* Add some failing tests around transient symbols
* Working, but slower
* A class is much faster, apparently
* This is probably best?
* Back to multimap
* Go back to single symbol cache
* Revert now-unnecessary generics
* Rename and reorganize
* Fix weird compound condition
* Clean up
* Enable module specifiers for all auto imports
* Use isIncomplete
* isIncomplete continuation
* Lots of fixes
* Merged/transient symbol fixes, resolve all ambient module specifiers up front, pull as many as we want from cache
* Fix existing tests
* Start testing
* Add more tests
* Set cache attempt limit, update API baseline
* Fix a few tests
* Fix contextToken
* Split getModuleSpecifiers
* Unexport function
* Clean up importFixes
* Clean up completions
* Delete transient symbol assertion - fixing later
* Switch index signature storage to 'indexInfos: IndexInfo[]' property
* Accept new baselines
* Remove another usage of IndexKind enum
* Update getIndexedAccessType and resolveMappedTypeMembers
* Accept new baselines
* Update grammar checking for index signatures
* Accept new baselines
* Consider all index signatures in mapped types and union types
* Accept new baselines
* Update getIndexType
* Accept new baselines
* Intersect multiple applicable index signatures
* Use getApplicableIndexInfo instead of hardwired string/number handling
* Update index signature relationship checking
* Report type for which index signature is missing
* Report type for which index signature is missing
* Accept new baselines
* Make 'number' index signatures consistently apply to numeric strings
* Accept new baselines
* Update fourslash test
* Revise index constraint checking
* Accept new baselines
* Update error messages
* Accept new baselines
* Update type inference from index signatures
* Update isKnownProperty
* Update contextual typing based on index signatures
* Accept new baselines
* Support union types in index signature declarations
* Accept new baselines
* Check duplicate index signatures / remove redundant template literals from unions with string
* Accept new baselines
* Include key type in diagnostic / check symbol-named properties
* Accept new baselines
* Minor fix
* Add tests
* Accept new baselines
* Add optimized findApplicableIndexInfoForName
* Accept new baselines
* Another place we don't need to obtain literal type for property name
* Accept new baselines
* Don't create literal types that are going to be discarded
* Individual maps for string, number, bigint, and enum literal types
* Remove ineffective optimizations
* Accept new baselines
* Permit intersections as key types in index signatures
* Index expression in element access is template literal context
* Add tests
* Accept new baselines
* Symbol index signatures from object literals with computed symbol properties
* Accept new baselines
* Add more tests
* Accept new baselines
* Implement Go To Definition for all applicable index signatures
* Add fourslash test
* Accept new API baselines
* Always issue cannot find name did-you-mean error
This PR issues "cannot find ${name}, did you mean ${name}" errors for
identifiers and propery access expressions in JS files *without*
`// @ts-check` and without `// @ts-nocheck`. This brings some benefits of
Typescript's binder to all Javascript users, even those who haven't
opted into Typescript checking.
```js
export var inModule = 1
inmodule.toFixed() // errors on exports
function f() {
var locals = 2
locale.toFixed() // errors on locals
}
var object = {
spaaace: 3
}
object.spaaaace // error on read
object.spaace = 2 // error on write
object.fresh = 12 // OK, no spelling correction to offer
```
To disable the errors, add `// @ts-nocheck` to the file. To get the
normal checkJs experience, add `// @ts-check`.
== Why This Works ==
In a word: precision. This change has low recall — it misses lots
of correct errors that would be nice to show — but it has high
precision: almost all the errors it shows are correct. And they come
with a suggested correction.
Here are the ingredients:
1. For unchecked JS files, the compiler suppresses all errors except
two did-you-mean name resolution errors.
2. Did-you-mean spelling correction is already tuned for high
precision/low recall, and doesn't show many bogus errors even in JS.
3. For identifiers, the error is suppressed for suggestions from global files.
These are often DOM feature detection, for example.
4. For property accesses, the error is suppressed for suggestions from
other files, for the same reason.
5. For property accesses, the error is suppressed for `this` property
accesses because the compiler doesn't understand JS constructor
functions well enough.
In particular, it doesn't understand any inheritance patterns.
== Work Remaining ==
1. Code cleanup.
2. Fix a couple of failures in existing tests.
3. Suppress errors on property access suggestions from large objects.
4. Combine (3) and (4) above to suppress errors on suggestions from other, global files.
5. A little more testing on random files to make sure that precision
is good there too.
