Skip to main content

Abysmal (Appallingly Basic Yet Somehow Mostly Adequate Language)

Project description

https://travis-ci.org/zillow/abysmal.svg?branch=master https://codecov.io/gh/zillow/abysmal/branch/master/graph/badge.svg

Abysmal stands for “appallingly basic yet somehow mostly adequate language”.

Abysmal is a programming language designed to allow non-programmers to implement simple business logic for computing prices, rankings, or other kinds of numeric values without incurring the security and stability risks that would normally result when non-professional coders contribute to production code. In other words, it’s a sandbox in which businesspeople can tinker with their business logic to their hearts’ content without involving your developers or breaking anything.

Features

  • Supports Python 3.3 and above

Dependencies

  • python3-dev native library including Python C header files

  • libmpdec-dev native library for decimal arithmetic

Language Reference

Abysmal programs are designed to be written by businesspeople, so the language foregoes almost all the features programmers want in a programming language in favor of mimicking something business people understand: flowcharts.

Just about the only way your businesspeople can “crash” an Abysmal program is by dividing by zero, because:

  • it’s not Turing-complete

  • it can’t allocate memory

  • it can’t access the host process or environment

  • it operates on one and only one type: arbitrary-precision decimal numbers

  • its only control flow construct is GOTO

  • it doesn’t even allow loops!

Example program

# input variables:
#
#    flavor:         VANILLA, CHOCOLATE, or STRAWBERRY
#    scoops:         1, 2, etc.
#    cone:           SUGAR or WAFFLE
#    sprinkles:      0 or 1
#    weekday:        MONDAY, TUESDAY, WEDNESDAY, THURSDAY, FRIDAY, SATURDAY, or SUNDAY

# output variables:
#
#    price:          total price, including tax

let TAX_RATE = 5.3%
let WEEKDAY_DISCOUNT = 25%
let GIVEAWAY_RATE = 1%

@start:
    random! <= GIVEAWAY_RATE => @giveaway_winner
    price = scoops * (flavor == STRAWBERRY ? 1.25 : 1.00)
    price = price + (cone == WAFFLE ? 1.00 : 0.00)
    price = price + (sprinkles * 0.25)
    weekday not in {SATURDAY, SUNDAY} => @apply_weekday_discount
    => @compute_total

@apply_weekday_discount:
    price = price * (1 - WEEKDAY_DISCOUNT)
    => @compute_total

@giveaway_winner:
    price = 0.00

@compute_total:
    price = price * (1 + TAX_RATE)

Control flow

An Abysmal program models a flowchart containing one or more steps, or states. Program execution begins at the beginning of the first state and continues until it reaches a dead end. Along the way, variables can be assigned new values, and execution can jump to other states. That’s it.

Every state has a name that starts with @. A state is declared like this:

@start:

A state declaration is followed by a sequence of actions. Each action appears on its own line, and is one of the following:

  1. an assignment of a value to a variable, like this:

price = scoops * flavor == STRAWBERRY ? 1.25 : 1.00
  1. a conditional jump to another state, like this:

weekday not in {SATURDAY, SUNDAY} => @apply_weekday_discount
  1. an unconditional jump to another state, like this:

=> @compute_total

When execution reaches a state, that state’s actions are executed in order. If execution reaches the end of a state without jumping to a new state, the program exits.

Programs are not allowed to contain loops or any other exeuction cycles. Any program containing a cycle will fail to compile.

Actions are typically indented to make the state labels easier to see, but this is just a stylistic convention and is not enforced by the language.

Comments

Anything following a # on a line is treated as a comment and is ignored.

Line continuations

A \ at the end of a line indicates that the next line is a continuation of the line. This makes it easy to format long lines readably by splitting them into multiple, shorter lines. Note that comments can appear after a \.

Numbers

Abysmal supports integer and fixed-point decimal literals like 123, 3.14159, etc. In addition, numbers can have the following suffixes:

suffix

meaning

%

percent (12.5% is equivalent to 0.125)

k or K

thousand (50k is equivalent to 50000)

m or M

million (1.2m is equivalent to 1200000)

b or B

billion (0.5b is equivalent to 500000000)

Scientific notation is not supported.

Booleans

Abysmal uses 1 and 0 to represent the result of any operation that yields a logical true/false value. When evaluating conditions in a conditional jump or a ? expression, zero is considered false and any non-zero value is considered true.

Expressions

Programs can evaluate expressions containing the following operators:

operator

precedence

meaning

example

( exp )

0 (highest)

grouping

(x + 1) * y

!

