miller/docs6/docs/programming-language.md.in

# Intro to Miller's programming language

In the [Miller in 10 minutes](10min.md) page we took a tour of some of Miller's most-used [verbs](reference-verbs.md) including `cat`, `head`, `tail`, `cut`, and `sort`. These are analogs of familiar system commands, but empowered by field-name indexing and file-format awareness: the system `sort` command only knows about lines and column names like `1,2,3,4`, while `mlr sort` knows about CSV/TSV/JSON/etc records, and field names like `color,shape,flag,index`.

We also caught a glimpse of Miller's `put` and `filter` verbs. These two are special since they let you express statements using Miller's programming language. It's a *embedded domain-specific language* since it's inside Miller: often referred to simply as the *Miller DSL*.

In the [DSL reference](reference-dsl.md) page we have a complete reference to Miller's programming language. For now, let's take a quick look at key features -- you can use as few or as many features as you like.

## Records and fields

Let's keep using the [example.csv](./example.csv) file:

GENMD_RUN_COMMAND
mlr --c2p put '$cost = $quantity * $rate' example.csv
GENMD_EOF

When we type that, a few things are happening:

* We refer to fields in the input data using a dollar sign and then the field name, e.g. `$quantity`. (If a field name contains special characters like a dot or slash, just use curly braces: `${field.name}`.)
* The expression `$cost = $quantity * $rate` is executed once per record of the data file. Our [example.csv](./example.csv) has 10 records so this expression was executed 10 times, with the field names `$quantity` and `$rate` each time bound to the current record's values for those fields.
* On the left-hand side we have the new field name `$cost` which didn't come from the input data. Assignments to new variables result in a new field being placed after all the other ones. If we'd assigned to an existing field name, it would have been updated in-place.
* The entire expression is surrounded by single quotes (with an adjustment needed on [Windows](miller-on-windows.md)), to get it past the system shell. Inside those, only double quotes have meaning in Miller's programming language.

## Multi-line statements, and statements-from-file

You can use more than one statement, separating them with semicolons, and optionally putting them on lines of their own:

GENMD_RUN_COMMAND
mlr --c2p put '$cost = $quantity * $rate; $index = $index * 100'  example.csv
GENMD_EOF

GENMD_RUN_COMMAND
mlr --c2p put '
  $cost = $quantity * $rate;
  $index *= 100
' example.csv
GENMD_EOF

One of Miller's key features is the ability to express data-transformation right there at the keyboard, interactively. But if you find yourself using expressions repeatedly, you can put everything between the single quotes into a file and refer to that using `put -f`:

GENMD_RUN_COMMAND
cat dsl-example.mlr
GENMD_EOF

GENMD_RUN_COMMAND
mlr --c2p put -f dsl-example.mlr example.csv
GENMD_EOF

This becomes particularly important on Windows. Quite a bit of effort was put into making Miller on Windows be able to handle the kinds of single-quoted expressions we're showing here, but if you get syntax-error messages on Windows using examples in this documentation, you can put the parts between single quotes into a file and refer to that using `mlr put -f` -- or, use the triple-double-quote trick as described in the [Miller on Windows page](miller-on-windows.md).

## Out-of-stream variables, begin, and end

Above we saw that your expression is executed once per record -- if a file has a million records, your expression will be executed a million times, once for each record. But you can mark statements to only be executed once, either before the record stream begins, or after the record stream is ended. If you know about [AWK](https://en.wikipedia.org/wiki/AWK), you might have noticed that Miller's programming language is loosely inspired by it, including the `begin` and `end` statements.

Above we also saw that names like `$quantity` are bound to each record in turn.

To make `begin` and `end` statements useful, we need somewhere to put things that persist across the duration of the record stream, and a way to emit them. Miller uses [**out-of-stream variables**](reference-dsl-variables.md#out-of-stream-variables) (or **oosvars** for short) whose names start with an `@` sigil, along with the [`emit`](reference-dsl-output-statements.md#emit-statements) keyword to write them into the output record stream:

GENMD_RUN_COMMAND
mlr --c2p --from example.csv put 'begin { @sum = 0 } @sum += $quantity; end {emit @sum}'
GENMD_EOF

If you want the end-block output to be the only output, and not include the records from the input data, you can use `mlr put -q`:

GENMD_RUN_COMMAND
mlr --c2p --from example.csv put -q 'begin { @sum = 0 } @sum += $quantity; end {emit @sum}'
GENMD_EOF

GENMD_RUN_COMMAND
mlr --c2j --from example.csv put -q 'begin { @sum = 0 } @sum += $quantity; end {emit @sum}'
GENMD_EOF

GENMD_RUN_COMMAND
mlr --c2j --from example.csv put -q '
  begin { @count = 0; @sum = 0 }
  @count += 1;
  @sum += $quantity;
  end {emit (@count, @sum)}
'
GENMD_EOF

We'll see in the documentation for [stats1](reference-verbs.md#stats1) that there's a lower-keystroking way to get counts and sums of things:

GENMD_RUN_COMMAND
mlr --c2j --from example.csv stats1 -a sum,count -f quantity
GENMD_EOF

So, take this sum/count example as an indication of the kinds of things you can do using Miller's programming language.

