123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431 |
- import {escapeDoubleQuotes} from 'sentry/utils';
- export enum TokenType {
- OPERATOR,
- FILTER,
- FREE_TEXT,
- }
- export type Token = {
- type: TokenType;
- value: string;
- key?: string;
- };
- function isOp(t: Token) {
- return t.type === TokenType.OPERATOR;
- }
- function isBooleanOp(value: string) {
- return ['OR', 'AND'].includes(value.toUpperCase());
- }
- function isParen(token: Token, character: '(' | ')') {
- return (
- token !== undefined &&
- isOp(token) &&
- ['(', ')'].includes(token.value) &&
- token.value === character
- );
- }
- // TODO(epurkhiser): This is legacy from before the existence of
- // searchSyntax/parser. We should absolutely replace the internals of this API
- // with `parseSearch`.
- export class MutableSearch {
- tokens: Token[];
- /**
- * Creates a MutableSearch from a string query
- */
- constructor(query: string);
- /**
- * Creates a mutable search query from a list of query parts
- */
- constructor(queries: string[]);
- constructor(tokensOrQuery: string[] | string) {
- const strTokens = Array.isArray(tokensOrQuery)
- ? tokensOrQuery
- : splitSearchIntoTokens(tokensOrQuery);
- this.tokens = [];
- for (let token of strTokens) {
- let tokenState = TokenType.FREE_TEXT;
- if (isBooleanOp(token)) {
- this.addOp(token.toUpperCase());
- continue;
- }
- if (token.startsWith('(')) {
- const parenMatch = token.match(/^\(+/g);
- if (parenMatch) {
- parenMatch[0].split('').map(paren => this.addOp(paren));
- token = token.replace(/^\(+/g, '');
- }
- }
- // Traverse the token and check if it's a filter condition or free text
- for (let i = 0, len = token.length; i < len; i++) {
- const char = token[i];
- if (i === 0 && (char === '"' || char === ':')) {
- break;
- }
- // We may have entered a filter condition
- if (char === ':') {
- const nextChar = token[i + 1] || '';
- if ([':', ' '].includes(nextChar)) {
- tokenState = TokenType.FREE_TEXT;
- } else {
- tokenState = TokenType.FILTER;
- }
- break;
- }
- }
- let trailingParen = '';
- if (token.endsWith(')') && !token.includes('(')) {
- const parenMatch = token.match(/\)+$/g);
- if (parenMatch) {
- trailingParen = parenMatch[0];
- token = token.replace(/\)+$/g, '');
- }
- }
- if (tokenState === TokenType.FREE_TEXT && token.length) {
- this.addFreeText(token);
- } else if (tokenState === TokenType.FILTER) {
- this.addStringFilter(token, false);
- }
- if (trailingParen !== '') {
- trailingParen.split('').map(paren => this.addOp(paren));
- }
- }
- }
- formatString() {
- const formattedTokens: string[] = [];
- for (const token of this.tokens) {
- switch (token.type) {
- case TokenType.FILTER:
- if (token.value === '' || token.value === null) {
- formattedTokens.push(`${token.key}:""`);
- } else if (/[\s\(\)\\"]/g.test(token.value)) {
- formattedTokens.push(`${token.key}:"${escapeDoubleQuotes(token.value)}"`);
- } else {
- formattedTokens.push(`${token.key}:${token.value}`);
- }
- break;
- case TokenType.FREE_TEXT:
- if (/[\s\(\)\\"]/g.test(token.value)) {
- formattedTokens.push(`"${escapeDoubleQuotes(token.value)}"`);
- } else {
- formattedTokens.push(token.value);
- }
- break;
- default:
- formattedTokens.push(token.value);
- }
- }
- return formattedTokens.join(' ').trim();
- }
- addStringFilter(filter: string, shouldEscape = true) {
- const [key, value] = parseFilter(filter);
- this.addFilterValues(key, [value], shouldEscape);
- return this;
- }
- addFilterValues(key: string, values: string[], shouldEscape = true) {
- for (const value of values) {
- // Filter values that we insert through the UI can contain special characters
- // that need to escaped. User entered filters should not be escaped.