1

logical NOT

!x

+

1

unary plus (has no effect)

+x

-

1

unary minus

-x

^

2

exponentiation (right associative)

x ^ 3

*

3

multiplication

x * 100

/

3

division

x / 2

+

4

addition

x + 5

-

4

subtraction

x - 3

in { exp, … }

5

is a member of the set

x in {0, y, -z}

not in { exp, … }

5

is not a member of the set

x not in {0, y, -z}

in [ low , high ]

5

falls within the interval (see Intervals)

x in [-3, 7]

not in [ low , high ]

5

does not fall within the interval

x not in [-3, 7]

<

6

is less than

x < y

<=

6

is less than or equal to

x <= y

>

6

is greater than

x > y

>=

6

is greater than or equal to

x >= y

==

7

is equal to

x == y

!=

7

is not equal to

x != y

&&

8

logical AND (short-circuiting)

x && (y / x > 0.8)

||

9

logical OR (short-circuiting)

x > 3 || y > 7

exp ? exp : exp

10 (lowest)

if-then-else

x < 0 ? -x : x

Intervals

Intervals support inclusive endpoints (specified with square brackets) and exclusive endpoints (specified with parentheses), and the two can be freely mixed. For example, the follwing are all valid checks:

  • x in (0, 1)

  • x in (0, 1]

  • x in [0, 1)

  • x in [0, 1]

Note that “backwards” intervals (where the first endpoint is greater than the second) are considered pathological and treated as empty. Therefore 2 in (1, 3) evaluates to 1 (aka true), but 2 in (3, 1) evaluates to 0 (aka false).

Functions

Expressions can take advantage of the following built-in functions:

function

returns

ABS(exp)

the absolute value of the specified value

CEILING(exp)

the nearest integer value greater than or equal to the specified value

FLOOR(exp)

the nearest integer value less than or equal to the specified value

MAX(exp1, exp2, …)

the maximum of the specified values

MIN(exp1, exp2, …)

the minimum of the specified values

ROUND(exp)

the specified value, rounded to the nearest integer

Variables

Abysmal programs can read from and write to variables that you define when you compile the program. Some of these variables will be inputs, whose values you will set before you run the program. Others will be outputs, whose values the program will compute so that those values can be examined after the program has terminated.

Abysmal does not distinguish between input and output variables.

All variables and constant values are decimal numbers. Abysmal does not have any concept of strings, booleans, null, or any other types.

If not explicitly set, variables default to 0.

random! is a special, read-only variable that yields a new, random value every time it is referenced.

You can also provide named constants to your programs when you compile them. Constants cannot be modified.

A program can also declare custom variables that it can use to store intermediate results while the model is being run, or simply to define friendlier names for values that are used within the model. Custom variables must be declared before the first state is declared.

Each custom variable is declared on its own line, like this:

let PI = 3.14159
let area = PI * r * r

Usage

An Abysmal program must be compiled before it can be run. The compiler needs to know the names of the variables that the program should have access to and names and values of any constants you want to define:

ICE_CREAM_VARIABLES = {
    # inputs
    'flavor',
    'scoops',
    'cone',
    'sprinkles',
    'weekday',

    # outputs
    'price',
}

ICE_CREAM_CONSTANTS = {
    # flavors
    'VANILLA': 1,
    'CHOCOLATE': 2,
    'STRAWBERRY': 3,

    # cones
    'SUGAR': 1,
    'WAFFLE': 2,

    # weekdays
    'MONDAY': 1,
    'TUESDAY': 2,
    'WEDNESDAY': 3,
    'THURSDAY': 4,
    'FRIDAY': 5,
    'SATURDAY': 6,
    'SUNDAY': 7,
}

compiled_program, source_map = abysmal.compile(source_code, ICE_CREAM_VARIABLES, ICE_CREAM_CONSTANTS)

Ignore the second value returned by abysmal.compile() for now (refer to the Measuring Coverage section to see what it’s useful for).

Next, we need to make a virtual machine for the compiled program to run on:

machine = compiled_program.machine()

Next, we can set any variables as we see fit:

# Variables can be set in bulk during reset()...
machine.reset(
    flavor=ICE_CREAM_CONSTANTS['CHOCOLATE'],
    scoops=2,
    cone=ICE_CREAM_CONSTANTS['WAFFLE']
)

# ... or one at a time (though this is less efficient)
machine['sprinkles'] = True  # automatically converted to '1'

Finally, we can run the machine and examine final variable values:

price = Decimal('0.00')
try:
    machine.run()
    price = round(Decimal(machine['price']), 2)
except abysmal.ExecutionError as ex:
    print('The ice cream pricing algorithm is broken: ' + str(ex))
else:
    print('Two scoops of chocolate ice cream in a waffle cone with sprinkles costs: ${0}'.format(price))

Note that the virtual machine exposes variable values as strings, which may be formatted in scientific or fixed-point notation.

Variables can be set from int, float, bool, Decimal, and string values but are converted to strings when assigned. When examining variables after running a machine, you need to convert to the values back to Decimal, float, or whatever numeric type you are interested in.

Random Numbers

By default, random! generates numbers between 0 and 1 with 9 decimal places of precision, and uses the default Python PRNG (random.randrange).

