Standardized helpful errors and methods for simpler, safer, and easier to read code.
Standardize on helpful errors for simpler, safer, easier to read code
- extend the
HelpfulError
for observable and actionable error messages - leverage the
UnexpectedCodePath
to eliminate complexity with narrowed codepaths - leverage the
BadRequestError
to make it clear when your logic successfully rejected a request - test that logic throws errors ergonomically with
getError
npm install --save @ehmpathy/error-fns
The UnexpectedCodePath
error is probably the most common type of error you'll throw.
It's common in business logic that you'll face a scenario that is technically possible but logically shouldn't occur.
For example, lets say you're writing on-vehicle code to check the tire pressures of a vehicle.
// given the tires to check
const tires = Tire[];
// first get the tire pressures for each
const tirePressures: number[] = tires.map(tire => getTirePressure(tire));
// now get the lowest tire pressure
const lowestTirePressure: number | undefined = tires.sort()[0];
In this case, its technically possible that lowestTirePressure
could be undefined: there could not be any tires.
However, this is definitely an unexpected code path for our application. We can just halt our logic if we reach here, since we dont need to solve for it.
// sanity check that we do have a tire pressure
if (lowestTirePressure === undefined)
throw new UnexpectedCodePath('no tire pressures found. can not compute lowest tire pressure', { tires });
With this, the type of lowestTirePressure
has been narrowed from number | undefined
to just number
, so you wont have any type errors anymore.
Further, if this case does occur in real life, then it will be really easy to debug what happened and why. Your error message will include the tires
input that caused the problem making this a breeze to debug. No more could not read property 'x' of undefined
!
The BadRequestError
is probably the next most common type of error you'll throw.
It's common in business logic that callers will try to execute your logic with inputs that are simply logically not valid. The user may not understand that their input is not valid or there may just be a bug upstream that is resulting in invalid requests.
For example, imagine you have an api that returns the liked songs of a user
const getLikedSongsByUser = ({ userUuid }: { userUuid: string }) => {
// lookup the user
const user = await userDao.findByUuid({ uuid: userUuid });
// if the user does not exist, this is an invalid request. we shouldn't be asked to lookup songs for fake users
if (!user)
throw new BadRequestError('user does not exist for uuid', { userUuid });
// use a property of the user to lookup their favorite songs
const songs = await spotifyApi.getLikesForUser({ spotifyUserId: user.spotifyUserId });
}
Whatever the reason for a caller making a logically invalid request, it's important to distinguish when your code is at fault versus when the request is at fault.
This is particularly useful when monitoring error rates. Its important to distinguish whether your software failed to execute
or whether it successfully rejected
the request for observability in monitoring for issues. The BadRequestError
enables us to do this easily
For example, libraries such as the simple-lambda-handlers leverage BadRequestErrors
to ensure that a bad request both successfully returns an error to the caller but is not marked as an lambda invocation error.
The HelpfulError
is the backbone of this pattern and is what you'll extend
whenever you want to create a custom error.
The purpose of this error is to be as helpful as possible to whoever has to read it when its thrown.
To fulfill this goal, the error makes it very easy to specify what the issue was as well as any other information that may be helpful to understanding why it occurred at the time. It then pretty prints this information to make it easy to read when observing.
throw new HelpfulError(
'the message of the error goes here',
{
context,
relevantInfo,
potentiallyHelpfulVariables,
goHere,
}
)
The getError
method is the cherry-on-top of this library.
When writing tests for logic that throws an error in certain situations, you may want to verify that the code indeed throws this error in a test.
The getError
utility makes it really easy to assert that the expected error is thrown.
Under the hood, it executes or awaits the logic or promise you give it as input, catches the error that is throw, or throws a NoErrorThrownError of its own. It does the legwork of handling all three cases you may need to use it in and defining the return type correctly.
usecase 1: synchronous logic
const doSomething = () => { throw new HelpfulError('found me'); }
const error = getError(() => doSomething())
expect(error).toBeInstanceOf(HelpfulError);
expect(error.message).toContain('found me')
usecase 2: asynchronous logic
const doSomething = async () => { throw new HelpfulError('found me'); }
const error = await getError(() => doSomething())
expect(error).toBeInstanceOf(HelpfulError);
expect(error.message).toContain('found me')
usecase 3: a promise
const doSomething = async () => { throw new HelpfulError('found me'); }
const error = await getError(doSomething())
expect(error).toBeInstanceOf(HelpfulError);
expect(error.message).toContain('found me')
The errors extended from the HelpfulError
include a .throw
static method for convenient usage with ternaries or condition chains
For example, instead of
const phone = customer.phoneNumber ?? (() => {
throw new UnexpectedCodePathError(
'customer has relationship without phone number. how is that possible?',
{ customer },
);
})();
You can simply write
const phone = customer.phoneNumber ?? UnexpectedCodePathError.throw(
'customer does not have a phone. how is that possible?',
{ customer },
);
The .cause parameter is a helpful feature of native errors. It allows you to chain errors together in a way that retains the full stack trace across errors.
For example, sometimes, the original error that your code experiences can be reworded to make it easier to debug. By using the .cause option, you're able to retain the stack trace and reference of the original error while throwing a new, more helpful, error.
// imagine you're using some api which throws an unhelpful error
const apiGetS3Object = async (input: { key: string }) => { throw new Error("no access") }
// you can catch and extend the error to add more context
const helpfulGetS3Object = async (input: { key: string }) => {
try {
await getS3Object();
} catch (error) {
if (error.message === "no access") throw HelpfulError("getS3Object.error: could not get object", {
cause: error, // !: by adding the "cause" here, we'll retain the stack trace of the original error
input,
})
}
}