tryF()
is a simple wrapper for code possibly throwing exceptions, based on the idea of scala.util.Try.
Given a function (that could throw), it returns either the result of the function, or the error thrown (see examples for details).
Same for async-function, i.e.: basically promises.
Given the general idea of pure functions, every function should return a value, pure functions should not interrupt the execution by throwing an error.
Functional programming languages implement this by some specific type (different programming languages use different names, with sometimes different semantics, i.e. scala also has Either).
A very tangible advantage is, when using statically typed languages, the function-signature indicates that a function could fail. There's no way (in typescript) to indicate that a function might throw an error. At least not checked by the compiler.
Leveraging typescript's union-type it's easy to bring this concept to JavaScript, since functions may now e.g. return the original value or an error.
function getDataFromApi(id: number): Response | Error { }
the calling code now needs to check if the result is an instance of Error
or the real result. But typescript forces you
to check this, so there's no way you can miss that.
A wrapper for types. If a function might fail, the return type may be Try<T, E>
. E
is optional, if you want to further
specify the type of the error.
Instead of
function getDataFromApi(id: number): Response | Error { }
one can write
function getDataFromApi(id: number): Try<Response> { }
or even
function getDataFromApi(id: number): Try<Response, MyCustomError> { }
Please be aware that no runtime error-type-checking happens.
trivial typeguard to dry up code a little bit, given a value const myValue: Try<number> = get();
if(myValue instanceof Error) { /* error handling */ }
one can also write
if(isError(myValue)) { /* error handling */ }
The actual wrapper. See examples.
The test-file also shows all use-cases, but let me add some more context here.
This is what you'll want to use to wrap third-party code — for your own code you'll want to avoid throw
completely
(and use return new Error()
with function-return-type Try<T>
), thus you won't need tryF
at all.
const myData: Try<any> = tryF(() => JSON.parse(input));
if(isError(myData)) {
// no way to recover here, let the calling function know there's an error
return myData;
}
Combined with async/await, passing promises to tryF
is rather neat. This example wraps request-promise-native,
which rejects for non-2xx-status-codes (if simple: true
is set — which is the default).
const result: Try<Response> = await tryF(rp("http://example.com"));
if(isError(result)) {
// error-handling here,
return some-sane-default;
}
// use result here. Typescript "knows" it's a Response here
Let's break it down:
const requestP: Promise<Response> = rp("http://example.com"); // this promise might be rejected
const resultP: Promise<Try<Response>> = tryF(rp("http://example.com")); // the resolved promise is now either a Response, or an Error
const result: Try<Response> = await resultP; // "unwrap" the promise.
Compare this to the code without Try<T>
, but using await, it's so ugly, it makes me sad:
let response: Response; // uag, "let"!!
try {
// awaiting a promise that rejects requires try/catch
response = await rp("http://example.com");
} catch (err) {
// error-handling here,
return some-sane-default;
}
// use result here
One can also wrap async blocks. I think you should not wrap giant lambdas in tryF
, but refactor your code to just
wrap the necessary calls/promises in tryF
. It's the very same reason you should narrow down try-blocks to the bare minimum.
However:
const result: Try<unkown> = tryF(async () => {
const a = getA();
const b = getB();
// this could of course throw
const data = JSON.parse(a);
const body = await rp("http://example.com");
return transform(body);
});
I didn't find a module on npmjs which does exactly what this does, but there are still alternatives.
Of course it's also an option to do this manually. For promises it's basically:
promise.then(x => x + x).catch(err => err);
for try-catch, you probably end up with almost the same wrapper-function.
fp-ts emulates Either from
Scala. Either wraps values into objects. Thus it has
more convenient access (via isLeft()
etc.), but involves runtime overhead. tryF
/Try<T>
is compile-time overhead only.
Besides that, Try<T>
is fixed to errors, Either
allows generic types for "left" (which is the "failure"-case by convention).
When programming scala, there is no Try
is always better then Either
or vice versa, it depends on the use case.
I also played around with type Try<T> = Either<Error, T>
, but didn't like the result that much (for my code).