This code passes the flow check:
/* #flow */
function test (list: ?Array<string>): Promise<number> {
if(list !== null && list !== undefined) {
return Promise.resolve(list.length)
} else {
return Promise.resolve(0)
}
}
console.log(test(null))
Whereas the following gets a null check error
/* #flow */
function test (list: ?Array<string>): Promise<number> {
if(list !== null && list !== undefined) {
return Promise.resolve().then(() => list.length)
} else {
return Promise.resolve(0)
}
}
console.log(test(null))
error:
property `length`. Property cannot be accessed on possibly null value
Clearly list cannot be null so there must be something about the code structure that makes flow unable to recognise this.
I would like to understand what limitation I am hitting and how I can work around it. Thanks!
Basically, Flow doesn't know that your type refinement (null check) will hold at the time when () => list.length is executed. Inside that callback Flow only looks at the type of list – which says it can be null.
The difference between first and second snippet is that in the second snippet list is crossing a function boundary – you're using it in a different function than where you refined its type.
One solution is to extract list.length into a variable, and use that variable in the callback.
var length = list.length;
return Promise.resolve().then(() => length)
This might also work:
var list2: Array<string> = list;
return Promise.resolve().then(() => list2.length)
Note that this problem exists even for immediately invoked callbacks, e.g. when using map or forEach. There is an issue on flow's github about this, but I couldn't find it after a quick search.
Related
I'm using koa which has middleware props typed as mixed, so I'm trying to do something along the lines of the following below but I'm getting an error that Cannot call `ctx.render` because mixed [1] is not a function.Flow(not-a-function)
app.use(async (ctx, next) => {
// some other code above it
await ctx.render('index');
});
My question is, what's the correct way to do a type refinement that this is a function and then allow me to call it?
You can refine this to a function, but calling it is another matter.
app.use(async (ctx, next) => {
if (typeof ctx.render === 'function') {
// Now we know that `ctx.render` is a function.
}
});
Flow actually has a special case for this, this is called an "unknown function." We know that ctx.render is a function, but we don't know anything about its arguments or return type so we can't safely do anything with it except pass it around. How can we safely call ctx.render(1) if we don't know that ctx.render takes a number?
What's more, we can't know anything about it. There is no reflection mechanism provided by JavaScript that we could interrogate for enough information about this function to be able to safely call it. The only thing we can find is the static arity (ctx.render.length) but this by itself is not reliable or sufficient.
If we had more information, like say if this were a union type instead of mixed, then we could use type refinement to do what we want:
(arg: boolean | (number => void)) => {
if (typeof arg === 'function') {
arg(1); // safe because if this is a function, we know it takes a number
}
};
In this case the most reasonable solution is to type through any. Assuming that we know that we should only ever receive one type of render function, then we just forcibly cast it to that type with all the blaring caveats one would expect:
// I don't know what the type of your render function is, but you would put it
// here:
type RenderFunction = string => void;
app.use(async (ctx, next) => {
// some other code above it
if (typeof ctx.render === 'function') {
// DANGER: We make a hard assumption here that the render function is
// always of this specific type. If it is ever of any other type then
// behavior is undefined!
await ((ctx.render: any): RenderFunction)('index');
}
});
Also it sounds to me like the koa libdef could probably be improved upon.
Why even with an IF check, Flow still complains about a possibly null value
if(document && document.getElementById("myID") && document.getElementById("myID").offsetWidth){
console.log(document.getElementById("myID").offsetWidth);
}
Gives this error
^ property `offsetWidth`. Property cannot be accessed on possibly null value
Flow has no way to know that the success of first call to getElementById means that the later ones will also succeed. For all it knows, reading the offsetWidth property could cause getElementById to start returning null the next time it is called.
You'll need to store the value, e.g.
const myIdEl = document && document.getElementById("myID");
if(myIdEl && myIdEl.offsetWidth) {
console.log(myIdEl.offsetWidth);
}
this way there is no way for myIdEl to become null after it has been referenced.
For HTMLElement (and extensions of HTMLElement like VideoHTMLElement) in FlowType, I'd recommend using instanceof to validate the Type and to validate that it's not null.
