Open asterkin opened 4 months ago
Apparently, the inflight member fields are documented in https://www.winglang.io/docs/language-reference#33-preflight-classes. I would recommend at least providing a clear reference from https://www.winglang.io/docs/concepts/inflights#inflight-code and a warning about potential brain split problems.
Hi,
This issue hasn't seen activity in 90 days. Therefore, we are marking this issue as stale for now. It will be closed after 7 days. Feel free to re-open this issue when there's an update or relevant information to be added. Thanks!
I tried this:
pub struct Ticket { ticketCode: str; carPlate: str; rateName: str; }
pub class MockDataStore { inflight _tickets: MutMap;
}
This happened:
bring cloud; bring expect; bring "./MockDataStore.w" as dataStore;
let _dataStore = new dataStore.MockDataStore(); let _carPlate = "6989GPJ"; let _rateName = "Green"; let _ticketCode = "0001";
let _buyTicket = new cloud.Function(inflight () => { let ticket = dataStore.Ticket { ticketCode: _ticketCode, carPlate: _carPlate, rateName: _rateName }; _dataStore.storeTicket(ticket); }) as "BuyTicket";
let _checkCar = new cloud.Function((inflight () => { let actual = _dataStore.findTickets(_carPlate, _rateName); let expected = Set[
dataStore.Ticket {
ticketCode: _ticketCode,
carPlate: _carPlate,
rateName: _rateName
}
];
log(Json.stringify(actual.toArray()));
log(Json.stringify(expected.toArray()));
expect.equal(actual, expected);
})) as "CheckCar";
test "it will fail even in local" { _buyTicket.invoke(); _checkCar.invoke(); }
I expected this:
As a naive user, I might expect that it will work, at least locally. Using RAM-only mock implementations of DataStore objects is common in mainstream programming.
Is there a workaround?
Using Table instead, but that inflates the "stuff vs value" ratio.
Anything else?
The feature is not documented in https://www.winglang.io/docs/concepts/inflights#inflight-code. It works and is useful for keeping intermediate test results and understanding why and where it works and where it does not, requires a good understanding of how the Cloud Function memory is organized. For example, it would work for a Lambdalith local always, and remote sometimes depending on load (e.g. if because of load there are two instances of the same Cloud Function, it will not work). The same is true for a Container-based implementation.
It's a tough case of potentially leaky abstraction. I would probably remove it altogether. For keeping intermediate test results, resolving the https://github.com/winglang/wing/issues/6673 would be enough.
Wing Version
0.74.68
Node.js Version
22.1.0
Platform(s)
No response
Community Notes