const state = {
// Match values case sensitively when filtering.
caseSensitive: false,
// Coerce values into regular expressions (instead of strings) when they're defined as atoms.
preferRegex: true,
// The key in the JS object that hosts the Date type against which we filter.
dateKey: 'Date',
};
// TODO(wpcarro): Support filtering by date (before, after).
// TODO(wpcarro): Support grouping with parentheses.
function select(query, xs) {
const predicate = compile(parse(query));
return xs.filter(predicate);
}
function compile(ast) {
if (ast.type === 'CONJUNCTION') {
const lhs = compile(ast.lhs);
const rhs = compile(ast.rhs);
if (ast.joint === 'AND') {
return function(x) {
return lhs(x) && rhs(x);
};
}
if (ast.joint === 'OR') {
return function(x) {
return lhs(x) || rhs(x);
};
}
}
if (ast.type === 'DATE_SELECTION') {
if (ast.key === 'before') {
return function(row) {
let t = new Date();
if (ast.val === 'yesterday') {
t.setDate(t.getDate() - 1);
console.log(t);
}
// MM/DD/YYYY
else {
t = new Date(ast.val);
}
return row[state.dateKey] < t;
};
}
if (ast.key === 'after') {
return function(row) {
let t = new Date();
if (ast.val === 'yesterday') {
t.setDate(t.getDate() - 1);
console.log(t);
}
// MM/DD/YYYY
else {
t = new Date(ast.val);
}
return row[state.dateKey] > t;
};
}
}
if (ast.type === 'SELECTION') {
const f = compile(ast.val);
return function(row) {
return ast.negate ? !f(row[ast.key]) : f(row[ast.key]);
};
}
if (ast.type === 'MATCH_ALL') {
if (ast.matchType === 'STRING') {
return function(row) {
return Object.values(row).some(x => {
if (state.caseSensitive) {
return x === ast.val;
} else {
return x.toLowerCase() === ast.val.toLowerCase();
}
})
};
}
if (ast.matchType === 'REGEX') {
return function(row) {
return Object.values(row).some(x => ast.val.test(x));
};
}
}
if (ast.type === 'STRING') {
return function(x) {
if (state.caseSensitive) {
return x === ast.val;
} else {
return x.toLowerCase() === ast.val.toLowerCase();
}
};
}
if (ast.type === 'REGEX') {
return function(x) {
return ast.val.test(x);
};
}
}
// A "selection" without a "$column:" prefix should fuzzy-search all columns.
//
// conjunction -> selection ( ( "AND" | "OR" )? selection )* ;
// selection -> "-"? COLUMN ":" ( regex | string ) | regex ;
// regex -> [_-a-zA-Z0-9] | "/" [ _-a-zA-Z0-9] "/" | string ;
// string -> "\"" [ _-a-zA-Z0-9] "\"" ;
// Whatever characters are valid for a JS regex.
const ATOM_REGEX = /[-_.\[\]a-zA-Z0-9*+^$]/;
function tokenize(x) {
const result = [];
let i = 0;
while (i < x.length) {
if (x[i] === ' ') {
i += 1;
while (i < x.length && x[i] === ' ') {
i += 1;
}
result.push(['WHITESPACE', null]);
continue;
}
if (x[i] === '-') {
result.push(['NEGATE', null]);
i += 1;
continue;
}
if (ATOM_REGEX.test(x[i])) {
let curr = x[i];
i += 1;
while (i < x.length && ATOM_REGEX.test(x[i])) {
curr += x[i];
i += 1;
}
result.push(['ATOM', curr]);
continue;
}
if (x[i] === ':') {
result.push(['COLON', null]);
i += 1;
continue;
}
if (x[i] === '(') {
result.push(['LPAREN', null]);
i += 1;
continue;
}
if (x[i] === ')') {
result.push(['RPAREN', null]);
i += 1;
continue;
}
if (x[i] === '/') {
let start = i;
let curr = '';
i += 1;
while (i < x.length && x[i] !== '/') {
curr += x[i];
i += 1;
}
// error
if (i >= x.length) {
throw `Tokenize Error: EOL while attempting to tokenize the regex beginning at column: ${start}`;
}
if (x[i] === '/') {
result.push(['REGEX', curr]);
i += 1;
}
continue;
}
if (x[i] === '"') {
let start = i;
let curr = '';
i += 1;
while (i < x.length && x[i] !== '"') {
// continue on \"
if (x[i] === '\\' && x[i + 1] === '"') {
curr += '\"';
i += 2;
} else {
curr += x[i];
i += 1;
}
}
if (i >= x.length) {
