-
-
Notifications
You must be signed in to change notification settings - Fork 2.8k
feat(eslint-plugin): add rule [strict-void-return] #9707
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
base: main
Are you sure you want to change the base?
Conversation
Thanks for the PR, @phaux! typescript-eslint is a 100% community driven project, and we are incredibly grateful that you are contributing to that community. The core maintainers work on this in their personal time, so please understand that it may not be possible for them to review your work immediately. Thanks again! 🙏 Please, if you or your company is finding typescript-eslint valuable, help us sustain the project by sponsoring it transparently on https://opencollective.com/typescript-eslint. |
❌ Deploy Preview for typescript-eslint failed.
|
View your CI Pipeline Execution ↗ for commit aad8977
☁️ Nx Cloud last updated this comment at |
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #9707 +/- ##
==========================================
+ Coverage 90.84% 90.91% +0.06%
==========================================
Files 501 504 +3
Lines 50919 51381 +462
Branches 8387 8506 +119
==========================================
+ Hits 46256 46711 +455
- Misses 4648 4655 +7
Partials 15 15
Flags with carried forward coverage won't be shown. Click here to find out more.
🚀 New features to boost your workflow:
|
👋 Exciting PR, really looking forward to the rule! Just marking as a draft because there are unit test failures. This keeps getting me excited when it pops up in my notifications 😄. Let us know if you want to talk or ask questions about any of them. Edit: ACK on the questions in the OP, I don't have the bandwidth to answer just now, but hopefully someone else does. Please ping us if those are blocking progress! |
Fixed and I'm not planning any more changes so I'm undrafting it I guess. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
OK! Very happy to have finally made it to this PR: it's a great piece of work. As you said, it covers a heck of a lot of functionality, and does so in ways that are really solid compared to previous approaches. Fantastic! 👏
It's also a lot of code that was hard to read through. For an initial version of the rule, I think nuance around suggestions aren't necessary. And the fixers would need to be suggestions given they change code behavior.
I left requests for simplification through the code: for messages, options, and the suggestions.
But, my advice would be to hold off applying that large set of removals until the conversation in https://github.com/typescript-eslint/typescript-eslint/pull/9707/files#r1741336663 is resolved. The consensus might end up being that the options are good and useful after all.
} | ||
}, | ||
AssignmentExpression: (node): void => { | ||
if (['=', '||=', '&&=', '??='].includes(node.operator)) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
[Testing] If I remove this if
and just leave its body, all unit tests still pass. That means there's at least a missing test case.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This is still the case.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Added a test but it doesn't matter:
declare let foo: () => void;
foo += () => 1;
still doesn't fail with condition removed.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Sounds to me like the if
can be removed, then! 🔪
Change request: remove the if
?
Just noting for posterity, node.operator
in an AssignmentExpression
can be any of:
"=" | "+=" | "-=" | "*=" | "**=" | "/=" | "%=" | "<<=" | ">>=" | ">>>=" | "&=" | "|=" | "||=" | "&&=" | "??=" | "^="
The operators not mentioned in the if
all produce TypeScript errors when given functions. So I agree with adding the valid
test case containing foo += () => 1
. That makes sure the removal of the if
doesn't break those cases.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
At this point this rule already did everything
no-misused-promises
'scheckVoidReturn
did, but better. It doesn't have problems #8054 or #8739. Maybe it's worth splittingno-misused-promises
into 3 separate rules in the future? (this one being one of them)
This is a great question. no-misused-promises
was already largely overlapped by no-unnecessary-condition
. This new strict-void-return
pretty much takes on the rest of no-misused-promises
, making no-misused-promises
redundant if you have both no-unnecessary-condition
and strict-void-return
...
I'd be in favor of deprecated no-misused-promises
in favor of using no-unnecessary-condition
+ strict-void-return
. The only benefit I can think of for no-misused-promises
would be projects that want to only apply the checks for Promises... Maybe these two rules could each be given some kind of "only check Promises" option?