Also, I don't believe you need to check if document exists, that is defined globally in flow (1)*
<HTMLElement> Example
const myIdEl: ?HTMLElement = document.getElementById('myID');
if (myIdEl instanceof HTMLElement) {
// continue
console.log(myIdEl.offsetWidth);
}
<HTMLSelectElement> Example
const selectEl: ?HTMLElement = document.getElementById('someSelectElement');
// Checks correct type (!null !undefined come for free)
if (selectEl instanceof HTMLSelectElement) {
const selectedVal = selectEl.options[selectEl.selectedIndex].value;
}
<HTMLVideoElement> Example using invariant
import invariant from 'invariant';
const videoContent = document.getElementById('video-player');
invariant(videoContent instanceof HTMLVideoElement, 'No video element');
// do stuff with video api
videoContent.volume = 0;
videoContent.plause();
https://github.com/facebook/flow/blob/f3f29f7fd8c5aa73ac5a8a546ccfbc29cd7505fe/lib/dom.js#L1288
In the following example, since I'm using matching over type of Message using the switch statement, I would like flow to recognise my incorrect case of 'ENUM_TYPO'. It currently doesn't.
type Message = 'BROADCAST_MESSAGE' | 'PRIVATE_MESSAGE';
const message: Message = 'BROADCAST_MESSAGE';
switch (message) {
case 'ENUM_TYPO':
// Do Broadcast
break;
default:
break;
}
As of v0.32.0, Flow does not complain about unreachable code, unless it's something like
// #flow
function foo() {
throw new Error();
return 123; // This will error
}.
However, consider the following code
// #flow
function foo(x: string): Object {
if (x === 123) {
return x;
}
return {};
}
Will currently will not error on this code. Flow does in fact notice that x === 123 will never be true. Inside the if block, Flow will refine the type of x to the empty type, since it doesn't believe that this code will ever be reached. That is why it doesn't complain about the return x statement.
One of the members of the Flow team is almost done with adding reachability analysis to Flow. Once this improvement lands (I'm guessing v0.34.0?), Flow will complain when it sees a conditional that it thinks will always fail. This will help you with your example, since switch statement cases are basically strict equality checks.
I'm trying to perform a custom sort using a comparator function from within a template helper in Meteor.
Here is my template helper:
Template.move_list.helpers({
assets() {
return Assets.find({}, { sort: sortFunction });
}
});
And here is the comparator function:
const sortFunction = function (doc1, doc2) {
const barcodes = Session.get('barcodesArray');
if (barcodes.indexOf(doc1.barcode) === -1 || barcodes.indexOf(doc2.barcode) === -1) {
return 0;
}
let last = null;
_.each(barcodes, function (barcode) {
if (barcode === doc1.barcode) last = doc1.barcode;
if (barcode === doc2.barcode) last = doc2.barcode;
});
return last === doc1.barcode ? 1 : -1;
}
Error
When the page loads, the following error is returned:
Exception in template helper: Error: Match error: Failed Match.OneOf, Match.Maybe or Match.Optional validation
I put a breakpoint in chrome into the sortFunction, however the function was never entered and the breakpoint never reached.
Of course, the error is not throw when I remove sort.
References
This feature is not very well documented, however here is the relevant part of the docs:
For local collections you can pass a comparator function which receives two document objects, and returns -1 if the first document comes first in order, 1 if the second document comes first, or 0 if neither document comes before the other. This is a Minimongo extension to MongoDB.
And the commit by mitar adding the functionality, with example code from the test:
var sortFunction = function (doc1, doc2) {
return doc2.a - doc1.a;
};
c.find({}, {sort: sortFunction})
Can anyone make sense of this error?
Edit:
This issue should be resolved in Meteor >= v1.3.3.1.
Local collections (i.e, client-side and in-memory server-side collections) will allow to pass a function as the sort clause.
The error comes from the mongo package, where the spec does not allow sort to be a function.
#mitar changed LocalCollection in the minimongo package. LocalCollection is part of the Mongo.Collection object on the client (its _collection attribute), but queries are still checked according to the original mongo spec. I believe this to be a bug, as the spec was not updated to reflect the change.
