Share
Embed
K2への道。コンパイラを作り直すってどういうこと?
今年の6月にリリースされたKotlin 1.7には、K2という丸ごと新しいKotlinコンパイラがプレビューされました。
これはJetBrainsの10年以上のKotlin開発経験を活かし、改めて、発展性、安定性、そして何よりもパフォーマンスを上げるのを目標にしたプロジェクトの成果なのです。
本セッションでは、以前のコンパイラとの仕組みを比較をしながら、一般のユーザーにとって何が変わるのか、について発表します。
IntelliJ IDEAのKotlinプラグインの開発者として、プラグイン側の改善についても話したいと思います。
具体的な説明も多少入りますが、コンパイラ内部をあんまり触ったことがない人でも理解出来ると思いますので、是非お越しください。
Yan
December 09, 2022
Transcript
76 slides
-
How does the Kotlin compiler work? And the Kotlin IntelliJ
IDEA Plugin? Why K2? -
Lexer fun factorial ( n : Int ) : Int
{ if ( n == 0 return 1 return 1 ( .. n ) . reduce { acc , i -> acc * i } } fun factorial(n: Int): Int { if (n == 0) return 1 return (1..n).reduce { acc, i -> acc * i } } -
AST Parser fun factorial(n: Int): Int { if (n ==
0) return 1 return (1..n).reduce { acc, i -> acc * i } } if (n == 0) return 1 fun factorial(n: Int): Int { … } { … } return (1..n).reduce { … } n: Int Int n == 0 return 1 1 (1..n).reduce { … } { acc, i -> acc * i } 1..n 1 n .. acc i acc * i n 0 == reduce { … } acc i * Int n -
Semantic Analyzer fun factorial(n: Int): Int { if (n ==
0) return 1 return (1..n).reduce { acc, i -> acc * i } } if (n == 0) return 1 fun factorial(n: Int): Int { … } { … } return (1..n).reduce { … } n: Int Int n == 0 return 1 1 (1..n).reduce { … } { acc, i -> acc * i } 1..n 1 n .. acc i acc * i n 0 == reduce { … } acc i * Int n kotlin.Int kotlin.ranges.IntRange kotlin.collections.reduce<Int, Int>() kotlin.Int.times() kotlin.Int.equals() -
Semantic Analyzer fun factorial(n: Int): Int { if (n ==
0) return 1 return (1..n).reduce { acc, i -> acc * i } } if (n == 0) return 1 fun factorial(n: Int): Int { … } { … } return (1..n).reduce { … } n: Int Int n == 0 return 1 1 (1..n).reduce { … } { acc, i -> acc * i } 1..n 1 n .. acc i acc * i n 0 == reduce { … } acc i * Int n kotlin.Int kotlin.ranges.IntRange kotlin.collections.reduce<Int, Int>() kotlin.Int.times() kotlin.Int.equals() Operator call resolution Type resolution Function call resolution Type inference <Int, Int> Return type checking -
Semantic Analyzer fun factorial(n: Int): Int { if (n ==
0) return 1 return (1..n).reduce { acc, i -> acc * i } } if (n == 0) return 1 fun factorial(n: Int): Int { … } { … } return (1..n).reduce { … } n: Int Int n == 0 return 1 1 (1..n).reduce { … } { acc, i -> acc * i } 1..n 1 n .. acc i acc * i n 0 == reduce { … } acc i * Int n kotlin.Int kotlin.ranges.IntRange kotlin.collections.reduce<Int, Int>() kotlin.Int.times() kotlin.Int.equals() Operator call resolution Type resolution Function call resolution Type inference <Int, Int> Return type checking The same is happening inside IntelliJ IDEA -
🙀 Double Life Compiler frontend IDE plugin frontend 🥴
🥵 🥶 🤬 😵💫 🤕 💀 🎃 fixme -
PSI Program Structure Interface PsiElement PsiClass PsiMethod PsiMember PsiField PsiExpression
PsiCallExpression PsiLiteralExpression PsiUnaryExpression PsiPoliadicExpression PsiNewExpression PsiMethodCallExpression PsiWhiteSpace PsiComment Java PSI (simpli fi ed) public interface PsiElement extends UserDataHolder, Iconable { Project getProject(); Language getLanguage(); PsiElement[] getChildren(); PsiElement getParent(); PsiFile getContainingFile(); PsiElement findElementAt(int offset); TextRange getTextRange(); String getText(); void accept(PsiElementVisitor visitor); void acceptChildren(PsiElementVisitor visitor); PsiElement copy(); PsiElement add(PsiElement element); PsiElement addBefore(PsiElement element, PsiElement anchor); PsiElement addAfter(PsiElement element, PsiElement anchor); void delete(); PsiElement replace(PsiElement newElement); } Root of PSI hierarchy PsiFile (simpli fi ed) -
Use IR? Intermediate Representation Without IR Lexer Parser Semantic Analyzer
Tokens AST Backend AST Semantics Sources Lexer Parser Semantic Analyzer Tokens Backend IR Sources IR Converter AST IR With IR -
Use IR? Intermediate Representation Without IR Lexer Parser Semantic Analyzer
Tokens AST Backend AST Semantics Sources Lexer Parser Semantic Analyzer Tokens Backend IR Sources IR Converter AST IR With IR -
Is AST enough? abstract class Base { fun perform() {}