Also of note is that no-misused-promises
's checkVoidReturns
is pretty configurable. Maybe, if this rule is to replace no-misused-promises
, it'd be useful to have each of those configurable options? Or, on the other hand, maybe those options are holdovers that real-world don't generally use? Investigation needed. I think those options can be a followup & shouldn't block this PR.
What do you think?
Also cc: @typescript-eslint/triage-team in general, and @kirkwaiblinger + @alythobani from #8765.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
no-misused-promises
was already largely overlapped byno-unnecessary-condition
. This newstrict-void-return
pretty much takes on the rest ofno-misused-promises
, makingno-misused-promises
redundant if you have bothno-unnecessary-condition
andstrict-void-return
...
Yeah the only thing left I think would be the checksSpreads
option:
const myPromise = Promise.resolve({num: 2, str: "2"});
const myObject = {...myPromise}; // Expected a non-Promise value to be spreaded in an object. eslint(@typescript-eslint/no-misused-promises)
I do agree it could make sense to replace checksVoidReturn
with strict-void-return
. Although there may be tradeoffs in terms of eng effort and/or UX complexity if we wanted to retain all the configurability on top of having an onlyChecksPromises
option.
As for checksConditionals
, I actually just found microsoft/TypeScript#34717 and microsoft/TypeScript#39175—looks like checkConditionals
has been covered by TypeScript for a couple years now :)
The only benefit I can think of for
no-misused-promises
would be projects that want to only apply the checks for Promises
Yeah e.g. one example I've seen when looking into this topic (void function assignability), is using push
with forEach
:
declare function forEach<T>(arr: T[], callback: (el: T) => void): void;
let target: number[] = [];
forEach([1, 2, 3], el => target.push(el)); // OK
It's possible some users would prefer to only check Promises so they can still use shorthands like the above without linter errors (and/or just mainly care about forgetting to await
Promises), in which case an onlyChecksPromises
option would be useful if we did replace checksVoidReturn
with strict-void-return
.
Maybe, if this rule is to replace
no-misused-promises
, it'd be useful to have each of those configurable options? Or, on the other hand, maybe those options are holdovers that real-world don't generally use?
It looks like #4619 was originally the impetus for adding the options (#4623); and based on the thread it looks like there are at least a few people who find the configurability you added very helpful!
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Maybe a lame response, but is there a compelling reason not to land this first, then consider the no-misused-promises deprecation, and which options we might need to port or create in order to do so, afterwards?
Just thinking, deprecating no-misused-promises might have some strings attached, such as some nontrivial updating of the docs in no-floating-promises that explain how to lint against promise antipatterns outside of ExpressionStatements.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Agreed with landing this first, then considering a deprecation as a followup.
In fact, this rule is pretty big and scary. We don't really have a process for declaring rules as "canary" or "experimental". #8676 is the closest we have to a feature request. Maybe we should set a precedent?
(I don't think this PR should be blocked on that)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Very cool PR, great job sticking it out and a lot of thought put into things!
Left some thoughts/questions, hope they're helpful 🌅
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
no-misused-promises
was already largely overlapped byno-unnecessary-condition
. This newstrict-void-return
pretty much takes on the rest ofno-misused-promises
, makingno-misused-promises
redundant if you have bothno-unnecessary-condition
andstrict-void-return
...
Yeah the only thing left I think would be the checksSpreads
option:
const myPromise = Promise.resolve({num: 2, str: "2"});
const myObject = {...myPromise}; // Expected a non-Promise value to be spreaded in an object. eslint(@typescript-eslint/no-misused-promises)
I do agree it could make sense to replace checksVoidReturn
with strict-void-return
. Although there may be tradeoffs in terms of eng effort and/or UX complexity if we wanted to retain all the configurability on top of having an onlyChecksPromises
option.
