sailfork / t3code-github / .repos/ alchemy-effect/ packages/ sql/ src/ Migrations/ Script.ts

/**
 * The dialects a SQL script can be split for. `sqlite` is split by
 * `splitSqliteStatements` (in `SqlFile.ts`), which also understands trigger
 * bodies.
 */
export type ScriptDialect = "postgres" | "mysql" | "googlesql";

/** One statement of a script, without its terminating `;`. */
export interface ScannedStatement {
  /**
   * The statement's exact source slice, comments and surrounding whitespace
   * included, so offsets into `code` index it directly.
   */
  readonly sql: string;
  /** The statement with comments removed: what to send to the database. */
  readonly text: string;
  /**
   * `sql` with comments blanked and every string, quoted identifier, and
   * dollar-quoted body masked to `_` plus spaces. Offsets match `sql`, so a
   * keyword found in `code` can be sliced out of `sql`.
   */
  readonly code: string;
  /** A comment or quoted span that never closed before the end of the script. */
  readonly unterminated?: "comment" | "quote";
}

interface Rules {
  /** `#` starts a line comment. */
  readonly hashComments: boolean;
  /** `--` is a comment only when followed by whitespace (MySQL). */
  readonly dashCommentNeedsSpace: boolean;
  /** Block comments nest (PostgreSQL). */
  readonly nestedBlockComments: boolean;
  /** `$tag$ … $tag

	
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
		
	.repos/alchemy-effect/packages/sql/src/Migrations/Script.ts at main · sailfork/t3code-github · Sailfork
	
