@kernlang/python 3.5.4-canary.157.1.43256f52 → 3.5.4-canary.161.2.d4dbfea4

package/dist/fastapi-response.js

@@ -7,7 +7,7 @@
  * addRespondImports     — add necessary imports for respond node
  */
 import { getProps } from '@kernlang/core';
-import { escapePyStr } from './fastapi-utils.js';
+import { escapePyStr, quoteObjectKeysOutsideStrings } from './fastapi-utils.js';
 import { toSnakeCase } from './type-map.js';
 export function generateRespondFastAPI(respondNode, indent) {
     const p = getProps(respondNode);
@@ -87,15 +87,584 @@ function lowerJsArrayMethods(expr) {
     }
     return next;
 }
-export function rewriteFastAPIExpr(expr, pathParams) {
-    let result = expr;
+// Index of the bracket that closes the one at `openIdx`, tracking ()[]{} depth
+// and skipping string/template literals. -1 if unbalanced.
+function matchBalancedParen(expr, openIdx) {
+    let depth = 0;
+    let quote = null;
+    for (let i = openIdx; i < expr.length; i++) {
+        const c = expr[i];
+        if (quote) {
+            if (c === '\\')
+                i += 1;
+            else if (c === quote)
+                quote = null;
+            continue;
+        }
+        if (c === '"' || c === "'" || c === '`')
+            quote = c;
+        else if (c === '(' || c === '[' || c === '{')
+            depth += 1;
+        else if (c === ')' || c === ']' || c === '}') {
+            depth -= 1;
+            if (depth === 0)
+                return i;
+        }
+    }
+    return -1;
+}
+// Split a call's inner argument text on top-level commas, ignoring commas
+// inside nested ()[]{} or string literals.
+function splitTopLevelArgs(inner) {
+    const args = [];
+    let depth = 0;
+    let quote = null;
+    let start = 0;
+    for (let i = 0; i < inner.length; i++) {
+        const c = inner[i];
+        if (quote) {
+            if (c === '\\')
+                i += 1;
+            else if (c === quote)
+                quote = null;
+            continue;
+        }
+        if (c === '"' || c === "'" || c === '`')
+            quote = c;
+        else if (c === '(' || c === '[' || c === '{')
+            depth += 1;
+        else if (c === ')' || c === ']' || c === '}')
+            depth -= 1;
+        else if (c === ',' && depth === 0) {
+            args.push(inner.slice(start, i).trim());
+            start = i + 1;
+        }
+    }
+    args.push(inner.slice(start).trim());
+    return args;
+}
+// Lower JSON.stringify(...) / JSON.parse(...) to json.dumps/loads. Uses a
+// balanced, string-aware scan because the single argument can itself contain
+// commas, nested parens, brackets, braces, or string literals — which regex
+// cannot reliably capture (three regex iterations were each holed by review).
+// Skips occurrences inside string literals and those that are a property of
+// another receiver (e.g. `myJSON.stringify`). Handles the pretty-print form
+// `JSON.stringify(x, null, n)` → `json.dumps(x, indent=n)`.
+function lowerJsonBuiltinCalls(expr, imports) {
+    let out = '';
+    let i = 0;
+    let quote = null;
+    while (i < expr.length) {
+        const c = expr[i];
+        if (quote) {
+            out += c;
+            if (c === '\\') {
+                out += expr[i + 1] ?? '';
+                i += 2;
+                continue;
+            }
+            if (c === quote)
+                quote = null;
+            i += 1;
+            continue;
+        }
+        if (c === '"' || c === "'" || c === '`') {
+            quote = c;
+            out += c;
+            i += 1;
+            continue;
+        }
+        const m = expr.slice(i).match(/^JSON\.(stringify|parse)\(/);
+        const prev = expr[i - 1];
+        if (m && !(prev && /[\w.]/.test(prev))) {
+            const method = m[1];
+            const openIdx = i + m[0].length - 1;
+            const closeIdx = matchBalancedParen(expr, openIdx);
+            if (closeIdx !== -1) {
+                const args = splitTopLevelArgs(expr.slice(openIdx + 1, closeIdx));
+                // Recurse so a nested builtin in the argument is lowered too, e.g.
+                // JSON.stringify(JSON.parse(x)) → json.dumps(json.loads(x)) (Codex
+                // review on 9d8ed8d0). Terminates: the argument is strictly shorter.