- const escaped = shouldEscape ? escapeFilterValue(value) : value;
- const token: Token = {type: TokenType.FILTER, key, value: escaped};
- this.tokens.push(token);
- }
- return this;
- }
- setFilterValues(key: string, values: string[], shouldEscape = true) {
- this.removeFilter(key);
- this.addFilterValues(key, values, shouldEscape);
- return this;
- }
- get filters() {
- type Filters = Record<string, string[]>;
- const reducer = (acc: Filters, token: Token) => ({
- ...acc,
- [token.key!]: [...(acc[token.key!] ?? []), token.value],
- });
- return this.tokens
- .filter(t => t.type === TokenType.FILTER)
- .reduce<Filters>(reducer, {});
- }
- getFilterValues(key: string) {
- return this.filters[key] ?? [];
- }
- getFilterKeys() {
- return Object.keys(this.filters);
- }
- hasFilter(key: string): boolean {
- return this.getFilterValues(key).length > 0;
- }
- removeFilter(key: string) {
- const removeErroneousAndOrOps = () => {
- let toRemove = -1;
- do {
- if (toRemove >= 0) {
- this.tokens.splice(toRemove, 1);
- toRemove = -1;
- }
- for (let i = 0; i < this.tokens.length; i++) {
- const token = this.tokens[i];
- const prev = this.tokens[i - 1];
- const next = this.tokens[i + 1];
- if (isOp(token) && isBooleanOp(token.value)) {
- if (prev === undefined || isOp(prev) || next === undefined || isOp(next)) {
- // Want to avoid removing `(term) OR (term)` and `term OR (term)`
- if (
- prev &&
- next &&
- (isParen(prev, ')') || !isOp(prev)) &&
- (isParen(next, '(') || !isOp(next))
- ) {
- continue;
- }
- toRemove = i;
- break;
- }
- }
- }
- } while (toRemove >= 0);
- };
- this.tokens = this.tokens.filter(token => token.key !== key);
- // Remove any AND/OR operators that have become erroneous due to filtering out tokens
- removeErroneousAndOrOps();
- // Now the really complicated part: removing parens that only have one element in them.
- // Since parens are themselves tokens, this gets tricky. In summary, loop through the
- // tokens until we find the innermost open paren. Then forward search through the rest of the tokens
- // to see if that open paren corresponds to a closed paren with one or fewer items inside.
- // If it does, delete those parens, and loop again until there are no more parens to delete.
- let parensToDelete: number[] = [];
- const cleanParens = (_, idx: number) => !parensToDelete.includes(idx);
- do {
- if (parensToDelete.length) {
- this.tokens = this.tokens.filter(cleanParens);
- }
- parensToDelete = [];
- for (let i = 0; i < this.tokens.length; i++) {
- const token = this.tokens[i];
- if (!isOp(token) || token.value !== '(') {
- continue;
- }
- let alreadySeen = false;
- for (let j = i + 1; j < this.tokens.length; j++) {
- const nextToken = this.tokens[j];
- if (isOp(nextToken) && nextToken.value === '(') {
- // Continue down to the nested parens. We can skip i forward since we know
- // everything between i and j is NOT an open paren.
- i = j - 1;
- break;
- } else if (!isOp(nextToken)) {
- if (alreadySeen) {
- // This has more than one term, no need to delete
- break;
- }
- alreadySeen = true;
- } else if (isOp(nextToken) && nextToken.value === ')') {
- // We found another paren with zero or one terms inside. Delete the pair.
- parensToDelete = [i, j];
- break;
- }
- }
- if (parensToDelete.length > 0) {
- break;
- }
- }
- } while (parensToDelete.length > 0);
- // Now that all erroneous parens are removed we need to remove dangling OR/AND operators.
- // I originally removed all the dangling properties in a single loop, but that meant that
- // cases like `a OR OR b` would remove both operators, when only one should be removed. So
- // instead, we loop until we find an operator to remove, then go back to the start and loop
- // again.