As for checksConditionals
, I actually just found microsoft/TypeScript#34717 and microsoft/TypeScript#39175—looks like checkConditionals
has been covered by TypeScript for a couple years now :)
The only benefit I can think of for
no-misused-promises
would be projects that want to only apply the checks for Promises
Yeah e.g. one example I've seen when looking into this topic (void function assignability), is using push
with forEach
:
declare function forEach<T>(arr: T[], callback: (el: T) => void): void;
let target: number[] = [];
forEach([1, 2, 3], el => target.push(el)); // OK
It's possible some users would prefer to only check Promises so they can still use shorthands like the above without linter errors (and/or just mainly care about forgetting to await
Promises), in which case an onlyChecksPromises
option would be useful if we did replace checksVoidReturn
with strict-void-return
.
Maybe, if this rule is to replace
no-misused-promises
, it'd be useful to have each of those configurable options? Or, on the other hand, maybe those options are holdovers that real-world don't generally use?
It looks like #4619 was originally the impetus for adding the options (#4623); and based on the thread it looks like there are at least a few people who find the configurability you added very helpful!
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Maybe a lame response, but is there a compelling reason not to land this first, then consider the no-misused-promises deprecation, and which options we might need to port or create in order to do so, afterwards?
Just thinking, deprecating no-misused-promises might have some strings attached, such as some nontrivial updating of the docs in no-floating-promises that explain how to lint against promise antipatterns outside of ExpressionStatements.
eeb8ae4
to
55b6378
Compare
Ready |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
OK! Sorry this took so long to get a review - it's a really wonderful new rule and has a lot of nuance that I wanted to give the attention it deserves. The functionality generally looks great to me, I think you nailed a lot of the hard-to-get-right cases. Awesome! 👏
I'm mostly requesting changes to address the logical branches that don't seem to impact unit tests when removed. For each, some code change should be done: either adding tests for a necessary runtime check or removing code (possibly also fiddling with types) for an unnecessary-at-runtime check.
I also took a pass at docs. They don't have to be perfect for merge, we can always touch them up as we go.
Thanks for bringing this back into review, it's a really exciting rule!
} | ||
}, | ||
AssignmentExpression: (node): void => { | ||
if (['=', '||=', '&&=', '??='].includes(node.operator)) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Sounds to me like the if
can be removed, then! 🔪
Change request: remove the if
?
Just noting for posterity, node.operator
in an AssignmentExpression
can be any of:
"=" | "+=" | "-=" | "*=" | "**=" | "/=" | "%=" | "<<=" | ">>=" | ">>>=" | "&=" | "|=" | "||=" | "&&=" | "??=" | "^="
The operators not mentioned in the if
all produce TypeScript errors when given functions. So I agree with adding the valid
test case containing foo += () => 1
. That makes sure the removal of the if
doesn't break those cases.
return ( | ||
// At least one return type is void | ||
returnTypes.some(type => | ||
tsutils.isTypeFlagSet(type, ts.TypeFlags.Void), |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
[Question] This uses Void
but the next check uses VoidLike
. Swapping to any other combination of the type flags doesn't produce any test failures. Was there a reason to choose one vs. the other?
I'm genuinely asking, I don't know 😄
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
dunno
if ( | ||
propNode.value.type === AST_NODE_TYPES.AssignmentPattern || | ||
propNode.value.type === AST_NODE_TYPES.TSEmptyBodyFunctionExpression | ||
) { | ||
return; | ||
} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
[Testing] Removing this if
check doesn't fail any unit tests. Either this check should exist and tests are missing, or this is dead code that can be removed.
Similar note to earlier on the type narrowing.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think in this case the types are wrong.
AFAIK it's not possible to have TSEmptyBodyFunctionExpression as an object's property value. { f() }
is an syntax error. This is only valid in an abstract class body.
AssignmentPattern is also not valid as value of Property of ObjectLiteral, but it is valid for Property of ObjectPattern. Types could be improved by splitting Property into 2 types.