6. Have people from the regular Code editor meeting test the code and
suggest ideas.
* all (most?) tests pass
* NOW they all pass
* add tonnes of semi-colons
* restore this.x check+add a test case
* make ts-ignore/no-check codefix work in unchecked js
* Issues errors only in the language service
* add a few more tests
* fix incorrect parentheses
* More cleanup in program.ts
* Improve readability of isExcludedJSError
* make diff in program.ts smaller via closure
* Switch unchecked JS did-you-mean to suggestion
Instead of selectively letting errors through.
* undo more missed changes
* disallow ignoring suggestions
* Issue different messages for plain JS than others
Straw text for the messages, I just changed the modals to avoid name
collisions.
Function extraction failed when using two identical `TypeLiteral`s ---
the problem is that the parameters' `.type.members` are shared in their
two occurences in the resulting code, and when the formatter bangs
position properties on the second, it's actually changing the first too.
* `typeToAutoImportableTypeNode`: wrap in `getSynthesizedDeepClone` to
avoid the offeding sharing.
* `getSynthesizedDeepCloneWorker`: add optional nodes/token visitors so
it doesn't skip cloning the above.
* A few very minor tweaks which I saw when tracing this (comment update,
two `!`s).
Fixes#44301
* Add cache invalidation for node_modules, config, and preferences changes
* Share watches with closed script info
* Update API baseline
* Small adjustments for fewer object allocations
* Document overloaded return value
* Update baselines
* Store isAutoImportable separately from modulePaths
* Add back missing return
* Return wrapped watcher instead of underlying one
* Make user preferences part of the cache key instead of implicitly clearing in editor services
* Fix missed merge conflict
* Fix discovery of more pnpm symlinks
* Add some tests
* Never show pnpm paths in auto imports, even if there’s no other path
* Import statement completions can return none
* Fix tests
* Add failing test showing poor symlink cache reuse
* Fix test, fails for right reasons now
* Preserve cache built up during program creation, then fill in with program resolutions
* Remove obsolete comment
* Remove obsolete type assertion
* Revert fully filtering out ignored paths
* Add @linkcode and @linkplain tags
They are just like @link tags but request fixed-width and normal
presentation, respectively.
Fixes#43935
* revert JSDocComment -> JSDoc SyntaxKind rename
* update API baselines
* fix lint
* Cache accessibe symbol chains, type parameter name generation
* Move signature declaration helper length approximation to start of function
* Add node result caching internal to `typeToTypeNodeHelper`
* Suggestion from PR
* WIP
* WIP
* Get completion details working
* Start unifying eager and lazy auto imports
* Fix export=
* Fix completion details for totally misspelled names
* Almost fixed duplication...
* Fix remaining completion tests
* Refactor to support multiple origins for same symbol
* Make import fixes make slightly more sense
* Add cache back in
* Set insertText based on import kind
* Update API baselines
* Add semicolons, snippet support, and sourceDisplay
* Add some tests
* Update baselines
* Fix pattern ambient modules appearing in auto imports
* Fix tests
* Remove commented code
* Switch to valueDeclaration for getting module source file
* Small optimizations
* Cache module specifiers / importableness and export map separately
* Fix and test cache invalidation logic
* Update API baselines
* Add separate user preference for snippet-formatted completions
* Require first character to match when resolving module specifiers
* Fix AutoImportProvider export map cache invalidation
* Really fix auto import provider export map invalidation
* Update test added in master
* Use logical or assignment
Co-authored-by: Daniel Rosenwasser <DanielRosenwasser@users.noreply.github.com>
* Simply conditional by reversing
Co-authored-by: Daniel Rosenwasser <DanielRosenwasser@users.noreply.github.com>
* When file is deleted need to marked correctly in the project as removed file
* Simplify hasAddedOrRemovedSymlinks with cherry-picked fix
* Ensure replacement range is on one line
* Update baselines
Co-authored-by: Daniel Rosenwasser <DanielRosenwasser@users.noreply.github.com>
Co-authored-by: Sheetal Nandi <shkamat@microsoft.com>
A misbehaved client can sometimes cause the server to reach
`deleteDeclaration` with the SourceFile, and it will crash due to no
`node.parent`. I couldn't find a good way to create a test for it, but
I could trigger it manually by having a file with just a `,`, and
sending an explicit `getCodeFixes` command to the server with
`errorCodes: [6133]`.
Do three things to improve this:
1. `textChanges.ts`: if we get here with the root node, delete it
instead of failing.
