Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix context stack misalignment caused by error replay #12508

Merged
merged 5 commits into from Apr 2, 2018
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
13 changes: 10 additions & 3 deletions packages/react-reconciler/src/ReactFiberHostContext.js
Expand Up @@ -60,12 +60,19 @@ export default function<T, P, I, TI, HI, PI, C, CC, CX, PL>(
// Push current root instance onto the stack;
// This allows us to reset root when portals are popped.
push(rootInstanceStackCursor, nextRootInstance, fiber);

const nextRootContext = getRootHostContext(nextRootInstance);

// Track the context and the Fiber that provided it.
// This enables us to pop only Fibers that provide unique contexts.
push(contextFiberStackCursor, fiber, fiber);

// Finally, we need to push the host context to the stack.
// However, we can't just call getRootHostContext() and push it because
// we'd have a different number of entries on the stack depending on
// whether getRootHostContext() throws somewhere in renderer code or not.
// So we push an empty value first. This lets us safely unwind on errors.
push(contextStackCursor, NO_CONTEXT, fiber);
const nextRootContext = getRootHostContext(nextRootInstance);
// Now that we know this function doesn't throw, replace it.
pop(contextStackCursor, fiber);
push(contextStackCursor, nextRootContext, fiber);
}

Expand Down
31 changes: 25 additions & 6 deletions packages/react-reconciler/src/ReactFiberScheduler.js
Expand Up @@ -263,10 +263,18 @@ export default function<T, P, I, TI, HI, PI, C, CC, CX, PL>(

let stashedWorkInProgressProperties;
let replayUnitOfWork;
let isReplayingFailedUnitOfWork;
let originalReplayError;
if (__DEV__ && replayFailedUnitOfWorkWithInvokeGuardedCallback) {
stashedWorkInProgressProperties = null;
replayUnitOfWork = (failedUnitOfWork: Fiber, isAsync: boolean) => {
// Retore the original state of the work-in-progress
isReplayingFailedUnitOfWork = false;
originalReplayError = null;
replayUnitOfWork = (
failedUnitOfWork: Fiber,
error: mixed,
isAsync: boolean,
) => {
// Restore the original state of the work-in-progress
assignFiberPropertiesInDEV(
failedUnitOfWork,
stashedWorkInProgressProperties,
Expand All @@ -290,12 +298,17 @@ export default function<T, P, I, TI, HI, PI, C, CC, CX, PL>(
break;
}
// Replay the begin phase.
isReplayingFailedUnitOfWork = true;
originalReplayError = error;
invokeGuardedCallback(null, workLoop, null, isAsync);
isReplayingFailedUnitOfWork = false;
originalReplayError = null;
if (hasCaughtError()) {
clearCaughtError();
} else {
// This should be unreachable because the render phase is
// idempotent
// If the begin phase did not fail the second time, set this pointer
// back to the original value.
nextUnitOfWork = failedUnitOfWork;
}
};
}
Expand Down Expand Up @@ -855,9 +868,15 @@ export default function<T, P, I, TI, HI, PI, C, CC, CX, PL>(
);
}
let next = beginWork(current, workInProgress, nextRenderExpirationTime);

if (__DEV__) {
ReactDebugCurrentFiber.resetCurrentFiber();
if (isReplayingFailedUnitOfWork) {
// Currently replaying a failed unit of work. This should be unreachable,
// because the render phase is meant to be idempotent, and it should
// have thrown again. Since it didn't, rethrow the original error, so
// React's internal stack is not misaligned.
throw originalReplayError;
Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should we maybe clear the variable just before throwing so we don't hold onto it until next replay?

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yeah good call

}
}
if (__DEV__ && ReactFiberInstrumentation.debugTool) {
ReactFiberInstrumentation.debugTool.onBeginWork(workInProgress);
Expand Down Expand Up @@ -935,7 +954,7 @@ export default function<T, P, I, TI, HI, PI, C, CC, CX, PL>(

if (__DEV__ && replayFailedUnitOfWorkWithInvokeGuardedCallback) {
const failedUnitOfWork = nextUnitOfWork;
replayUnitOfWork(failedUnitOfWork, isAsync);
replayUnitOfWork(failedUnitOfWork, thrownValue, isAsync);
}

const sourceFiber: Fiber = nextUnitOfWork;
Expand Down
@@ -0,0 +1,59 @@
/**
* Copyright (c) 2013-present, Facebook, Inc.
*
* This source code is licensed under the MIT license found in the
* LICENSE file in the root directory of this source tree.
*
* @emails react-core
* @jest-environment node
*/

'use strict';

let React;
let ReactNoop;

describe('ReactIncrementalErrorReplay', () => {
beforeEach(() => {
jest.resetModules();
React = require('react');
ReactNoop = require('react-noop-renderer');
});

function div(...children) {
children = children.map(c => (typeof c === 'string' ? {text: c} : c));
return {type: 'div', children, prop: undefined};
}

function span(prop) {
return {type: 'span', children: [], prop};
}

it('should fail gracefully on error in the host environment', () => {
ReactNoop.simulateErrorInHostConfig(() => {
ReactNoop.render(<span />);
expect(() => ReactNoop.flush()).toThrow('Error in host config.');
Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This one causes an unexpected pop because replay logic assumes we can always safely pop context after "begin" phase failed:

switch (failedUnitOfWork.tag) {
case HostRoot:
popHostContainer(failedUnitOfWork);
popTopLevelLegacyContextObject(failedUnitOfWork);
break;
case HostComponent:
popHostContext(failedUnitOfWork);
break;
case ClassComponent:
popLegacyContextProvider(failedUnitOfWork);
break;
case HostPortal:
popHostContainer(failedUnitOfWork);
break;
case ContextProvider:
popProvider(failedUnitOfWork);
break;
}

But this doesn't necessarily need to be the case if pushing context involves running user or renderer code:

const nextContext = getChildHostContext(context, fiber.type, rootInstance);

const nextRootContext = getRootHostContext(nextRootInstance);

In that case there's a chance we won't get to push, and the stack won't line up.

Curiously, for legacy class context we seem to avoid this problem because we first push without running user code and then "replace" it. So even if user code failed, the stack would match up. Maybe we could always do something like this?

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yeah something like this should work. Seems easy to mess up, though.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sounds fine, we'll have some inline comments so the next person knows what's up.

});
});

it('should fail gracefully on error that does not reproduce on replay', () => {
let didInit = false;

function badLazyInit() {
const needsInit = !didInit;
didInit = true;
if (needsInit) {
throw new Error('Hi');
}
}

class App extends React.Component {
render() {
badLazyInit();
return <div />;
}
}
ReactNoop.render(<App />);
expect(() => ReactNoop.flush()).toThrow('Hi');
Copy link
Collaborator Author

@gaearon gaearon Apr 2, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This one fails because we assume rendering is idempotent:

// This should be unreachable because the render phase is
// idempotent

But there are cases (like lazy init) where it might throw the first time but not on subsequent calls. Or just a buggy render implementation where we may render something completely different. We should handle this gracefully. When this happens, when we exit this block, failedUnitOfWork is not the same as nextUnitOfWork (which might be null):

});
});