’ve shared the script as a Gist here. The script must be placed at ~/.ycm_extra_conf.py for the Emacs configuration to find and use it. If you do not want to use the script, then remove the lines

1
2
3
(set-variable 'ycmd-extra-conf-whitelist '("~/.ycm_extra_conf.py"))

(set-variable 'ycmd-global-config "~/.ycm_extra_conf.py")

from the Emacs init file.

Another thing to be aware of is that getting YCMD, or more specifically libclang, to play nicely with precompiled headers took some work. See this issue on the YCMD GitHub for details. The short story is that if you use precompiled headers you might have to build YCMD using your system libclang by passing --system-libclang to the build script.

Git Source Code Management

I also perform all my git operations inside Emacs using magit, and use git-gutter to show me which lines I’ve changed in the current file. Magit is extremely feature rich and I highly recommend reading through the lengthy documentation. You will very quickly save the time you invest by no longer typing out long git commands. There are also screencasts available on the magit site, so I won’t show any here.

Fast Startup Time

Note: This section was added on Feb. 10, 2018. The functionality of the configuration does not change, but it is faster to load.

I don’t do much python development and I especially try to avoid notebooks, but sometimes I have no choice. I find web browsers to be a terrible development environment and so I’ve started using EIN for editing Jupyter notebooks. This makes the process much more tolerable. Unfortunately, starting a Jupyter server when starting Emacs takes ~2.5s, which is a long time. I use an Emacs server on my development machine, but when working on a supercomputer I typically don’t, or still have to have it load when I log in. I decided to do some research and use use-package to speed up the load time. The result is that my Emacs startup time is down to ~0.6s from ~4.7s without loss of functionality. While I’d love to have it be 0.1s, 0.6s makes the startup time very tolerable. I won’t say more about this other than that I’ve shared my most recent version of my ~/.emacs.el file in the Gist.

My init File/Installation

I’ve shared my init file as a Gist here. The only things you will need to do to have it completely replace your current configuration are install ycmd and change the path to where you installed ycmd (search for /home/nils/Research/ycmd in the .emacs.el file). To ensure Emacs loads the new init file make sure that you do not have an ~/.emacs or ~/.emacs.d/init.el file. Make sure you have an internet connection when you start Emacs because it will download and install any missing or outdated packages automatically for you. If everything is correct you should not receive any errors or warnings in the *Messages*, *warnings*, and *ycmd-server* buffer.

I’ll now provide a step-by-step guide to installing my configuration.

Prerequisites:

  • Emacs 24.5 or newer
  • YCMD and its requirements (e.g. glibc version 2.14 or newer)
  • ripgrep (optional, remove from init if you don’t have it)

Installation:

  1. Make a copy of your ~/.emacs or ~/.emacs.el, as well as ~/.emacs.d
  2. Delete ~/.emacs, ~/.emacs.el, or ~/.emacs.d/init.el, whichever you use
  3. Copy the Gist to ~/.emacs.el
  4. Start Emacs. If Emacs does not start installing packages immediately press M-x and run list-packages, then close and start Emacs again.
  5. If any packages fail to install automatically, press M-x and run list-packages to install them manually, though most will install automatically.
  6. Edit ~/.emacs.el replacing /home/nils/Research/ycmd/ycmd with the path to wherever you’ve chosen to install ycmd and restart Emacs.
  7. Copy the ycmd Gist to ~/.ycm_extra_conf.py

Note: On one machine I installed this configuration on I had to comment out the (require 'yasnippet) section, start Emacs and let everything install, then uncomment that section and restart Emacs.

Summary

In this post I gave a fairly brief overview of the major changes I’ve made to my Emacs configuration. I’ve put in a fair amount of effort into cleaning up and organizing my Emacs init file, so hopefully sharing it here is enough to get you started. If not, please leave a comment on what you want an explanation on and I’ll update this post with more information.