+                const a0 = lowerJsonBuiltinCalls(args[0] ?? '', imports);
+                imports?.add('import json');
+                if (method === 'parse') {
+                    out += `json.loads(${a0})`;
+                }
+                else if (args.length >= 3 && /^(None|null)$/.test(args[1]) && /^\d+$/.test(args[2])) {
+                    out += `json.dumps(${a0}, indent=${args[2]})`;
+                }
+                else {
+                    out += `json.dumps(${a0})`;
+                }
+                i = closeIdx + 1;
+                continue;
+            }
+        }
+        out += c;
+        i += 1;
+    }
+    return out;
+}
+// Build the Python comprehension for one `Array.from(...)` call's argument list,
+// or return null if the call isn't a lowerable length-form. Uses the balanced
+// helpers (not regex) so a length value or arrow params containing braces/parens
+// don't desync (codex/gemini review of cd7c40ae).
+function tryLowerArrayFrom(args) {
+    if (args.length < 2)
+        return null;
+    // arg0 must be an object literal whose `length` property gives the count.
+    const arg0 = args[0].trim();
+    if (!arg0.startsWith('{') || matchBalancedParen(arg0, 0) !== arg0.length - 1)
+        return null;
+    let count = null;
+    for (const prop of splitTopLevelArgs(arg0.slice(1, -1))) {
+        const mm = prop.match(/^(?:length|["']length["'])\s*:\s*([\s\S]+)$/);
+        if (mm) {
+            count = mm[1].trim();
+            break;
+        }
+    }
+    if (count === null)
+        return null;
+    // arg1 must be an arrow `(params) => body` or `param => body`.
+    const arrowStr = args[1].trim();
+    let params;
+    let body;
+    if (arrowStr.startsWith('(')) {
+        const pClose = matchBalancedParen(arrowStr, 0);
+        if (pClose === -1)
+            return null;
+        const after = arrowStr.slice(pClose + 1).trim();
+        if (!after.startsWith('=>'))
+            return null;
+        params = splitTopLevelArgs(arrowStr.slice(1, pClose))
+            .map((s) => s.trim())
+            .filter(Boolean);
+        body = after.slice(2).trim();
+    }
+    else {
+        const am = arrowStr.match(/^([A-Za-z_$][\w$]*)\s*=>\s*([\s\S]+)$/);
+        if (!am)
+            return null;
+        params = [am[1]];
+        body = am[2].trim();
+    }
+    // Loop var = the INDEX (2nd param). The 1st param is the element, which is
+    // undefined for the length form, so it is NOT promoted to the loop variable
+    // (doing so would diverge from JS — `(x) => x` is [undefined…], not [0,1,…]).
+    // A non-simple index (destructuring) isn't a valid Python loop target → bail.
+    const idxVar = params[1] || '_';
+    if (!/^[A-Za-z_$][\w$]*$/.test(idxVar))
+        return null;
+    // `(_, i) => ({...})` parenthesizes the object body to disambiguate it from a
+    // block; unwrap ONLY when the enclosed body is an object literal, so a comma
+    // operator `(1, 2)` or grouped expr isn't mis-stripped (codex review).
+    if (body.startsWith('(') && matchBalancedParen(body, 0) === body.length - 1) {
+        const inner = body.slice(1, -1).trim();
+        if (inner.startsWith('{'))
+            body = inner;
+    }
+    // Recurse so a nested Array.from in the count or body is lowered too.
+    return `[${lowerArrayFromCalls(body)} for ${idxVar} in range(${lowerArrayFromCalls(count)})]`;
+}
+// Expand JS object-literal shorthand properties to explicit `key: key` so the
+// dict-key quoting pass can quote them: `{ items, page }` → `{ items: items,
+// page: page }`. Bracket/string-aware: only an object-literal entry that is a
+// bare identifier is expanded; `key: value`, `**spread`, computed keys, and
+// array/comprehension contents (`[]`) are left alone, and nested objects are
+// handled by recursing into each entry. Runs just before key quoting.
