[Flight] Parse Stack Trace from Structured CallSite if available (facebook#33135)

This is first step to include more enclosing line/column in the parsed
data.

We install our own `prepareStackTrace` to collect structured callsite
data and only fall back to parsing the string if it was already
evaluated or if `prepareStackTrace` doesn't work in this environment.

We still mirror the default V8 format for encoding the function name
part. A lot of this is covered by tests already.

Author: sebmarkbage

packages/react-server/src/ReactFlightServer.js

@@ -152,11 +152,27 @@ function defaultFilterStackFrame(
   );
 }
 
+// DEV-only cache of parsed and filtered stack frames.
+const stackTraceCache: WeakMap<Error, ReactStackTrace> = __DEV__
+  ? new WeakMap()
+  : (null: any);
+
 function filterStackTrace(
   request: Request,
   error: Error,
   skipFrames: number,
 ): ReactStackTrace {
+  const existing = stackTraceCache.get(error);
+  if (existing !== undefined) {
+    // Return a clone because the Flight protocol isn't yet resilient to deduping
+    // objects in the debug info. TODO: Support deduping stacks.
+    const clone = existing.slice(0);
+    for (let i = 0; i < clone.length; i++) {
+      // $FlowFixMe[invalid-tuple-arity]
+      clone[i] = clone[i].slice(0);
+    }
+    return clone;
+  }
   // Since stacks can be quite large and we pass a lot of them, we filter them out eagerly
   // to save bandwidth even in DEV. We'll also replay these stacks on the client so by
   // stripping them early we avoid that overhead. Otherwise we'd normally just rely on
@@ -183,6 +199,7 @@ function filterStackTrace(
       i--;
     }
   }
+  stackTraceCache.set(error, stack);
   return stack;
 }
 

packages/react-server/src/ReactFlightStackConfigV8.js

@@ -9,23 +9,104 @@
 
 import type {ReactStackTrace} from 'shared/ReactTypes';
 
-import DefaultPrepareStackTrace from 'shared/DefaultPrepareStackTrace';
+let framesToSkip: number = 0;
+let collectedStackTrace: null | ReactStackTrace = null;
 
-function getStack(error: Error): string {
-  // We override Error.prepareStackTrace with our own version that normalizes
-  // the stack to V8 formatting even if the server uses other formatting.
-  // It also ensures that source maps are NOT applied to this since that can
-  // be slow we're better off doing that lazily from the client instead of
-  // eagerly on the server. If the stack has already been read, then we might
-  // not get a normalized stack and it might still have been source mapped.
-  const previousPrepare = Error.prepareStackTrace;
-  Error.prepareStackTrace = DefaultPrepareStackTrace;
-  try {
-    // eslint-disable-next-line react-internal/safe-string-coercion
-    return String(error.stack);
-  } finally {
-    Error.prepareStackTrace = previousPrepare;
+const identifierRegExp = /^[a-zA-Z_$][0-9a-zA-Z_$]*$/;
+
+function getMethodCallName(callSite: CallSite): string {
+  const typeName = callSite.getTypeName();
+  const methodName = callSite.getMethodName();
+  const functionName = callSite.getFunctionName();
+  let result = '';
+  if (functionName) {
+    if (
+      typeName &&
+      identifierRegExp.test(functionName) &&
+      functionName !== typeName
+    ) {
+      result += typeName + '.';
+    }
+    result += functionName;
+    if (
+      methodName &&
+      functionName !== methodName &&
+      !functionName.endsWith('.' + methodName) &&
+      !functionName.endsWith(' ' + methodName)
+    ) {
+      result += ' [as ' + methodName + ']';
+    }
+  } else {
+    if (typeName) {
+      result += typeName + '.';
+    }
+    if (methodName) {
+      result += methodName;
+    } else {
+      result += '<anonymous>';
+    }
+  }
+  return result;
+}
+
+function collectStackTrace(
+  error: Error,
+  structuredStackTrace: CallSite[],
+): string {
+  const result: ReactStackTrace = [];
+  // Collect structured stack traces from the callsites.
+  // We mirror how V8 serializes stack frames and how we later parse them.
+  for (let i = framesToSkip; i < structuredStackTrace.length; i++) {
+    const callSite = structuredStackTrace[i];
+    let name = callSite.getFunctionName() || '<anonymous>';
+    if (name === 'react-stack-bottom-frame') {
+      // Skip everything after the bottom frame since it'll be internals.
+      break;
+    } else if (callSite.isNative()) {
+      result.push([name, '', 0, 0]);
+    } else {
+      // We encode complex function calls as if they're part of the function
+      // name since we cannot simulate the complex ones and they look the same
+      // as function names in UIs on the client as well as stacks.
+      if (callSite.isConstructor()) {
+        name = 'new ' + name;
+      } else if (!callSite.isToplevel()) {
+        name = getMethodCallName(callSite);
+      }
+      if (name === '<anonymous>') {
+        name = '';
+      }
+      let filename = callSite.getScriptNameOrSourceURL() || '<anonymous>';
+      if (filename === '<anonymous>') {
+        filename = '';
+      }
+      if (callSite.isEval() && !filename) {
+        const origin = callSite.getEvalOrigin();
+        if (origin) {
+          filename = origin.toString() + ', <anonymous>';
+        }
+      }
+      const line = callSite.getLineNumber() || 0;
+      const col = callSite.getColumnNumber() || 0;
+      result.push([name, filename, line, col]);
+    }
   }
+  // At the same time we generate a string stack trace just in case someone
+  // else reads it. Ideally, we'd call the previous prepareStackTrace to
+  // ensure it's in the expected format but it's common for that to be
+  // source mapped and since we do a lot of eager parsing of errors, it
+  // would be slow in those environments. We could maybe just rely on those
+  // environments having to disable source mapping globally to speed things up.
+  // For now, we just generate a default V8 formatted stack trace without
+  // source mapping as a fallback.
+  const name = error.name || 'Error';
+  const message = error.message || '';
+  let stack = name + ': ' + message;
+  for (let i = 0; i < structuredStackTrace.length; i++) {
+    stack += '\n    at ' + structuredStackTrace[i].toString();
+  }
+  collectedStackTrace = result;
+  return stack;
 }
 
 // This matches either of these V8 formats.
@@ -39,7 +120,32 @@ export function parseStackTrace(
   error: Error,
   skipFrames: number,
 ): ReactStackTrace {
-  let stack = getStack(error);
+  // We override Error.prepareStackTrace with our own version that collects
+  // the structured data. We need more information than the raw stack gives us
+  // and we need to ensure that we don't get the source mapped version.
+  collectedStackTrace = null;
+  framesToSkip = skipFrames;
+  const previousPrepare = Error.prepareStackTrace;
+  Error.prepareStackTrace = collectStackTrace;
+  let stack;
+  try {
+    // eslint-disable-next-line react-internal/safe-string-coercion
+    stack = String(error.stack);
+  } finally {
+    Error.prepareStackTrace = previousPrepare;
+  }
+
+  if (collectedStackTrace !== null) {
+    const result = collectedStackTrace;
+    collectedStackTrace = null;
+    return result;
+  }
+
+  // If the stack has already been read, or this is not actually a V8 compatible
+  // engine then we might not get a normalized stack and it might still have been
+  // source mapped. Regardless we try our best to parse it. This works best if the
+  // environment just uses default V8 formatting and no source mapping.
+
   if (stack.startsWith('Error: react-stack-top-frame\n')) {
     // V8's default formatting prefixes with the error message which we
     // don't want/need.