throw `Tokenize Error: EOL while attempting to tokenize the string starting at column: ${start}`;
}
if (x[i] === '"') {
result.push(['STRING', curr]);
i += 1;
}
continue;
}
else {
i += 1;
}
}
return result;
}
function expect(f, expectation, p) {
const [type, val] = p.tokens[p.i];
if (f(type, val)) {
p.i += 1;
} else {
throw `Parse Error: expected ${expectation}, but got ${p.tokens[p.i]}; ${JSON.stringify(p)}`
}
}
function matches(f, p) {
const [type, val] = p.tokens[p.i];
if (f(type, val)) {
return true;
}
return false;
}
function match(f, expectation, p) {
const [type, val] = p.tokens[p.i];
if (f(type, val)) {
p.i += 1;
return val;
}
throw `Parse Error: expected ${expectation}, but got: ${p.tokens[p.i]}; ${JSON.stringify(p)}`;
}
function skipWhitespace(p) {
while (p.i < p.tokens.length && matches((type, _) => type === 'WHITESPACE', p)) {
p.i += 1;
}
}
function parser(tokens) {
return { i: 0, tokens };
}
function parse(x) {
const tokens = tokenize(x);
const p = parser(tokens);
return conjunction(p);
}
function conjunction(p) {
skipWhitespace(p);
const lhs = selection(p);
skipWhitespace(p);
if (p.i >= p.tokens.length) {
return lhs;
}
let joint = 'AND';
if (matches((type, val) => type === 'ATOM' && val === 'AND', p)) {
joint = 'AND';
p.i += 1;
} else if (matches((type, val) => type === 'ATOM' && val === 'OR', p)) {
joint = 'OR';
p.i += 1;
}
skipWhitespace(p);
let rhs = conjunction(p);
return {
type: 'CONJUNCTION',
joint,
lhs,
rhs,
};
}
function peekType(n, p) {
if (p.i + n < p.tokens.length) {
return p.tokens[p.i + n][0];
}
return null;
}
function selection(p) {
// column:value OR -column:value
if ((peekType(0, p) === 'ATOM' && peekType(1, p) === 'COLON') ||
(peekType(0, p) === 'NEGATE' && peekType(1, p) === 'ATOM' && peekType(2, p) === 'COLON')) {
let negate = false;
if (p.tokens[p.i][0] === 'NEGATE') {
negate = true;
p.i += 1;
}
const key = match((type, _) => type === 'ATOM', 'a column label', p);
expect((type, val) => type === 'COLON', 'a colon', p);
if (key === 'before' || key === 'after') {
const val = date(p);
return {
type: 'DATE_SELECTION',
key,
val,
};
} else {
const val = value(p);
return {
type: 'SELECTION',
negate,
key,
val,
};
}
} else {
return matchAll(p);
}
}
function matchAll(p) {
const [type, val] = p.tokens[p.i];
// Cast atoms into strings or regexes depending on the current state.
if (type === 'ATOM') {
p.i += 1;
if (state.preferRegex) {
const regex = state.caseSensitive ? new RegExp(val) : new RegExp(val, "i");
return { type: 'MATCH_ALL', matchType: 'REGEX', val: regex };
} else {
return { type: 'MATCH_ALL', matchType: 'STRING', val }
}
}
if (type === 'STRING') {
p.i += 1;
return { type: 'MATCH_ALL', matchType: 'STRING', val };
}
if (type === 'REGEX') {
p.i += 1;
const regex = state.caseSensitive ? new RegExp(val) : new RegExp(val, "i");
return { type: 'MATCH_ALL', matchType: 'REGEX', val: regex };
}
throw `Parse Error: Expected a regular expression or a string, but got: ${p.tokens[p.i]}; ${JSON.stringify(p)}`;
}
function value(p) {
const [type, val] = p.tokens[p.i];
// Cast atoms into strings or regexes depending on the current state.
if (type === 'ATOM') {
p.i += 1;
if (state.preferRegex) {
const regex = state.caseSensitive ? new RegExp(val) : new RegExp(val, "i");
return { type: 'REGEX', val: regex };
} else {
return { type: 'STRING', val }
}
}
if (type === 'STRING') {
p.i += 1;
return { type, val };
}
if (type === 'REGEX') {
p.i += 1;
const regex = state.caseSensitive ? new RegExp(val) : new RegExp(val, "i");
return { type, val: regex };
}
throw `Parse Error: Expected a regular expression or a string, but got: ${p.tokens[p.i]}; ${JSON.stringify(p)}`;
}
function date(p) {
const [type, val] = p.tokens[p.i];
p.i += 1;
return val;
}