+function expandObjectShorthand(expr) {
+    let out = '';
+    let i = 0;
+    let quote = null;
+    while (i < expr.length) {
+        const c = expr[i];
+        if (quote) {
+            out += c;
+            if (c === '\\') {
+                out += expr[i + 1] ?? '';
+                i += 2;
+                continue;
+            }
+            if (c === quote)
+                quote = null;
+            i += 1;
+            continue;
+        }
+        if (c === '"' || c === "'" || c === '`') {
+            quote = c;
+            out += c;
+            i += 1;
+            continue;
+        }
+        if (c === '{') {
+            const close = matchBalancedParen(expr, i);
+            if (close !== -1) {
+                const rebuilt = splitTopLevelArgs(expr.slice(i + 1, close)).map((entry) => {
+                    const t = entry.trim();
+                    if (t === '')
+                        return entry;
+                    if (/^[A-Za-z_$][\w$]*$/.test(t))
+                        return `${t}: ${t}`;
+                    return expandObjectShorthand(entry);
+                });
+                out += `{${rebuilt.join(', ')}}`;
+                i = close + 1;
+                continue;
+            }
+        }
+        out += c;
+        i += 1;
+    }
+    return out;
+}
+// Lower `Array.from({ length: N }, (_, i) => BODY)` to a Python list
+// comprehension `[BODY for i in range(N)]` (Express keeps Array.from — valid
+// JS). Balanced, string-aware scan; runs BEFORE the ref/key/template passes so
+// they lower N and BODY in place. Only the `{ length: N }` form is handled;
+// `Array.from(iterable, fn)` (map form) is left untouched. A call immediately
+// followed by a method chain (`.map`, `.filter`, …) is left raw rather than
+// lowered, because the array-method pass cannot consume a comprehension
+// receiver and would emit malformed Python (codex review of cd7c40ae).
+function lowerArrayFromCalls(expr) {
+    let out = '';
+    let i = 0;
+    let quote = null;
+    while (i < expr.length) {
+        const c = expr[i];
+        if (quote) {
+            out += c;
+            if (c === '\\') {
+                out += expr[i + 1] ?? '';
+                i += 2;
+                continue;
+            }
+            if (c === quote)
+                quote = null;
+            i += 1;
+            continue;
+        }
+        if (c === '"' || c === "'" || c === '`') {
+            quote = c;
+            out += c;
+            i += 1;
+            continue;
+        }
+        const m = expr.slice(i).match(/^Array\.from\(/);
+        const prev = expr[i - 1];
+        if (m && !(prev && /[\w.]/.test(prev))) {
+            const openIdx = i + m[0].length - 1;
+            const closeIdx = matchBalancedParen(expr, openIdx);
+            if (closeIdx !== -1 && expr[closeIdx + 1] !== '.') {
+                const lowered = tryLowerArrayFrom(splitTopLevelArgs(expr.slice(openIdx + 1, closeIdx)));
+                if (lowered !== null) {
+                    out += lowered;
+                    i = closeIdx + 1;
+                    continue;
+                }
+            }
+        }
+        out += c;
+        i += 1;
+    }
+    return out;
+}
+function scanQuotedString(expr, startIndex, quote) {
+    for (let i = startIndex + 1; i < expr.length; i++) {
+        if (expr[i] === '\\') {
+            i += 1;
+            continue;
+        }
+        if (expr[i] === quote)
+            return i;
+    }
+    return -1;
+}
+function scanTemplateInterpolationEnd(expr, startIndex) {
+    let depth = 1;
+    for (let i = startIndex; i < expr.length; i++) {
+        const c = expr[i];
+        if (c === '"' || c === "'") {
+            const quotedEnd = scanQuotedString(expr, i, c);
+            if (quotedEnd === -1)
+                return -1;
+            i = quotedEnd;
+            continue;
+        }
+        if (c === '`') {
+            const templateEnd = scanTemplateLiteralEnd(expr, i);
+            if (templateEnd === -1)
+                return -1;
+            i = templateEnd;
+            continue;
+        }
+        if (c === '{')
+            depth += 1;
+        else if (c === '}') {
+            depth -= 1;
+            if (depth === 0)
+                return i;
+        }
+    }
+    return -1;
+}
+function scanTemplateLiteralEnd(expr, startIndex) {
+    for (let i = startIndex + 1; i < expr.length; i++) {
+        const c = expr[i];
+        if (c === '\\') {
+            i += 1;
+            continue;
+        }
+        if (c === '`')
+            return i;
+        if (c === '$' && expr[i + 1] === '{') {
+            const interpolationEnd = scanTemplateInterpolationEnd(expr, i + 2);
+            if (interpolationEnd === -1)
+                return -1;
+            i = interpolationEnd;
+        }
+    }
+    return -1;
+}
+function parseTemplateLiteral(expr, startIndex) {
+    const textParts = [];
+    const interpolationParts = [];
+    let text = '';
