@oat-sa/tao-calculator

v0.7.1

Published

13 days ago

A calculator's engine for TAO

Downloads

1,309

tao-calculator

A calculator's engine for TAO.

Calculator's engine overview

The calculator's engine relies on a formal parser for evaluating mathematical expressions. It respects the mathematical order of operations. Mathematical precision is handled by an arbitrary precision library which avoid the well known round-off issue.

Requirements

node.js version 14 and above

Usage

Add the dependency by running

npm i @oat-sa/tao-calculator

Then import the engine:

import { engineFactory } from '@oat-sa/tao-calculator';

const calculator = engineFactory({
    expression: '', // The current expression
    position: 0, // The current position in the expression
    instant: false, // Should the engine compute the expression as soon as an operation is complete?
    corrector: false, // Should the engine auto-correct incomplete expression?
    variables: {}, // A list of variables
    commands: {}, // A list of commands that can be invoked
    plugins: {}, // A list of plugins to load
    maths: {} // Additional config for the internal maths evaluator
});

See the Configuration section for more info.

API

The API is described in the dedicated page.

The engine emits events for each noticeable operation applied to the calculator. In other words, each time a modification is applied, a related event is emitted.

The expression is managed internally through a list of extracted tokens. For easing the expression management, the numbers are tokenized digit by digit. This helps adding or removing elements up to a single digit or symbol.

The engine can be extended through two distinct mechanisms: commands and plugins.

It is also accompanied by a set of helpers for manipulating expression terms.

Project's structure

├── .github
│   └── workflows       # github action workflows
├── build               # standalone package generated
├── dist                # bundle generated from the source code
├── sandbox             # sandbox for playing with the engine locally
├── src                 # root of the source files
│   ├── core            # source files for the core engine
│   │   └── strategies  # a set of strategies for internal features
│   ├── plugins         # a set of plugins that can be added on top of the engine
│   ├── utils           # a set of utilities
│   └── index.js        # the engine entry point
├── *.config.js         # shared tools configurations
├── API.md
├── HISTORY.md
├── LICENSE
├── package.json
├── package-lock.json
└── README.md

Third-party libraries

The calculator engine relies on several third-party libraries, which are listed here.

Parser

The expression is parsed and evaluated by expr-eval, which was forked in a OAT repository as we needed to extend its abilities to ease the implementation of the Nth root function.

The purpose of this fork was to bring the ability to use any two-entry functions as a regular binary operator, with the same operator precedence than the function. The feature is relying on a prefix that transforms the function into an operator.

This parser accepts mathematical expressions, including functions, and is able to process them. It also provides a way to add custom functions, and, last but not least, it accepts to replace the existing implementation of built-in operators and functions by an arbitrary version.

This library is used to parse and evaluate the expressions built inside the calculator, so it is a very important piece in the architecture. The precedence of operators is managed by the parser, as well as the operations are evaluated and computed. The parser can be seen as a function accepting expressions and returning the result of this expression once evaluated.

Expressions syntax

The parser expects expressions to be ASCII strings, with the following rules:

Numerical values are expressed with digits, the decimal separator being the dot .: 3.1415.
Scientific notation using exponent part is allowed: 3.1415e10.
Regular arithmetic operators are supported, with natural precedence: +, -, *, /, !. Operators can be either unary or binary, depending on their meaning.
Expressions can be wrapped in parenthesis: 3*(5+2).
Terms are separated by spaces, however regular operators and parenthesis also act as separators.
Functions and constants are supported: cos PI.
Variables are supported, but require to be defined: x=10, 3 * x.
Any functions can be turned into a binary operator by prefixing them with a @: 4 @nthrt 16 is equivalent to nthrt(4, 16). Obviously, only functions that accept two parameters can work properly with this trick.

Numbers

In order to avoid the well known round-off issue, the default number representation has been replaced in the parser by an arbitrary precision implementation. We chose Decimal.js, which is a good compromise regarding precision and performances. It exposes a comprehensive API to manipulate and compute decimal numbers.

As the parser in use is not resilient against the round-off issue, every built-in operation is replaced by the equivalent provided by Decimal.js. This allows us to support a better numerical precision, but it also means the result of the parser is now a Decimal object instead of a native Number.

To ensure the parser is still working well after having replaced its built-in computation functions, some wrappers have been added. They convert the data type on the fly if required. The benefit is that we can feed the parser with native types and rely on the wrappers to always ensure a compatible format. See the implementation in mathsEvaluator.

Lexer

Manipulating the expression requires to be able to recognize its elements, and to do so the calculator component is relying on a lexer, that is able to cut the expression into atomic and clearly identifiable parts.

The tokenization relies on moo, which is a well supported and offers an optimized tokenizer. It accepts rational expressions to define the patterns it is meant to recognize.

It is worth mentioning this tokenizer is not utilized by the parser to recognize the expression, as the parser is bringing its own way to do that. The purpose of this tokenizer, inside the calculator, is to bring the ability to identify the elements of the expression, in order to manipulates them separately, up to each digit of a number. Among other things, this allows to:

display properly each element with respect to their meaning (mathematical symbols or special tokens)
navigate freely between elements, and add or remove them easily, including each digit of a number
apply some business logic on top of them (for instance change the sign of the current operand, or recognize exponents)

Development

Setup

Clone the repository

git clone [email protected]:oat-sa/tao-calculator-fe.git

Install the package with npm:

cd tao-calculator-fe
npm ci

Useful Commands

To run the sandbox:

npm run dev

To run unit tests:

npm test

To update the test snapshots:

npm run test:update

To run unit tests while developing:

npm run test:watch

To run unit tests with coverage report:

npm run test:cov

All commands

npm run dev: run a sandbox and watch for changes. It also opens it in the browser
npm run test: run the test suite
npm run test <testname>: run a test suite
- testname (optional): Specific test to run. If it is not provided, all will be ran.
npm run test:update: run tests and update the snapshots
npm run test:watch: run tests after each change in the code
npm run test:cov: run tests and report the code coverage in the terminal
npm run coverage:html: show coverage report in browser
npm run coverage:clover: build a code coverage report for continuous integration
npm run build: build for production into dist directory
npm run build:watch: build for production into dist directory and watch for changes
npm run lint: check syntax of code
npm run lint:report: build a syntax check report
npm run format: correct the code style

Known issues and drawbacks

Thanks to the numbers representation engine, the calculator is able to give a good computation precision. However, due to the nature of computation, some known issues are still there, and cannot be simply addressed.

Loss of precision of inlined variables

Internally, the computed value have hundreds of decimal digits, and only a few is displayed. So, if a variable is inlined, i.e. its displayed value is used instead of the variable itself, the result won't be accurate. For instance, compute the square root of 2. If you immediately elevate it at a power of 2, you will retrieve the former value, said 2. However, if you take the displayed value and elevate it to the power of 2, the result will be different, and could be considered wrong if you expected to retrieve the former value.

Loss of precision in some irrational numbers

Mathematically, it is impossible to have a bijective computation with the inverse of 3: 1/3 gives 0.3333333333333, and we can continue indefinitely with the 3 after the decimal point. Now if you multiply this value by 3, no matter the amount of 3 you will add after the decimal point, you will never retrieve 1, but 0.9999999999 instead. A mathematical trick is to add a 4 as a last digit, but unfortunately this won't work with the calculator's engine. In fact, it can only work by doing: 0.333333333 + 0.333333333 + 0.333333334. Which is not the same.

History

Changes are detailed in the history page.

This repository has been migrated on 2023/01/31 from:

Please consult those repositories for history prior to this date.

License

Licensed under the terms of the GNU GPL v2