protected abstract fun perform(ctx: Context) } class Impl: Base() { override fun perform(ctx: Context) {} } public? protected? -
No, it’s not abstract class Base { fun perform() {}
protected abstract fun perform(ctx: Context) } class Impl: Base() { protected override fun perform(ctx: Context) {} } Redundant visibility modi fi er -
Semantics -Types of everything -Calls -Implicit declaration properties (visibility, modality)
-Data fl ow -A lot more open, abstract, final etc. -
BindingContext public interface BindingContext { <K, V> V get(ReadOnlySlice<K, V>
slice, K key); } public interface BindingTrace { BindingContext getBindingContext(); <K, V> void record(WritableSlice<K, V> slice, K key, V value); } (simpli fi ed) And its mutable version, BindingTrace -
Compiler Work fl ow Lexer & Parser Semantic Analyzer .kt
PSI Backend PSI BindingContext Binary -
Compiler IDE Full analysis Analyzes fi les completely Partial analysis
Analyzes only requested declarations -
Compiler IDE Full analysis Analyzes fi les completely Partial analysis
Analyzes only requested declarations Single module at a time Either module sources or binary libraries All project at once Cross-module dependencies -
Compiler IDE Full analysis Analyzes fi les completely Partial analysis
Analyzes only requested declarations Single module at a time Either module sources or binary libraries All project at once Cross-module dependencies Simple error handling Report and die Sustainable error handling Cope with local errors here and there -
Compiler IDE Full analysis Analyzes fi les completely Partial analysis
Analyzes only requested declarations Single module at a time Either module sources or binary libraries All project at once Cross-module dependencies Simple error handling Report and die Sustainable error handling Cope with local errors here and there Compile and exit No caches, simple fl ow Live forever Cache invalidation -
Compiler IDE Full analysis Analyzes fi les completely Partial analysis
Analyzes only requested declarations Single module at a time Either module sources or binary libraries All project at once Cross-module dependencies Simple error handling Report and die Sustainable error handling Cope with local errors here and there Compile and exit No caches, simple fl ow Live forever Cache invalidation -
StorageManager interface StorageManager { fun <K, V : Any> createMemoizedFunction(compute:
(K) -> V): MemoizedFunctionToNotNull<K, V> fun <K, V : Any> createMemoizedFunctionWithNullableValues(compute: (K) -> V?): MemoizedFunctionToNullable<K, V> fun <T : Any> createLazyValue(computable: () -> T): NotNullLazyValue<T> fun <T : Any> createNullableLazyValue(computable: () -> T?): NullableLazyValue<T> } (simpli fi ed) -
public void resolveMemberHeaders() { ForceResolveUtil.forceResolveAllContents(getAnnotations()); ForceResolveUtil.forceResolveAllContents(getDanglingAnnotations()); getCompanionObjectDescriptor(); getDescriptorsForExtraCompanionObjects(); getConstructors(); getContainingDeclaration();
getThisAsReceiverParameter(); getKind(); getModality(); getName(); getOriginal(); getScopeForClassHeaderResolution(); getScopeForMemberDeclarationResolution(); DescriptorUtils.getAllDescriptors(getUnsubstitutedMemberScope()); getScopeForInitializerResolution(); getUnsubstitutedInnerClassesScope(); getTypeConstructor().getSupertypes(); for (TypeParameterDescriptor typeParameterDescriptor : getTypeConstructor().getParameters()) { typeParameterDescriptor.getUpperBounds(); } getUnsubstitutedPrimaryConstructor(); getVisibility(); getContextReceivers(); } -
New type inference Kotlin 1.4 New JVM backend Kotlin 1.5
New JS backend Kotlin 1.6 K2 🙂 Not the fi rst But a big one -
BindingContext public interface BindingContext { <K, V> V get(ReadOnlySlice<K, V>
slice, K key); } Maps are slow ☕ -
BindingContext public interface BindingContext { <K, V> V get(ReadOnlySlice<K, V>
slice, K key); } Maps are slow Maps are huge ☕ ☕ -
BindingContext public interface BindingContext { <K, V> V get(ReadOnlySlice<K, V>
slice, K key); } Maps are slow Maps are huge Bad locality ☕ ☕ ☕ -
