Installation, portability, dependencies, and testing

Miller license

Prebuilt executables

Please see where there are builds for OSX Yosemite, Linux i686 (statically linked), and Linux x86-64 (dynamically linked).

Homebrew installation support for OSX is available via

brew update && brew install miller

You may already have mlr available in your platform’s package manager on NetBSD, Debian Linux, Ubuntu Xenial and upward, Arch Linux, or perhaps other distributions. For example, on Debian/Ubuntu, you can try

sudo apt-get install miller

Building from source

From release tarball using autoconfig

Miller allows you the option of using GNU autoconfigure to build portably.

Grateful acknowledgement: Miller’s GNU autoconfig work was done by the generous and expert efforts of Thomas Klausner.

  • Obtain mlr-i.j.k.tar.gz from, replacing i.j.k with the desired release, e.g. 2.2.1.
  • tar zxvf mlr-i.j.k.tar.gz
  • cd mlr-i.j.k
  • Install the following packages using your system’s package manager (apt-get, yum install, etc.): automake autoconf libtool flex (if you need to run autoreconf -fiv) else just libtool flex
  • Various configuration options of your choice, e.g.
    • autoreconf -fiv (no longer necessary as of 5.2.0)
    • ./configure
    • ./configure --prefix=/usr/local
    • ./configure --prefix=$HOME/pkgs
    • ./configure CC=clang
    • ./configure --disable-shared (to make a statically linked executable)
    • ./configure 'CFLAGS=-Wall -std=gnu99 -O3'
    • etc.
  • make creates the c/mlr executable
  • make check
  • make install copies the c/mlr executable to your prefix’s bin subdirectory.

From git clone using autoconfig

  • git clone
  • cd miller
  • Then continue from “Install the following ... ” as above.

Without using autoconfig

GNU autoconfig is familiar to many users, and indeed plenty of folks won’t bother to use an open-source software package which doesn’t have autoconfig support. And this is for good reason: GNU autoconfig allows us to build software on a wide diversity of platforms. For this reason I’m happy that Miller supports autoconfig.

But, many others (myself included!) find autoconfig confusing: if it works without errors, great, but if not, the ./configure && make output can be exceedingly difficult to decipher. And this also can be a turn-off for using open-source software: if you can’t figure out the build errors, you may just keep walking. For this reason I’m happy that Miller allows you to build without autoconfig. (Of course, if you have any build errors, feel free to contact me at, — or, better, open an issue with “New Issue” at


  • Obtain a release tarball or git clone.
  • cd into the c subdirectory.
  • Edit the INSTALLDIR in
  • To change the C compiler, edit the CC= lines in and dsls/
  • make -f creates the mlr executable and runs unit/regression tests (i.e. the equivalent of both make and make check using autoconfig).
  • make install copies the mlr executable to your install directory.

The is simple: little more than gcc *.c. Customzing is less automatic than autoconfig, but more transparent. I expect this makefile to work with few modifications on a large fraction of modern Linux/BSD-like systems: I’m aware of successful use with gcc and clang, on Ubuntu 12.04 LTS, SELinux, Darwin (MacOS Yosemite), and FreeBSD.

Windows (beta)

Status as of 2017-04-15: Miller is built on Windows using MSYS2: You can install MSYS2 and build Miller from its source code within MSYS2, and then you can use the binary from outside MSYS2. You can also use a precompiled binary.

Precompiled binaries

Miller releases from 5.1.0w onward will have a precompiled Windows binary, in addition to the MacOSX, Linux 32-bit static, and Linux 64-bit precompiled binaries as on previous releases.

Important caveat: due to the issue tracked in, you will need to download three files: the mlr.exe file as well as libpcre-1.dll and libpcreposix-0.dll. This situation is unfortunate and inconvenient; I felt it worthwhile to get a Windows beta out to users (due to demand) while this issue is being resolved.

For example: C:\> mkdir \mbin. Then place libpcreposix-0.dll, libpcre-1.dll, and mlr.exe all into C:\mbin. Then put that into your %PATH%, e.g. with C:\> set PATH=%PATH%;\mbin.

Building from source

You will first need to install MSYS2:

Then, start an MSYS2 shell, e.g. (supposing you installed MSYS2 to C:\msys2\) run C:\msys2\mingw64.exe. Within the MSYS2 shell, you can run the following to install dependent packages:

pacman -Syu
pacman -Su
pacman -S base-devel
pacman -S msys2-devel
pacman -S mingw-w64-x86_64-toolchain
pacman -S mingw-w64-x86_64-pcre
pacman -S msys2-runtime

There are (at present) two manual edits, not yet (as of 2017-04-15) automated into the configuration process. One is to edit c/lib/mlr_arch.h and add #define MLR_ON_MSYS2 on its own line right above #ifdef MLR_ON_MSYS2. The other is to edit c/ and c/unit_test/ and add a space and -lpcreposix after -lm.