+    for (let i = startIndex + 1; i < expr.length;) {
+        const c = expr[i];
+        if (c === '\\') {
+            text += c;
+            if (i + 1 < expr.length)
+                text += expr[i + 1];
+            i += 2;
+            continue;
+        }
+        if (c === '`') {
+            textParts.push(text);
+            return { endIndex: i, textParts, interpolationParts };
+        }
+        if (c === '$' && expr[i + 1] === '{') {
+            textParts.push(text);
+            text = '';
+            const interpolationEnd = scanTemplateInterpolationEnd(expr, i + 2);
+            if (interpolationEnd === -1)
+                return undefined;
+            interpolationParts.push(expr.slice(i + 2, interpolationEnd));
+            i = interpolationEnd + 1;
+            continue;
+        }
+        text += c;
+        i += 1;
+    }
+    return undefined;
+}
+// Re-encode JS-template literal text (kept raw by parseTemplateLiteral, with `\x`
+// as two characters) for a Python double-quoted string. Most JS escapes are
+// ALSO valid Python escapes (`\n \t \r \b \f \v \\ \" \uXXXX \xXX \0`), so they
+// are preserved verbatim — decoding then re-encoding them only risks corrupting
+// the exotic ones (Codex reviews on 678e6bc1 and the escape-decoder commit).
+// Only the JS-specific escapes that Python does not recognise are converted to
+// the bare character: `\`` → backtick, `\$` → `$`, `\'` → `'`. A bare `"` (or a
+// bare trailing backslash, or raw control char) is escaped so the literal stays
+// valid.
+function escapeJsTemplateTextForPy(raw) {
+    let out = '';
+    for (let i = 0; i < raw.length; i++) {
+        const c = raw[i];
+        if (c === '\\' && i + 1 < raw.length) {
+            const next = raw[i + 1];
+            if (next === '`' || next === '$' || next === "'") {
+                out += next; // JS-only escape → bare char (Python has no such escape)
+            }
+            else {
+                out += `\\${next}`; // valid Python escape (\n, \uXXXX, \0, ...) — keep
+            }
+            i += 1;
+            continue;
+        }
+        if (c === '\\')
+            out += '\\\\'; // lone trailing backslash
+        else if (c === '"')
+            out += '\\"';
+        else if (c === '\n')
+            out += '\\n';
+        else if (c === '\r')
+            out += '\\r';
+        else if (c === '\t')
+            out += '\\t';
+        else
+            out += c;
+    }
+    return out;
+}
+function escapePythonTemplateText(text, forFormatTemplate) {
+    const escaped = escapeJsTemplateTextForPy(text);
+    if (!forFormatTemplate)
+        return escaped;
+    // str.format treats { } as field markers, so literal braces must be doubled.
+    return escaped.replace(/{/g, '{{').replace(/}/g, '}}');
+}
+function lowerTemplateLiteralToPython(parsed, pathParams, bodyFields, authUser, imports) {
+    if (parsed.interpolationParts.length === 0) {
+        return `"${escapePythonTemplateText(parsed.textParts.join(''), false)}"`;
+    }
+    const rewrittenInterpolations = parsed.interpolationParts.map((part) => rewriteFastAPIExpr(part.trim(), pathParams, bodyFields, authUser, imports));
+    let fmt = '';
+    for (let i = 0; i < parsed.textParts.length; i++) {
+        fmt += escapePythonTemplateText(parsed.textParts[i], true);
+        if (i < parsed.interpolationParts.length)
+            fmt += '{}';
+    }
+    return `"${fmt}".format(${rewrittenInterpolations.join(', ')})`;
+}
+function extractTemplateLiterals(expr, pathParams, bodyFields, authUser, imports) {
+    let maskedExpr = '';
+    const replacements = [];
+    let quote = null;
+    for (let i = 0; i < expr.length;) {
+        const c = expr[i];
+        if (quote) {
+            maskedExpr += c;
+            if (c === '\\') {
+                maskedExpr += expr[i + 1] ?? '';
+                i += 2;
+                continue;
+            }
+            if (c === quote)
+                quote = null;
+            i += 1;
+            continue;
+        }
+        if (c === '"' || c === "'") {
+            quote = c;
+            maskedExpr += c;
+            i += 1;
+            continue;
+        }
+        if (c === '`') {
+            const parsed = parseTemplateLiteral(expr, i);
+            if (!parsed) {
+                maskedExpr += c;
+                i += 1;
+                continue;
+            }
+            const placeholder = `__KERN_TEMPLATE_${replacements.length}__`;
+            const lowered = lowerTemplateLiteralToPython(parsed, pathParams, bodyFields, authUser, imports);
+            replacements.push({ placeholder, lowered });
+            maskedExpr += placeholder;
+            i = parsed.endIndex + 1;
+            continue;
+        }
+        maskedExpr += c;
+        i += 1;
+    }
+    return { maskedExpr, replacements };
+}
+// Lower JS spread elements to Python unpacking, choosing the operator from the
+// enclosing bracket: `{...x}` → `{**x}`, `[...x]` / `f(...x)` → `[*x]` / `f(*x)`.