- removeErroneousAndOrOps();
- return this;
- }
- removeFilterValue(key: string, value: string) {
- const values = this.getFilterValues(key);
- if (Array.isArray(values) && values.length) {
- this.setFilterValues(
- key,
- values.filter(item => item !== value)
- );
- }
- }
- addFreeText(value: string) {
- const token: Token = {type: TokenType.FREE_TEXT, value: formatQuery(value)};
- this.tokens.push(token);
- return this;
- }
- addOp(value: string) {
- const token: Token = {type: TokenType.OPERATOR, value};
- this.tokens.push(token);
- return this;
- }
- get freeText(): string[] {
- return this.tokens.filter(t => t.type === TokenType.FREE_TEXT).map(t => t.value);
- }
- set freeText(values: string[]) {
- this.tokens = this.tokens.filter(t => t.type !== TokenType.FREE_TEXT);
- for (const v of values) {
- this.addFreeText(v);
- }
- }
- copy() {
- const q = new MutableSearch([]);
- q.tokens = [...this.tokens];
- return q;
- }
- isEmpty() {
- return this.tokens.length === 0;
- }
- }
- /**
- * Splits search strings into tokens for parsing by tokenizeSearch.
- *
- * Should stay in sync with src.sentry.search.utils:split_query_into_tokens
- */
- function splitSearchIntoTokens(query: string) {
- const queryChars = Array.from(query);
- const tokens: string[] = [];
- let token = '';
- let endOfPrevWord = '';
- let quoteType = '';
- let quoteEnclosed = false;
- for (let idx = 0; idx < queryChars.length; idx++) {
- const char = queryChars[idx];
- const nextChar = queryChars.length - 1 > idx ? queryChars[idx + 1] : null;
- token += char;
- if (nextChar !== null && !isSpace(char) && isSpace(nextChar)) {
- endOfPrevWord = char;
- }
- if (isSpace(char) && !quoteEnclosed && endOfPrevWord !== ':' && !isSpace(token)) {
- tokens.push(token.trim());
- token = '';
- }
- if (["'", '"'].includes(char) && (!quoteEnclosed || quoteType === char)) {
- quoteEnclosed = !quoteEnclosed;
- if (quoteEnclosed) {
- quoteType = char;
- }
- }
- if (quoteEnclosed && char === '\\' && nextChar === quoteType) {
- token += nextChar;
- idx++;
- }
- }
- const trimmedToken = token.trim();
- if (trimmedToken !== '') {
- tokens.push(trimmedToken);
- }
- return tokens;
- }
- /**
- * Checks if the string is only spaces
- */
- function isSpace(s: string) {
- return s.trim() === '';
- }
- /**
- * Splits a filter on ':' and removes enclosing quotes if present, and returns
- * both sides of the split as strings.
- */
- function parseFilter(filter: string) {
- const idx = filter.indexOf(':');
- const key = removeSurroundingQuotes(filter.slice(0, idx));
- const value = removeSurroundingQuotes(filter.slice(idx + 1));
- return [key, value];
- }
- function removeSurroundingQuotes(text: string) {
- const length = text.length;
- if (length <= 1) {
- return text;
- }
- let left = 0;
- for (; left <= length / 2; left++) {
- if (text.charAt(left) !== '"') {
- break;
- }
- }
- let right = length - 1;
- for (; right >= length / 2; right--) {
- if (text.charAt(right) !== '"' || text.charAt(right - 1) === '\\') {
- break;
- }
- }
- return text.slice(left, right + 1);
- }
- /**
- * Strips enclosing quotes and parens from a query, if present.
- */
- function formatQuery(query: string) {
- return query.replace(/^["\(]+|["\)]+$/g, '');
- }
- /**
- * Some characters have special meaning in a filter value. So when they are
- * directly added as a value, we have to escape them to mean the literal.
- */
- export function escapeFilterValue(value: string) {
- // TODO(txiao): The types here are definitely wrong.
- // Need to dig deeper to see where exactly it's wrong.
- //
- // astericks (*) is used for wildcard searches
- return typeof value === 'string' ? value.replace(/([\*])/g, '\\$1') : value;
- }
|