-
Notifications
You must be signed in to change notification settings - Fork 785
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
feat(typescript): typescript 4.5 support #3205
Conversation
update client-patch-browser.ts patching of `performance.mark()` & `performance.measure()` to be ignored by the type checker. this patch allows stencil to support Safari 10; since we don't have access to a non-existent performance buffer and changing the return value for either of these functions could be considered a breaking chamge, defer altering the (non-compliant) functions until stencil drops support for Safari 10
`reload` has no parameter, although Firefox supports a non-standard 'forceGet' boolean parameter for bypassing its cache. for all other browsers, this is otherwise ignored. these arguments are used in the context of stencil's dev-server (IE are only used at dev time). given these two statements, the risk of a truly breaking change is deemed minimal here
set the return type of `whenDefined` to be `Promise<CustomElementConstructor>` over `Promise<void>` to meet new type checking expectations with TS v4.4. this leads to a small functional change in that the return value for this function when the internal registry contains an entry for a given `tagName`, the function now returns the constructor associated with the tag name, rather than void.
update the function signatures of mock-doc's performance implementation by declaring the return type on `mark()` and `measure()` to align with the `Performance` interface, but _continue to return undefined_, ignorning the type checker errors. stencil's stubs for these functions assume access to underlying browser api's (namely a performance buffer). given that older browsers may not support this functionality and implementating a mock version of these functions falls well outside the scope of a typescript update, simply ignore the errors. we take this approach, rather than: ```typescript // @ts-ignore mark(): void { /* no-op */ } ``` as stencil can (and does) instantiate an instance of `MockPerformance` and attempts to assign it to variable(s) that are of type `Performance`. this circumvents the type checker (although stencil was lying to it already) to prevent additional `// @ts-ignore` pragmas this may result in typing errors for consumers, however the actual values shall remain the same.
enable `useUnknownInCatchVariables` in all TSConfig files, with the exception of the main `tsconfig.json`. Although flipping that bit was originally going to be a part of the TS 4.4 upgrade, our decision to upgrade from 4.3 -> 4.5 directly increased scope a bit. That and the fact there are ~90 places where we want to rework the code, I feel more comfortable splitting that off into a separate effort
i'm concerned about the `matchFile` call and how the method changes for v4.4, then is reverted in v4.5. based on the conversations/issues in GH surrounding these changes, I'm strongly considering moving directly to v4.5. pause here to investiage those changes
update karma-typescript with monounity/karma-typescript@f4cdc43 update TypeScript to v4.5.4 remove the temporary hack during the TS v4.4 upgrade for `matchFile` call to succeed. update single unit test that was failing following the upgrade. update typescript import statements that were broken as a result of the upgrade to handle `AssertClause`s. move typescript calls to use factory methods
tagName = tagName.toLowerCase(); | ||
|
||
if (this.__registry != null && this.__registry.has(tagName) === true) { | ||
return Promise.resolve(); | ||
return this.__registry.get(tagName).cstr; |
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'm not completely sure what the type of cstr
is, but assuming it's CustomElementConstructor
I would expect this to still return a promise:
return this.__registry.get(tagName).cstr; | |
return Promise.resolve<CustomElementConstructor>(this.__registry.get(tagName).cstr); |
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.
Good call - technically it may not be a promise. Updated in b365de1
Pull request checklist
Please check if your PR fulfills the following requirements:
npm run build
) was run locally and any changes were pushednpm test
) were run locally and passednpm run test.karma.prod
) were run locally and passednpm run prettier
) was run locally and passedPull request type
Please check the type of change your PR introduces:
What is the current behavior?
We currently support v4.3 of TypeScript
GitHub Issue Numbers:
'useUnknownInCatchVariables'
. #3170What is the new behavior?