If you require a more secure PRNG, or different precision, or if you want to force certain values to be produced for testing purposes, you can supply your own random number iterator before running a machine:

# force random! to yield 0, 1, 0, 1, ...
machine.random_number_iterator = itertools.cycle([0, 1])

The values you return are not required to fall within any particular range, but [0, 1] is recommended, for consistency with the default behavior.

Limits

Decimal values are constrained in accordance with the IEEE 754 decimal128 format. This provides 34 digits of precision and an exponent range of -6143 to +6144.

Infinity, negative infinity, and NaN (not-a-number) are not allowed. Calculations that would give rise to one of these will instead trigger an error.

In addition, a calculation can result in overflow or underflow if its result is too large or too small to fit into the decimal128 range.

Errors

abysmal.CompilationError

raised by abysmal.compile() if the source code cannot be compiled

abysmal.ExecutionError

raised by machine.run() and machine.run_with_coverage() if a program encounters an error while running; this includes conditions such as: division by zero, invalid exponentiation, stack overflow, floating-point overflow, floating-point underflow, out-of-space, and failure to generate a random number

abysmal.InstructionLimitExceededError

raised by machine.run() and machine.run_with_coverage() if a program exceeds its allowed instruction count and is aborted; this error is a subclass of abysmal.ExecutionError

Performance Tips

Abysmal programs run very quickly once compiled, and the virtual machine is optimized to make repeated runs with different inputs as cheap as possible.

As always, decide on your performance goals and measure before optimizing.

To get the best performance, follow these tips:

Avoid recompilation

Compiling a program is orders of magnitude slower than actually running it.

Save the compiled program and reuse it rather than recompiling every time. Compiled programs are pickleable, so they are easy to cache.

Use baseline images

When you create a machine, you can pass keyword arguments to set the machine’s variables to initial values. The state of the variables at this moment is called a baseline image. When you reset a machine, it restores all variables to the baseline image very efficiently. Therefore, if you are going to run a particular program repeatedly with some inputs having the same values for all the runs, you should specify those input values in the baseline.

For example:

def compute_shipping_costs(product, weight, zip_codes, compiled_program):
    shipping_costs = {}
    machine = compiled_program.machine(product=product, weight=weight)
    for zip_code in zip_codes:
        machine.reset(zip=zip_code).run()
        shipping_costs[zip_code] = round(Decimal(machine['shippingCost']), 2)
    return shipping_costs

Set multiple variables at once

Override baseline variable values by passing keywords to machine.reset() rather than assigning variables one-by-one. The overhead of making multiple Python function calls is non-trivial if your scenario needs performance!

Only read and write variables you need

Initializing variables before a program runs and reading variables afterwards can easily add up to more time it takes to actually run a typical program. If performance is critical for your scenario, you can save time by only examining variables whose values you really need.

Limit instruction execution

Since Abysmal does not support loops, it is very difficult to create a program that runs for very long. However, you can impose an additional limit on the number of instructions that a program can execute by setting the instruction_limit attribute of a machine:

machine.instruction_limit = 5000

If a program exceeds its instruction limit, it will raise an abysmal.InstructionLimitExceededError.

The default instruction limit is 10000.

The run() method returns the number of instructions that were run before the program exited.

Measuring Coverage

In addition to run(), virtual machines expose a run_with_coverage() method which can be used in conjunction with the source map returned by abysmal.compile() to generate coverage reports for Abysmal programs.

coverage_tuples = [
    machine.reset(**test_case_inputs).run_with_coverage()
    for test_case_inputs in test_cases
]
coverage_report = abysmal.get_uncovered_lines(source_map, coverage_tuples)
print('Partially covered lines: ' + ', '.join(map(str, coverage_report.partially_covered_line_numbers)))
print('Totally uncovered lines: ' + ', '.join(map(str, coverage_report.uncovered_line_numbers))

How coverage works:

run_with_coverage() returns a coverage tuple whose length is equal to the number of instructions in the compiled program. The value at index i in the coverage tuple will be True or False depending on whether instruction i was executed during the program’s run.

The source map is another tuple, with the same length as the coverage tuple. The value at index i in the source map indicates which line or lines in the source code generated instruction i of the compiled program. There are three possibilities:

  • None - the instruction was not directly generated by any source line

  • int - the instruction was generated by a single source line

  • (int, int, …) - the instruction was generated by multiple source lines (due to line continuations being used)

Installation

Note that native library dependencies must be installed BEFORE you install the abysmal library.

pip install abysmal

Development

# Install system-level dependencies on Debian/Ubuntu
make setup

# Run unit tests
make test

# Check code cleanliness
make pylint

# Check code coverage
make cover

# Create sdist package
make package

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

abysmal-1.2.0.tar.gz (43.7 kB view hashes)

Uploaded Source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page