Then (still within the MSYS2 shell) proceed as above with autoreconf.fiv, ./configure, and make. There is a unit-test false-negative issue involving the semantics of the mkstemp library routine but a make -k in the c subdirectory has been producing a mlr.exe for me.

Within MSYS2 you can run mlr: simply copy it from the c subdirectory to your desired location somewhere within your MSYS2 $PATH. To run mlr outside of MSYS2, just as with precompiled binaries as described above, you’ll need the two DLLs. One way to do this is to augment your path:

C:\> set PATH=%PATH%;\msys64\mingw64\bin

Another way to do it is to copy the Miller executable and the two DLLs all to the same directory:

C:\> mkdir \mbin
C:\> copy \msys64\mingw64\bin\libpcreposix-0.dll \mbin
C:\> copy \msys64\mingw64\bin\libpcre-1.dll \mbin
C:\> copy \msys64\wherever\you\installed\miller\c\mlr.exe \mbin
C:\> set PATH=%PATH%;\mbin

The Windows port is still beta: please open an issue at if you encounter any problems.

Continuous build using Appveyor

Currently Miller is autorun using Travis on every build ( This was set up by the generous assistance of SikhNerd on Github, tracked in There is a similar continuous-build system for Windows. As of 2017-04-15 is not correctly set up, but in the future, it will auto-build Miller commits and releases.

In case of problems


Required external dependencies

These are necessary to produce the mlr executable.

  • gcc, clang, etc. (or presumably other compilers; please open an issue or send me a pull request if you have information for me about other 21st-century compilers)
  • The standard C library
  • flex
  • automake, autoconf, and libtool, if you build with autoconfig

Optional external dependencies

This documentation pageset is built using Poki:
docs here, source code here. Note that is nothing more than Miller’s doc directory served up by a web server. You’ll need poki if you modify documents, or if you modify the code in a way that affects the documents (there are auto-run snippets inserted into the doc). The best way to discover this is to run make install as above, then run poki in Miller’s doc subdirectory, then run git diff to see if docs were affected by the code mod. (If so, commit and push them.)

Internal dependencies

These are included within the
Miller source tree and do not need to be separately installed (and in fact any separate installation will not be picked up in the Miller build):

Creating a new release: for developers

At present I’m the primary developer so this is just my checklist for making new releases.

In this example I am using version 3.4.0; of course that will change for subsequent revisions.

  • Update version found in mlr --version and man mlr:
  • Create the release tarball and SRPM:
    • On buildbox: autoreconf -fiv && ./configure && make distcheck
    • On buildbox: make SRPM as in
    • On all buildboxes: cd c and make -f mlr.static. Then copy mlr.static to ../mlr.{arch}.
    • Pull back release tarball mlr-3.4.0.tar.gz and SRPM mlr-3.4.0-1.el6.src.rpm from buildbox, and mlr.{arch} binaries from whatever buildboxes.
  • Create the Github release tag:
    • Write the release notes
    • Attach the release tarball, SRPM, and binaries
    • Publish the release
  • Create the release-specific docs:
    • Pull the Github-generated source .tgz from Github to ISP miller-releases directory and unpack and cd into it.
    • cd into doc subdir. Edit template.html’s titleinbody tag from Miller to Miller 3.4.0 and poki-build.
    • ISP upload
    • At the ISP site, update the latest-revision symlink ln -s pub_http_internet/miller-releases/miller-3.4.0 pub_http_internet/miller. (This is the main doc URL linked to from the top of the Github repo so it must point to this newly published revision.)
    • Back in the Github repo, edit doc/content-for-release-docs.html to include new release, poki-build, push/commit, and ISP upload.
  • Notify:
    • Submit brew pull request; notify any other distros which don’t appear to have autoupdated since the previous release.
      git remote add upstream # one-time setup only
      git fetch upstream
      git rebase upstream/master
      git checkout -b miller-3.4.0
      shasum -a 256 /path/to/mlr-3.4.0.tar.gz
      edit Formula/miller.rb
      git add Formula/miller.rb
      git commit -m 'miller 3.4.0'
      git push -u origin miller-3.4.0
      (submit the pull request)
    • Social-media updates.
  • Afterwork:
    • Edit and c/mlrvers.h to change version from 3.4.0 to 3.4.0-dev.
    • make -C c -f installhome && make -C doc -f all installhome
    • Commit and push.

Misc. development notes

I use terminal width 120 and tabwidth 4.