This PR adds support for TypeScript 4.5. I originally intended to only bump Stencil to v4.4, but there were some changes to a private TypeScript API Stencil uses (and shouldn't, more on that later) in TS v4.4 that were reverted in v4.5 (again, more on that later). Below is a summary of each commit by name (should I rebase this and mess up the SHA links)
chore(dependencies): upgrade typescript to v4.4.4
Version bump us to v4.4.4 - this was before I realized we'd have a problem with 4.4. This commit will later be superseded by the upgrade to 4.5.
update client browser patching
update client-patch-browser.ts patching of
performance.mark()
&performance.measure()
to be ignored by the type checker. this patchallows stencil to support Safari 10; since we don't have access to a
non-existent performance buffer and changing the return value for either
of these functions could be considered a breaking change, defer altering
the (non-compliant) functions until stencil drops support for Safari 10
remove non-compliant window.location.reload() arguments
reload
has no parameter, although Firefox supports a non-standard'forceGet' boolean parameter for bypassing its cache. for all other
browsers, this is otherwise ignored. these arguments are used in the
context of stencil's dev-server (IE are only used at dev time). given
these two statements, the risk of a truly breaking change is deemed
minimal here
update fn signature of mock-doc's customElementRegistry#whenDefined
set the return type of
whenDefined
to bePromise<CustomElementConstructor>
overPromise<void>
to meet newtype checking expectations with TS v4.4. this leads to a small
functional change in that the return value for this function when the
internal registry contains an entry for a given
tagName
, the functionnow returns the constructor associated with the tag name, rather than
void.
update mock-doc performance function signatures
update the function signatures of mock-doc's performance implementation
by declaring the return type on
mark()
andmeasure()
to align withthe
Performance
interface, but continue to return undefined,ignorning the type checker errors.
stencil's stubs for these functions assume access to underlying browser
api's (namely a performance buffer). given that older browsers may not
support this functionality and implementating a mock version of these
functions falls well outside the scope of a typescript update, simply
ignore the errors.
we take this approach, rather than:
as stencil can (and does) instantiate an instance of
MockPerformance
and attempts to assign it to variable(s) that are of type
Performance
. this circumvents the type checker (although stencil waslying to it already) to prevent additional
// @ts-ignore
pragmasthis may result in typing errors for consumers, however the actual
values shall remain the same.
window - add ts-ignore to handle mock-window & window type diffs
In various parts of the codebase, Window is used as a typing when the actual concrete type is a MockWindow - this leads to some tricky spots where when the TS compiler sees these as two distinct, unrelated types, calling a MockWindow method on an object typed as Window leads to compiler errors/adding @ts-ignore pragmas to the code. Created a ticket for us to revisit this
useUnknownInCatchVariables
enable
useUnknownInCatchVariables
in all TSConfig files, with theexception of the main
tsconfig.json
. Although flipping that bit wasoriginally going to be a part of the TS 4.4 upgrade, our decision to
upgrade from 4.3 -> 4.5 directly increased scope a bit. That and the
fact there are ~90 places where we want to rework the code, I feel more
comfortable splitting that off into a separate effort
[PAUSE] Stop current line of work, look into TS 4.5 level of effort
i'm concerned about the
matchFile
call and how the method changes forv4.4, then is reverted in v4.5. based on the conversations/issues in GH
surrounding these changes, I'm strongly considering moving directly to
v4.5. pause here to investigate those changes
TO BE CLEAR - Stencil needs to move off this API. I have a story in the backlog for us to do this
TS 4.5
update karma-typescript with monounity/karma-typescript@f4cdc43 - this is necessary due to internal changes made in TypeScript 4.5.
update TypeScript to v4.5.4
remove the temporary hack during the TS v4.4 upgrade for
matchFile
call to succeed.
update single unit test that was failing following the upgrade.
update typescript import statements that were broken as a result of the
upgrade to handle
AssertClause
s. move typescript calls to use factorymethods
Does this introduce a breaking change?
Yes
No - I'm following the same guidance we have for previous upgrades here. We treat TS upgrades as 'minor'/'non-breaking', just as the TS team does.
TypeScript 4.4 breaking changes
TypeScript 4.5 breaking changes
Because there's a little bit of a chance there could be a less than smooth upgrade here, I wanna hold off on adding this to the Stencil v2.13 release, and slate this for v2.14
Testing
In addition to unit/karma tests helping us out here, I also installed this branch's generated tarball (
npm run build && npm pack
) in Ionic Framework's core module, then built Framework/ran its tests. For better or worse, the source of most breaking changes would come from the TS upgrade itself (rather than changes we made to accommodate the upgrade, since there are a fewts-ignore
pragmas in here)