## Context variables

Also inspired by [AWK](https://en.wikipedia.org/wiki/AWK), the Miller DSL has the following special [**context variables**](reference-dsl-variables.md#built-in-variables):

* `FILENAME` -- the filename the current record came from. Especially useful in things like `mlr ... *.csv`.
* `FILENUM` -- similarly, but integer 1,2,3,... rather than filenam.e
* `NF` -- the number of fields in the current record. Note that if you assign `$newcolumn = some value` then `NF` will increment.
* `NR` -- starting from 1, counter of how many records processed so far.
* `FNR` -- similar, but resets to 1 at the start of each file.

GENMD_RUN_COMMAND
cat context-example.mlr
GENMD_EOF

GENMD_RUN_COMMAND
cat data/a.csv
GENMD_EOF

GENMD_RUN_COMMAND
cat data/b.csv
GENMD_EOF

GENMD_RUN_COMMAND
mlr --c2p put -f context-example.mlr data/a.csv data/b.csv
GENMD_EOF

## Functions and local variables

You can [define your own functions](reference-dsl-user-defined-functions.md):

GENMD_RUN_COMMAND
cat factorial-example.mlr
GENMD_EOF

GENMD_RUN_COMMAND
mlr --c2p --from example.csv put -f factorial-example.mlr -e '$fact = factorial(NR)'
GENMD_EOF

Note that here we used the `-f` flag to `put` to load our function
definition, and also the `-e` flag to add another statement on the command
line. (We could have also put `$fact = factorial(NR)` inside
`factorial-example.mlr` but that would have made that file less flexible for our
future use.)

## If-statements, loops, and local variables

Suppose you want to only compute sums conditionally -- you can use an `if` statement:

GENMD_RUN_COMMAND
cat if-example.mlr
GENMD_EOF

GENMD_RUN_COMMAND
mlr --c2p --from example.csv put -q -f if-example.mlr
GENMD_EOF

Miller's else-if is spelled `elif`.

As we'll see more of in the [control-structures reference
page](reference-dsl-control-structures.md#for-loops), Miller has a few kinds of
for-loops. In addition to the usual 3-part `for (i = 0; i < 10; i += 1)` kind
that many programming languages have, Miller also lets you loop over
[maps](reference-main-maps.md) and [arrays](reference-main-arrays.md). We
haven't encountered maps and arrays yet in this introduction, but for now it
suffices to know that `$*` is a special variable holding the current record as
a map:

GENMD_RUN_COMMAND
cat for-example.mlr
GENMD_EOF

GENMD_RUN_COMMAND
mlr --csv cat data/a.csv
GENMD_EOF

GENMD_RUN_COMMAND
mlr --csv --from data/a.csv put -qf for-example.mlr
GENMD_EOF

Here we used the local variables `k` and `v`. Now we've seen four kinds of variables:

* Record fields like `$shape`
* Out-of-stream variables like `@sum`
* Local variables like `k`
* Built-in context variables like `NF` and `NR`

If you're curious about scope and extent of local variables, you can read more in the [section on variables](reference-dsl-variables.md).

## Arithmetic

Numbers in Miller's programming language are intended to operate with the principle of least surprise:

* Internally, numbers are either 64-bit signed integers or double-precision floating-point.
* Sums, differences, and products of integers are also integers (so `2*3=6` not `6.0`) -- unless the result of the operation would overflow a 64-bit signed integer in which case the result is automatically converted to float. (If you ever want integer-to-integer arithmetic, use `x .+ y`, `x .* y`, etc.)
* Quotients of integers are integers if the division is exact, else floating-point:  so `6/2=3` but `7/2=3.5`.

You can read more about this in the [arithmetic reference](reference-main-arithmetic.md).

## Absent data

In addition to types including string, number (int/float), maps, and arrays,
Miller varibles can also be **absent**. This is when a variable never had a
value assigned to it. Miller's treatment of absent data is intended to make it
easy for you to handle non-heterogeneous data. We'll see more in the [null-data
reference](reference-main-null-data.md) but the basic idea is:

* Adding a number to absent gives the number back. This means you don't have to put `@sum = 0` in your `begin` blocks.
* Any variable which has the absent value is not assigned. This means you don't have to check presence of things from one record to the next.

For example, you can sum up all the `$a` values across records without having to check whether they're present or not:

GENMD_RUN_COMMAND
mlr --json cat absent-example.json
GENMD_EOF

GENMD_RUN_COMMAND
mlr --json put '@sum_of_a += $a; end {emit @sum_of_a}' absent-example.json
GENMD_EOF