Limit unsound indexed access type relations - #27490
Limit unsound indexed access type relations#27490Anders Hejlsberg (ahejlsberg) merged 4 commits into
Conversation
Sorry, something went wrong.
There was a problem hiding this comment.
While I wholeheartedly agree with chipping away at how unsound we are here (and this looks fine in that regard), I do worry about the knock-on effects in our ecosystem (especially around, eg, react). What I've seen is that when types start getting complex (eg, generic indexes + other generic type constructors) people stop understanding them and simply (over)specify them until they work without quite understanding what shape/procedure they've described. We should probably do a diff on DT with this change to see if it's as bad as I think it might be, or if I'm overthinking it. Not to see if it's a good change, but to see if we need to identify and fix any libraries which embed a similar problem to this and accidentally or intentionally rely on it before we ship it in a full release.
Sorry, something went wrong.
| function f90<T extends S2, K extends keyof S2>(x1: S2[keyof S2], x2: T[keyof S2], x3: S2[K]) { | ||
| x1 = x2; | ||
| x1 = x3; | ||
| x1 = x4; |
There was a problem hiding this comment.
Might be worth keeping all these variants (maybe in a different file or function) just to document that they're expected to fail now (and so we know if they suddenly start succeeding again).
Sorry, something went wrong.
There was a problem hiding this comment.
They're basically covered by the new tests in keyofAndIndexedAccessErrors.ts, so I think it is fine to get rid of them.
Sorry, something went wrong.
|
Wesley Wigham (@weswigham) Do we have an automated way of running the DT tests? (Similar to the RWC bot.) Otherwise, what's the best way to go about it? |
Sorry, something went wrong.
|
Anders Hejlsberg (@ahejlsberg) No, the bot doesn't have a build for baselining dt. I can probably whip one up, though I don't think it'll be able to have an automatic PR/diff viewing system like RWC does. In the absence of that, running jake runtests-parallel ru=dt on master, accepting the initial baseline snapshot, then running it again on this branch would be a way to go about it locally; but you probably want someone with a real beefy machine to do it, since they take quite a long time compared to RWC, even. |
Sorry, something went wrong.
Like yours? (Hint, hint) |
Sorry, something went wrong.
|
I knew what I was getting into when I said what I said. XD |
Sorry, something went wrong.
|
And 4 hours later the results are in. My fears seem overblown, this change doesn't seem to affect anything on DT. For reference, my motivating issue for my worry was #27201, wherein what people are trying to do might be affected by this. TypeScript Bot (@typescript-bot) test this just in case tho |
Sorry, something went wrong.
|
Heya Wesley Wigham (@weswigham), I've started to run the extended test suite on this PR at 69cd6c0. You can monitor the build here. It should now contribute to this PR's status checks. |
Sorry, something went wrong.
|
I just tried this in my project, and it indeed does fix #27470! 🎉 |
Sorry, something went wrong.
|
I think rwc on this PR is fine; this PR is just missing a change in master (semicolon in a helper) which caused an RWC change. Everything looks good. |
Sorry, something went wrong.
Fixes #27470.