+// Bracket-aware (a stack) and string-aware (skips quoted contents) so a literal
+// "..." inside a string is left intact. Runs BEFORE the request-ref rewrites so
+// that, e.g., `...user.roles` becomes `*user.roles` and the auth rewrite's
+// `(?<!\.)` lookbehind no longer sees the spread's trailing dot.
+function lowerSpreadElements(expr) {
+    let out = '';
+    const stack = [];
+    let i = 0;
+    while (i < expr.length) {
+        const ch = expr[i];
+        if (ch === '"' || ch === "'") {
+            const q = ch;
+            out += ch;
+            i++;
+            while (i < expr.length) {
+                out += expr[i];
+                if (expr[i] === '\\') {
+                    i++;
+                    if (i < expr.length)
+                        out += expr[i];
+                    i++;
+                    continue;
+                }
+                if (expr[i] === q) {
+                    i++;
+                    break;
+                }
+                i++;
+            }
+            continue;
+        }
+        if (ch === '{' || ch === '[' || ch === '(') {
+            stack.push(ch);
+            out += ch;
+            i++;
+            continue;
+        }
+        if (ch === '}' || ch === ']' || ch === ')') {
+            stack.pop();
+            out += ch;
+            i++;
+            continue;
+        }
+        if (ch === '.' && expr[i + 1] === '.' && expr[i + 2] === '.') {
+            out += stack[stack.length - 1] === '{' ? '**' : '*';
+            i += 3;
+            // Collapse whitespace after the operator so `{ ... body }` yields tight
+            // `{**body}` — the model_dump pass matches `**body`, not `** body` (Codex).
+            while (i < expr.length && /\s/.test(expr[i]))
+                i++;
+            continue;
+        }
+        out += ch;
+        i++;
+    }
+    return out;
+}
+export function rewriteFastAPIExpr(expr, pathParams, bodyFields = new Set(), authUser = false, imports) {
+    const { maskedExpr, replacements } = extractTemplateLiterals(expr, pathParams, bodyFields, authUser, imports);
+    let result = maskedExpr;
+    // Spread → unpacking first, so the request-ref rewrites below see clean
+    // operands (e.g. `*user.roles`, not `...user.roles`).
+    result = lowerSpreadElements(result);
+    // Expand object shorthand BEFORE Array.from lowering, so a shorthand length
+    // object `Array.from({ length }, …)` becomes `{ length: length }` and is
+    // recognised (codex review of d75a9d05). No later pass creates new object
+    // literals, so this single early pass covers length objects, arrow bodies,
+    // and every other object.
+    result = expandObjectShorthand(result);
+    // Array.from(length, arrow) → list comprehension. Runs before the ref/key
+    // passes so they lower the count and body of the produced comprehension.
+    result = lowerArrayFromCalls(result);
     // params.X → X (function param) for path params
     for (const param of pathParams) {
         result = result.replace(new RegExp(`\\bparams\\.${param}\\b`, 'g'), param);
     }
     // Fallback: any remaining params.X → X (for query params not in pathParams)
     result = result.replace(/\bparams\.([A-Za-z_]\w*)/g, '$1');
-    // body.X → body.X (Pydantic model — already correct)
+    // user.X → user["X"]: with auth, `user` is the decoded JWT payload (a dict
+    // returned by auth_required/auth_optional), so attribute access would raise
+    // AttributeError. Only applied when the route declares auth (Codex review).