		
bodies (PostgreSQL). */ readonly dollarQuotes: boolean; /** `'''…'''` and `"""…"""` (GoogleSQL). */ readonly tripleQuotes: boolean; /** Quote characters that open a string or quoted identifier. */ readonly quotes: ReadonlyArray<string>; /** Backslash escapes inside every quoted span. */ readonly backslashEscapes: boolean; } const RULES: Record<ScriptDialect, Rules> = { postgres: { hashComments: false, dashCommentNeedsSpace: false, nestedBlockComments: true, dollarQuotes: true, tripleQuotes: false, quotes: ["'", '"'], backslashEscapes: false, }, mysql: { hashComments: true, dashCommentNeedsSpace: true, nestedBlockComments: false, dollarQuotes: false, tripleQuotes: false, quotes: ["'", '"', "`"], backslashEscapes: true, }, googlesql: { hashComments: true, dashCommentNeedsSpace: false, nestedBlockComments: false, dollarQuotes: false, tripleQuotes: true, quotes: ["'", '"', "`"], backslashEscapes: true, }, }; /** Index just past a line comment starting at `i` (the newline is kept). */ const endOfLine = (sql: string, i: number) => { const end = sql.indexOf("\n", i); return end < 0 ? sql.length : end; }; interface Span { readonly end: number; readonly closed: boolean; } /** The end of a block comment starting at `i`. */ const blockComment = (sql: string, i: number, nested: boolean): Span => { let depth = 1; let j = i + 2; while (j < sql.length && depth > 0) { if (nested && sql.startsWith("/*", j)) { depth++; j += 2; } else if (sql.startsWith("*/", j)) { depth--; j += 2; } else j++; } return { end: j, closed: depth === 0 }; }; /** * The end of a quoted span starting at `i`. A doubled closing quote * escapes itself in every dialect except GoogleSQL; backslashes escape when * the dialect (or a PostgreSQL `E'…'` string) says so, never in a GoogleSQL * raw string (`r'…'`). */ const quoted = ( sql: string, i: number, close: string, options: { backslashEscapes: boolean; doubledEscapes: boolean }, ): Span => { let j = i + close.length; while (j < sql.length) { if (options.backslashEscapes && sql[j] === "\\") { j += 2; continue; } if (sql.startsWith(close, j)) { j += close.length; if (options.doubledEscapes && close.length === 1 && sql[j] === close) { j++; continue; } return { end: j, closed: true }; } j++; } return { end: sql.length, closed: false }; }; const isWordChar = (ch: string | undefined) => ch !== undefined && /[A-Za-z0-9_]/.test(ch); /** * Split a SQL script into statements on top-level `;`. Semicolons inside * strings, quoted identifiers, dollar-quoted bodies, and comments never * split, and each dialect's quoting and comment rules apply. Empty * statements are dropped. * * This is a lexer, not a parser: it does not understand statement bodies * that contain their own `;` outside quotes (procedural blocks), which * migrations for these engines don't use. */ export const scanSqlScript = ( sql: string, dialect: ScriptDialect, ): ReadonlyArray<ScannedStatement> => { const rules = RULES[dialect]; const statements: ScannedStatement[] = []; let start = 0; let text = ""; let code = ""; let unterminated: ScannedStatement["unterminated"]; const push = (end: number) => { if (code.trim().length > 0) { statements.push({ sql: sql.slice(start, end), text: text.trim(), code, ...(unterminated ? { unterminated } : {}), }); } start = end + 1; text = ""; code = ""; }; const blank = (span: string) => span.replace(/[^\n]/g, " "); const mask = (span: string) => `_${" ".repeat(Math.max(0, span.length - 1))}`; let i = 0; while (i < sql.length) { const ch = sql[i]!; const next = sql[i + 1]; const dashComment = ch === "-" && next === "-" && (!rules.dashCommentNeedsSpace || i + 2 >= sql.length || /\s/.test(sql[i + 2]!)); if (dashComment || (rules.hashComments && ch === "#")) { const end = endOfLine(sql, i); code += blank(sql.slice(i, end)); text += " "; i = end; continue; } if (ch === "/" && next === "*") { const span = blockComment(sql, i, rules.nestedBlockComments); if (!span.closed) unterminated = "comment"; code += blank(sql.slice(i, span.end)); text += " "; i = span.end; continue; } if (rules.dollarQuotes && ch === "$" && !isWordChar(sql[i - 1])) { const tag = /^\$(?:[A-Za-z_][A-Za-z_0-9]*)?\$/.exec(sql.slice(i))?.[0]; if (tag) { const close = sql.indexOf(tag, i + tag.length); if (close < 0) unterminated = "quote"; const end = close < 0 ? sql.length : close + tag.length; code += mask(sql.slice(i, end)); text += sql.slice(i, end); i = end; continue; } } if (rules.quotes.includes(ch)) { // A prefix letter right before the quote: E'…' (PostgreSQL escapes), // r'…' / rb'…' (GoogleSQL raw strings). const prefix = /(?:^|[^A-Za-z0-9_])([A-Za-z]{1,2})$/ .exec(sql.slice(Math.max(0, i - 3), i))?.[1] ?.toLowerCase() ?? ""; const raw = dialect === "googlesql" && ["r", "rb", "br"].includes(prefix); const escaped = dialect === "postgres" && prefix === "e" && ch === "'"; const close = rules.tripleQuotes && ch !== "`" && sql.startsWith(ch.repeat(3), i) ? ch.repeat(3) : ch; const span = quoted(sql, i, close, { backslashEscapes: !raw && (rules.backslashEscapes || escaped), doubledEscapes: dialect !== "googlesql", }); if (!span.closed) unterminated = "quote"; code += mask(sql.slice(i, span.end)); text += sql.slice(i, span.end); i = span.end; continue; } if (ch === ";") { push(i); i++; continue; } code += ch; text += ch; i++; } push(sql.length); return statements; }; /** Statements of a SQL script with comments removed; see {@link scanSqlScript}. */ export const splitSqlScript = (sql: string, dialect: ScriptDialect): string[] => scanSqlScript(sql, dialect).map((statement) => statement.text);
.repos/alchemy-effect/packages/sql/src/Migrations/Script.ts at main · sailfork/t3code-github · Sailfork