2. `fixUnusedIdentifier.ts`: check that we don't `delete` a node that is
the whole source file, so the error is more focused (should have more
similar failure stacks).
3. `session.ts`: when there was any failure in `getCodeFixes`, check if
the input had a diag code that does not appear in the requested text
range, and throw an error saying that the failure is probably a
result of a bad request.
Closes#33726 (probably not fixing it, but making it easier to find the
cause)
* Revert "Revert "Editor support for link tag (#41877)" (#43302)"
This reverts commit 451d4354b9.
* Fix parsing @link at end of comment
* Parse comments as string when no @link occurs
* fix lint
* Initial scribbles
* Compiles but provides spans instead of location pairs
Probably need to fork the services/server types and provide a conversion
with Session.toFileSpan. Not sure where to put the conversion.
* Switch to DocumentSpan
In theory this is already better supported, but not sure practise bears
that out.
* Builds w/protocol types + conversions
* cleanup:better names and scrub TODOs
* fix test harness too
* Misc
1. Simplify protocol after talking to @mjbvz.
2. Add more tests.
3. Initial notes about where to add parsing.
* Parse and store links in the compiler
The text of the link is still stored in the comment text, but that's now
kept in an object instead of just a string. Each link has the parse for
the entity reference, if there is one.
Needs lots more tests -- this just makes all the existing jsdoc tests
pass.
* more tests and some fixes
* Fix other failing tests
* fix bad merge
* polish parser
* improve names and array types
* slight tweaks
* remove some done TODOs
* more tests + resulting fixes
* add+fix cross-module tests
* Support `@see {@link`
Plus find-all-refs support equivalent to @see's.
* add server test
* Make comments actually part of the AST
* Add span for link.name in language service/protocol
* Make checker optional in getJSDocTags
Also change to JSDocCommentText from JSDocCommentComment
* Use getTokenValue instead of getTokenText
Measure twice, slice once
* Add missing support for top-level links
The language service and protocol were missing support for top-level
links. This commit adds that plumbing.
* add string back to comment type in node constructors
* Full parse of link tags and jsdoc comment text
- Doesn't pass fourslash yet, I'm going to switch to baselines for
failures there.
- Still needs some work on the protocol to convert file+offset to
file+line+offset.
* fix lint
* Fix missing newlines in inferFromUsage codefix
* Parse jsdoc comments as text node/link array
And switch to line+character offsets in the protocol
* Fix fourslash tests
Mostly ones that can't be baselined, but I switched a couple more over
to baselines
* Improve types and documentation
* Test+fix @link emit, scrub other TODOs
* update API baselines
* test that goto-def works with @link
* Split link displaypart into 3
One for link prefix and suffix, one for link name, and one for link
text.
* update baselines
* Provide JSDocTagInfo.text: string to full clients by default
Instead of upgrading them to displayparts.
* Real server tests
* Disambiguate {@link} and @param x {type}
They are ambiguous; previously the parser preferred the type
interpretation, but will now look ahead and parse links instead when the
prefix is `{@link`.
* Add explanatory comment in test
* fix location in richResponse in protocol
* update API baseline
* Address PR comments
1. Add a cross-file goto-def test.
2. Switch from per-message args to UserPreference.
* use arraysEqual from core
* About halfway through the checker
I'm going to merge with master to avoid clashing with the declaration
fix.
* Add undefined to Symbol.valueDeclaration
Also add undefined to a number of utility functions that have always
accepted it, but never added it to their type.
* Fix lint from code review
Co-authored-by: Daniel Rosenwasser <DanielRosenwasser@users.noreply.github.com>
* remove obsoleted fix from inferFromUsage
Co-authored-by: Daniel Rosenwasser <DanielRosenwasser@users.noreply.github.com>
* Add undefined to Symbol.declarations' type
Symbol.declarations now has type `Declaration[] | undefined`.
I made a mistake somewhere in the checker related to JS checking, so
there are quite a few test failures right now.
* undo clever change to getDeclaringConstructor
* Address PR comments
1. More early-returns.
2. More line breaks.
* Refactor emit substitution into transform
* Add reusable state machine for binary expressions
* Allow emitBinary to use state machine for comments/sourcemaps
* Switch core trampoline state back to arrays
* Switch binder to parallel stacks, temporarily partially revert emitBinary
* Add link to benchmark when posting perf results
* Ensure work stacks are per-execution
* Reenable comments and sourcemaps