+    // Skip text inside string literals so `{{"user.id"}}` isn't corrupted to
+    // `"user["id"]"` (Codex review on 02ecb2fa), and require `user` NOT be a
+    // property of something else (negative lookbehind `(?<!\.)`) so a nested
+    // body access like `body.user.id` is left intact (Kimi review on 02ecb2fa).
+    if (authUser) {
+        const USER_FIELD_RE = new RegExp(`${STRING_LITERAL_ALT}|(?<!\\.)\\buser\\.([A-Za-z_]\\w*)`, 'g');
+        result = result.replace(USER_FIELD_RE, (match, field) => (field ? `user["${field}"]` : match));
+    }
+    // body.X → body.<snake_case(X)>: the generated Pydantic model snake-cases
+    // every field, so a camelCase access would raise AttributeError at runtime.
+    // Only remap fields the model actually declares; leave unknown `body.X`
+    // (e.g. external validate schemas) untouched.
+    result = result.replace(/\bbody\.([A-Za-z_]\w*)/g, (match, field) => bodyFields.has(field) ? `body.${toSnakeCase(field)}` : match);
+    // Spreading the whole request body: `{**body}` raises TypeError because a
+    // Pydantic model is not a mapping, so unpack its dict form instead. This is
+    // unconditional: whenever the `body` symbol exists it is a Pydantic model
+    // (inline `RequestBody`, or an external `validate` schema typed `body: X` for
+    // POST/PUT/PATCH) — there is no `body: dict` codegen path, so model_dump() is
+    // always correct. Keying on bodyFields would wrongly skip external schemas
+    // (their field names are unknown but the param is still a model). A
+    // `**body.field` member spread is left alone via the `(?!\s*\.)` guard.
+    result = result.replace(/\*\*body\b(?!\s*\.)/g, '**body.model_dump()');
     // query.X → X (function param)
     result = result.replace(/\bquery\.([A-Za-z_]\w*)/g, '$1');
     // headers.X → request.headers.get("X")
@@ -127,6 +696,37 @@ export function rewriteFastAPIExpr(expr, pathParams) {
             return 'False';
         return match; // quoted string
     });
+    // ── Host-builtin lowering (JS globals → Python stdlib) ────────────────
+    // crypto / Date are fixed forms matched by regex with a `(?<![\w.])` guard so
+    // a custom receiver (`some.crypto.randomUUID()`) is left untouched. The JSON
+    // calls need balanced argument parsing (regex can't), so they go through the
+    // string-aware scanner `lowerJsonBuiltinCalls`.
+    // crypto.randomUUID() → str(uuid.uuid4())
+    result = result.replace(new RegExp(`${STRING_LITERAL_ALT}|(?<![\\w.])crypto\\.randomUUID\\(\\)`, 'g'), (match) => {
+        if (match === 'crypto.randomUUID()') {
+            imports?.add('import uuid');
+            return 'str(uuid.uuid4())';
+        }
+        return match; // string literal — leave untouched
+    });
+    // new Date().toISOString() → datetime.now(timezone.utc).isoformat()
+    result = result.replace(new RegExp(`${STRING_LITERAL_ALT}|(?<![\\w.])new Date\\(\\)\\.toISOString\\(\\)`, 'g'), (match) => {
+        if (match === 'new Date().toISOString()') {
+            imports?.add('from datetime import datetime, timezone');
+            return 'datetime.now(timezone.utc).isoformat()';
+        }
+        return match;
+    });
+    // JSON.stringify(...) → json.dumps(...) / JSON.parse(...) → json.loads(...)
+    result = lowerJsonBuiltinCalls(result, imports);
+    // Object-literal keys → quoted Python dict keys (`{userId: x}` →
+    // `{"userId": x}`). Applied last, mirroring the raw `res.json(...)` path's
+    // outer quote-after-lower order; runs after array-method lowering so dicts
+    // produced inside list comprehensions are quoted too.
+    result = quoteObjectKeysOutsideStrings(result);
+    for (const replacement of replacements) {
+        result = result.split(replacement.placeholder).join(replacement.lowered);
+    }
     return result;
 }
 export function extractExprCode(prop) {