To overcome this (in the meantime), either have the function accept a sub-field, such that the sort value is an object:
var sortFunction = function (x, y) {
return x - y;
};
c.find({}, {sort: {a: sortFunction}});
or use the c._collection.find() instead, which will work (as far as I can tell), except it will not apply any transformations defined for the collection.
var sortFunction = function (doc1, doc2) {
return doc2.a - doc1.a;
};
c._collection.find({}, {sort: sortFunction});
I'm trying to do this relatively complex operation in BaconJs.
Basically, the idea is keep trying each check until you have a 'pass' status or they all fail. The catch is that 'pending' statuses have a list of Observables (built from jquery ajax requests) that will resolve the check. For performance reasons, you need to try each Observable in order until either they all pass or one fails.
Here's the full pseudo algorithm:
Go thru each check. A check contains an id and status = fail/pass/pending. If pending, it contains a list of observables.
If status = pass, then return the id (you're done!)
if status = fail, then try the next check
if status = pending
try each observable in order
if observable result is 'false', then try the next check
if reach end of observable list and result is 'true', then return the id (you're done!)
Here's the Bacon code. It doesn't work when the Observables are Ajax requests.
Basically, what happens is that it skips over pending checks....it doesn't wait for the ajax calls to return. If I put a log() right before the filter(), it doesn't log pending requests:
Bacon.fromArray(checks)
.flatMap(function(check) {
return check.status === 'pass' ? check.id :
check.status === 'fail' ? null :
Bacon.fromArray(check.observables)
.flatMap(function(obs) { return obs; })
.takeWhile(function(obsResult) { return obsResult; })
.last()
.map(function(obsResult) { return obsResult ? check.id : null; });
})
.filter(function(contextId) { return contextId !== null; })
.first();
UPDATE: the code works when the checks look like this: [fail, fail, pending]. But it doesn't work when the checks look like this: [fail, pending, pass]
I am more familiar with RxJS than Bacon, but I would say the reason you aren't seeing the desired behavior is because flatMap waits for no man.
It passes [fail, pending, pass] in quick succession, fail returns null and is filtered out. pending kicks off an observable, and then receives pass which immediately returns check.id (Bacon may be different, but in RxJS flatMap won't accept a single value return). The check.id goes through filter and hits first at which point it completes and it just cancels the subscription to the ajax request.
A quick fix would probably be to use concatMap rather than flatMap.
In RxJS though I would refactor this to be (Disclaimer untested):
Rx.Observable.fromArray(checks)
//Process each check in order
.concatMap(function(check) {
var sources = {
//If we pass then we are done
'pass' : Rx.Observable.just({id : check.id, done : true}),
//If we fail keep trying
'fail' : Rx.Observable.just({done : false}),
'pending' : Rx.Observable.defer(function(){ return check.observables;})
.concatAll()
.every()
.map(function(x) {
return x ? {done : true, id : check.id} :
{done : false};
})
};
return Rx.Observable.case(function() { return check.status; }, sources);
})
//Take the first value that is done
.first(function(x) { return x.done; })
.pluck('id');
What the above does is:
Concatenate all of the checks
Use the case operator to propagate instead of nested ternaries.
Fail or pass fast
If pending create a flattened observable out of check.observables, if they are all true then we are done, otherwise continue to the next one
Use the predicate value of first to get the first value returned that is done
[Optionally] strip out the value that we care about.
I agree with #paulpdaniels Rx-based answer. The problem seems to be that when using flatMap, Bacon.js won't wait for your first "check-stream" to complete before launching a new one. Just replace flatMap with flatMapConcat.
Thanks to #raimohanska and #paulpdaniels. The answer is to use #flatMapConcat. This turns what is basically a list of async calls done in parallel into a sequence of calls done in order (and note that the last "check" is programmed to always pass so that this always outputs something):
Bacon.fromArray(checks)
.flatMapConcat(function(check) {
var result = check();
switch(result.status) {
case 'pass' :
case 'fail' :
return result;
case 'pending' :
return Bacon.fromArray(result.observables)
.flatMapConcat(function(obs) { return obs; })
.takeWhile(function(obsResult) { return obsResult.result; })
.last()
.map(function (obsResult) { return obsResult ? {id: result.id, status: 'pass'} : {status: 'fail'}; });
}
})
.filter(function(result) { return result.status === 'pass'; })
.first()
.map('.id');