Skip to content
Quick links:   Flags   Verbs   Functions   Glossary   Release docs


The Miller REPL (read-evaluate-print loop) is an interactive counterpart to record-processing using the put/filter language. (A REPL is anything that evaluates what you type into it -- like python with no arguments, or Ruby's irb, or node with no arguments, etc.)

Miller's REPL isn't a source-level debugger which lets you execute one source-code statement at a time -- however, it does let you operate on one record at a time. Further, it lets you use "immediate expressions", namely, you can interact with the Miller programming language without having to provide data from an input file.

mlr repl

[mlr] 1 + 2

[mlr] apply([1,2,3,4,5], func(e) {return e ** 3})
[1, 8, 27, 64, 125]

[mlr] :open example.csv

[mlr] :read

[mlr] $*
  "color": "yellow",
  "shape": "triangle",
  "flag": "true",
  "k": 1,
  "index": 11,
  "quantity": 43.6498,
  "rate": 9.8870

Using Miller without the REPL

Using put and filter, you can do the following as we've seen above:

  • Specify input format (e.g. --icsv), output format (e.g. --ojson), etc. using command-line flags.
  • Specify filenames on the command line.
  • Define begin {...} blocks which are executed before the first record is read.
  • Define end {...} blocks which are executed after the last record is read.
  • Define user-defined functions/subroutines using func and subr.
  • Specify statements to be executed on each record -- which are anything outside of begin/end/func/subr.
  • Example:
mlr --icsv --ojson --from example.csv head -n 2 \
  then put 'begin {print "HELLO"} $qr = $quantity / $rate; end {print "GOODBYE"}'
  "color": "yellow",
  "shape": "triangle",
  "flag": "true",
  "k": 1,
  "index": 11,
  "quantity": 43.6498,
  "rate": 9.8870,
  "qr": 4.414868008496004
  "color": "red",
  "shape": "square",
  "flag": "true",
  "k": 2,
  "index": 15,
  "quantity": 79.2778,
  "rate": 0.0130,
  "qr": 6098.292307692308

Using Miller with the REPL

Using the REPL, by contrast, you get interactive control over those same steps:

  • Specify input format (e.g. --icsv), output format (e.g. --ojson), etc. using command-line flags.
  • REPL-only statements (non-DSL statements) start with :, such as :help or :quit or :open.
  • Specify filenames either on the command line or via :open at the Miller REPL.
  • Read records one at a time using :read.
  • Write the current record (maybe after you've modified it with things like $z = $x + $y) using :write. This goes to the terminal; you can use :> {filename} to make writes go to a file, or :>> {filename} to append.
  • You can type :reopen to go back to the start of the same file(s) you specified with :open.
  • Skip ahead using statements :skip 10 or :skip until NR == 100 or :skip until $status_code != 200.
  • Similarly, but processing records rather than skipping past them, using :process rather than :skip. Like :write, these go to the screen; use :> {filename} or :>> {filename} to log to a file instead.
  • Define begin {...} blocks; invoke them at will using :begin.
  • Define end {...} blocks; invoke them at will using :end.
  • Define user-defined functions/subroutines using func/subr; call them from other statements.
  • Interactively specify statements to be executed immediately on the current record.
  • Load any of the above from Miller-script files using :load.

The input "record" by default is the empty map but you can do things like $x=3, or unset $y, or $* = {"x": 3, "y": 4} to populate it. Or, :open foo.dat followed by :read to populate it from a data file.

Non-assignment expressions, such as 7 or true, operate as filter conditions in the put DSL: they can be used to specify whether a record will or won't be included in the output-record stream. But here in the REPL, they are simply printed to the terminal, e.g. if you type 1+2, you will see 3.

Entering multi-line statements

  • To enter multi-line statements, enter < on a line by itself, then the code (taking care for semicolons), then > on a line by itself. These will be executed immediately.
  • If you enter << on a line by itself, then the code, then >> on a line by itself, the statements will be remembered for executing on records with :main, as if you had done :load to load statements from a file.


Use the REPL to look at arithmetic:

mlr repl

[mlr] 6/3

[mlr] 6/5

[mlr] typeof(6/3)

[mlr] typeof(6/5)

Read the first record from a small file:

mlr repl

[mlr] :open foo.dat

[mlr] :read

[mlr] :context

[mlr] $*
  "a": "eks",
  "b": "wye",
  "i": 4,
  "x": 0.38139939387114097,
  "y": 0.13418874328430463

[mlr] $z = $x + $i

[mlr] :write

Skip until deep into a larger file, then inspect a record:

mlr repl --csv

[mlr] :open data/colored-shapes.csv
[mlr] :skip until NR == 10000
[mlr] :r
[mlr] $*
  "color": "yellow",
  "shape": "circle",
  "flag": 1,
  "i": 496422,
  "u": 0.6530503199545348,
  "v": 0.23908588907834516,
  "w": 0.4799125551304738,
  "x": 6.379888206335166


No command-line-history-editing feature is built in but rlwrap mlr repl is a delight. You may need brew install rlwrap, sudo apt-get install rlwrap, etc. depending on your platform.

Suggestion: alias mrpl='rlwrap mlr repl' in your shell's startup file.

Online help

After mlr repl, type :help to see more about your options. In particular, :help examples.

Back to top