I will add type assertions for now.
if (propNode.value == null) { | ||
return; | ||
} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
[Testing] Removing this if
check doesn't fail any unit tests. Either this check should exist and tests are missing, or this is dead code that can be removed.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
There is already a test for this and if I remove this check then the types are wrong. TS functions say they don't accept null but if you pass null they just do nothing and return more null so nothing crashes.
if ( | ||
methodNode.value.type === AST_NODE_TYPES.TSEmptyBodyFunctionExpression | ||
) { | ||
return; | ||
} | ||
if (methodNode.kind !== 'method') { | ||
return; | ||
} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
[Testing] Removing this if
checks doesn't fail any unit tests. Either they should exist and tests are missing, or this is dead code that can be removed.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Added test for empty body func.
Removing kind == 'method'
actually makes the rule catch more cases.
JSXAttribute: (node): void => { | ||
if ( | ||
node.value?.type === AST_NODE_TYPES.JSXExpressionContainer && | ||
node.value.expression.type !== AST_NODE_TYPES.JSXEmptyExpression |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
[Testing] Removing the && node.value... !== JSXEmptyExpression
doesn't fail any unit tests. Either this check should exist and tests are missing, or this is dead code that can be removed.
If the reason it's here is to narrow node.value.expression
's type down to TSESTree.Expression
then maybe either the types are suggesting a legit case to handle, or are wrong?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Added valid tests for JSXEmptyExpression (prop={}
) and JSXElement (prop=<elem/>
) but I dunno how to get JSXSpreadChild.
I tried prop={...value}
but it doesn't work.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Oh if that's just not doable in the AST then we can just ignore it. A good followup issue would be to tighten up the AST types to not include it.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Agreed with landing this first, then considering a deprecation as a followup.
In fact, this rule is pretty big and scary. We don't really have a process for declaring rules as "canary" or "experimental". #8676 is the closest we have to a feature request. Maybe we should set a precedent?
(I don't think this PR should be blocked on that)
Co-authored-by: Josh Goldberg ✨ <git@joshuakgoldberg.com>
Co-authored-by: Josh Goldberg ✨ <git@joshuakgoldberg.com>
Co-authored-by: Josh Goldberg ✨ <git@joshuakgoldberg.com>
Co-authored-by: Josh Goldberg ✨ <git@joshuakgoldberg.com>
Co-authored-by: Josh Goldberg ✨ <git@joshuakgoldberg.com>
Co-authored-by: Josh Goldberg ✨ <git@joshuakgoldberg.com>
Co-authored-by: Josh Goldberg ✨ <git@joshuakgoldberg.com>
PR Checklist
Overview
So basically I implemented the void checking as requested by #2988. For every checked node (arguments, assignments, returns, etc) I take the actual function type and the contextual function type and compare return types. Only object method shorthand required slightly different logic.
That already worked pretty well, but I also found #1744 and decided to include it in this rule as well, since I already had a similar thing implemented for object shorthand methods.
I added this as an optionconsiderBaseClass
andconsiderImplementedInterfaces
, enabled by default.Then I noticed that callback for
addEventListener
is not detected as void context. That's because it has another signature where the callback can returnany
. I was stuck on this for a long time. Ultimately I looked at howno-misused-promises
does this and implemented something similarasconsiderOtherSignatures
option, enabled by default.At this point this rule already did everything
no-misused-promises
'scheckVoidReturn
did, but better. It doesn't have problems like #8054 or #8739. Maybe it's worth splittingno-misused-promises
into 3 separate rules in the future? (this one being one of them)EDIT: autofixes removed for now
I also added many autofixes and suggestions. They are possible when the provided function is a function literal and we can inspect its body. Some of them are the same as in
no-confusing-void-expression
so I moved them into utils. It might make sense to change some autofixes into suggestions instead so they don't accidentally remove a big chunk of code. Let me know if that's a good idea.The biggest feature is automatic suggestions which I and probably others had to type manually a thousand of times:
into
or
To allow the second suggestion without the need of ignoring the line I added the option
allowReturnPromiseIfTryCatch
. It's just a simple extraif
near the end of the long routine that checks everything that could be wrong in the function body. I hope it can stay.