bodies (PostgreSQL). */\n readonly dollarQuotes: boolean;\n /** `'''…'''` and `\"\"\"…\"\"\"` (GoogleSQL). */\n readonly tripleQuotes: boolean;\n /** Quote characters that open a string or quoted identifier. */\n readonly quotes: ReadonlyArray\u003cstring>;\n /** Backslash escapes inside every quoted span. */\n readonly backslashEscapes: boolean;\n}\n\nconst RULES: Record\u003cScriptDialect, Rules> = {\n postgres: {\n hashComments: false,\n dashCommentNeedsSpace: false,\n nestedBlockComments: true,\n dollarQuotes: true,\n tripleQuotes: false,\n quotes: [\"'\", '\"'],\n backslashEscapes: false,\n },\n mysql: {\n hashComments: true,\n dashCommentNeedsSpace: true,\n nestedBlockComments: false,\n dollarQuotes: false,\n tripleQuotes: false,\n quotes: [\"'\", '\"', \"`\"],\n backslashEscapes: true,\n },\n googlesql: {\n hashComments: true,\n dashCommentNeedsSpace: false,\n nestedBlockComments: false,\n dollarQuotes: false,\n tripleQuotes: true,\n quotes: [\"'\", '\"', \"`\"],\n backslashEscapes: true,\n },\n};\n\n/** Index just past a line comment starting at `i` (the newline is kept). */\nconst endOfLine = (sql: string, i: number) => {\n const end = sql.indexOf(\"\\n\", i);\n return end \u003c 0 ? sql.length : end;\n};\n\ninterface Span {\n readonly end: number;\n readonly closed: boolean;\n}\n\n/** The end of a block comment starting at `i`. */\nconst blockComment = (sql: string, i: number, nested: boolean): Span => {\n let depth = 1;\n let j = i + 2;\n while (j \u003c sql.length && depth > 0) {\n if (nested && sql.startsWith(\"/*\", j)) {\n depth++;\n j += 2;\n } else if (sql.startsWith(\"*/\", j)) {\n depth--;\n j += 2;\n } else j++;\n }\n return { end: j, closed: depth === 0 };\n};\n\n/**\n * The end of a quoted span starting at `i`. A doubled closing quote\n * escapes itself in every dialect except GoogleSQL; backslashes escape when\n * the dialect (or a PostgreSQL `E'…'` string) says so, never in a GoogleSQL\n * raw string (`r'…'`).\n */\nconst quoted = (\n sql: string,\n i: number,\n close: string,\n options: { backslashEscapes: boolean; doubledEscapes: boolean },\n): Span => {\n let j = i + close.length;\n while (j \u003c sql.length) {\n if (options.backslashEscapes && sql[j] === \"\\\\\") {\n j += 2;\n continue;\n }\n if (sql.startsWith(close, j)) {\n j += close.length;\n if (options.doubledEscapes && close.length === 1 && sql[j] === close) {\n j++;\n continue;\n }\n return { end: j, closed: true };\n }\n j++;\n }\n return { end: sql.length, closed: false };\n};\n\nconst isWordChar = (ch: string | undefined) => ch !== undefined && /[A-Za-z0-9_]/.test(ch);\n\n/**\n * Split a SQL script into statements on top-level `;`. Semicolons inside\n * strings, quoted identifiers, dollar-quoted bodies, and comments never\n * split, and each dialect's quoting and comment rules apply. Empty\n * statements are dropped.\n *\n * This is a lexer, not a parser: it does not understand statement bodies\n * that contain their own `;` outside quotes (procedural blocks), which\n * migrations for these engines don't use.\n */\nexport const scanSqlScript = (\n sql: string,\n dialect: ScriptDialect,\n): ReadonlyArray\u003cScannedStatement> => {\n const rules = RULES[dialect];\n const statements: ScannedStatement[] = [];\n let start = 0;\n let text = \"\";\n let code = \"\";\n let unterminated: ScannedStatement[\"unterminated\"];\n const push = (end: number) => {\n if (code.trim().length > 0) {\n statements.push({\n sql: sql.slice(start, end),\n text: text.trim(),\n code,\n ...(unterminated ? { unterminated } : {}),\n });\n }\n start = end + 1;\n text = \"\";\n code = \"\";\n };\n const blank = (span: string) => span.replace(/[^\\n]/g, \" \");\n const mask = (span: string) => `_${\" \".repeat(Math.max(0, span.length - 1))}`;\n\n let i = 0;\n while (i \u003c sql.length) {\n const ch = sql[i]!;\n const next = sql[i + 1];\n\n const dashComment =\n ch === \"-\" &&\n next === \"-\" &&\n (!rules.dashCommentNeedsSpace || i + 2 >= sql.length || /\\s/.test(sql[i + 2]!));\n if (dashComment || (rules.hashComments && ch === \"#\")) {\n const end = endOfLine(sql, i);\n code += blank(sql.slice(i, end));\n text += \" \";\n i = end;\n continue;\n }\n if (ch === \"/\" && next === \"*\") {\n const span = blockComment(sql, i, rules.nestedBlockComments);\n if (!span.closed) unterminated = \"comment\";\n code += blank(sql.slice(i, span.end));\n text += \" \";\n i = span.end;\n continue;\n }\n if (rules.dollarQuotes && ch === \"$\" && !isWordChar(sql[i - 1])) {\n const tag = /^\\$(?:[A-Za-z_][A-Za-z_0-9]*)?\\$/.exec(sql.slice(i))?.[0];\n if (tag) {\n const close = sql.indexOf(tag, i + tag.length);\n if (close \u003c 0) unterminated = \"quote\";\n const end = close \u003c 0 ? sql.length : close + tag.length;\n code += mask(sql.slice(i, end));\n text += sql.slice(i, end);\n i = end;\n continue;\n }\n }\n if (rules.quotes.includes(ch)) {\n // A prefix letter right before the quote: E'…' (PostgreSQL escapes),\n // r'…' / rb'…' (GoogleSQL raw strings).\n const prefix =\n /(?:^|[^A-Za-z0-9_])([A-Za-z]{1,2})$/\n .exec(sql.slice(Math.max(0, i - 3), i))?.[1]\n ?.toLowerCase() ?? \"\";\n const raw = dialect === \"googlesql\" && [\"r\", \"rb\", \"br\"].includes(prefix);\n const escaped = dialect === \"postgres\" && prefix === \"e\" && ch === \"'\";\n const close =\n rules.tripleQuotes && ch !== \"`\" && sql.startsWith(ch.repeat(3), i) ? ch.repeat(3) : ch;\n const span = quoted(sql, i, close, {\n backslashEscapes: !raw && (rules.backslashEscapes || escaped),\n doubledEscapes: dialect !== \"googlesql\",\n });\n if (!span.closed) unterminated = \"quote\";\n code += mask(sql.slice(i, span.end));\n text += sql.slice(i, span.end);\n i = span.end;\n continue;\n }\n if (ch === \";\") {\n push(i);\n i++;\n continue;\n }\n code += ch;\n text += ch;\n i++;\n }\n push(sql.length);\n return statements;\n};\n\n/** Statements of a SQL script with comments removed; see {@link scanSqlScript}. */\nexport const splitSqlScript = (sql: string, dialect: ScriptDialect): string[] =>\n scanSqlScript(sql, dialect).map((statement) => statement.text);\n","preview":null,"history":null,"version":"6af0fef55874fc0ea202db1fcd4a02bcafc04010"},"/_session?shell=1":null}