PSI public interface PsiElement extends UserDataHolder, Iconable { Project getProject();
Language getLanguage(); PsiElement[] getChildren(); PsiElement getParent(); PsiFile getContainingFile(); PsiElement findElementAt(int offset); TextRange getTextRange(); String getText(); void accept(PsiElementVisitor visitor); void acceptChildren(PsiElementVisitor visitor); PsiElement copy(); PsiElement add(PsiElement element); PsiElement addBefore(PsiElement element, PsiElement anchor); PsiElement addAfter(PsiElement element, PsiElement anchor); void delete(); PsiElement replace(PsiElement newElement); } -
PSI public interface PsiElement extends UserDataHolder, Iconable { /** *
The empty array of PSI elements which can be reused to avoid unnecessary allocations. */ PsiElement[] EMPTY_ARRAY = new PsiElement[0]; ArrayFactory<PsiElement> ARRAY_FACTORY = count -> count == 0 ? EMPTY_ARRAY : new PsiElement[count]; /** * Returns the project to which the PSI element belongs. * * @return the project instance. * @throws PsiInvalidElementAccessException if this element is invalid */ @NotNull @Contract(pure=true) Project getProject() throws PsiInvalidElementAccessException; /** * Returns the language of the PSI element. * * @return the language instance. */ @NotNull @Contract(pure=true) Language getLanguage(); /** * Returns the PSI manager for the project to which the PSI element belongs. * * @return the PSI manager instance. */ @Contract(pure=true) PsiManager getManager(); /** * Returns the array of children for the PSI element. * Important: In some implementations children are only composite elements, i.e. not a leaf elements * * @return the array of child elements. */ @Contract(pure=true) @NotNull PsiElement @NotNull [] getChildren(); /** * Returns the parent of the PSI element. * * @return the parent of the element, or null if the element has no parent. */ @Contract(pure=true) PsiElement getParent(); /** * Returns the first child of the PSI element. * * @return the first child, or null if the element has no children. */ @Contract(pure=true) PsiElement getFirstChild(); /** * Returns the last child of the PSI element. * * @return the last child, or null if the element has no children. */ @Contract(pure=true) PsiElement getLastChild(); /** * Returns the next sibling of the PSI element. * * @return the next sibling, or null if the node is the last in the list of siblings. */ @Contract(pure=true) PsiElement getNextSibling(); /** * Returns the previous sibling of the PSI element. * * @return the previous sibling, or null if the node is the first in the list of siblings. */ @Contract(pure=true) PsiElement getPrevSibling(); /** * Returns the file containing the PSI element. * <p></p> * Note: this method might need to traverse the whole AST up, which can be slow in deep trees, so invoking this method should be avoided if possible. * * @return the file instance, or null if the PSI element is not contained in a file (for example, * the element represents a package or directory). * @throws PsiInvalidElementAccessException * if this element is invalid */ @Contract(pure=true) PsiFile getContainingFile() throws PsiInvalidElementAccessException; /** * Returns the text range in the document occupied by the PSI element. * <p></p> * Note: it works in <i>O(tree_depth)</i> time, which can be slow in deep trees, so invoking this method should be avoided if possible. * * @return the text range. */ @Contract(pure=true) TextRange getTextRange(); /** * @return text range of this element relative to its parent */ @Contract(pure = true) @NotNull default TextRange getTextRangeInParent() { return TextRange.from(getStartOffsetInParent(), getTextLength()); } /** * Returns the text offset of the PSI element relative to its parent. * * @return the relative offset. */ @Contract(pure=true) int getStartOffsetInParent(); /** * Returns the length of text of the PSI element. * * @return the text length. */ @Contract(pure=true) int getTextLength(); /** * Finds a leaf PSI element at the specified offset from the start of the text range of this node. * * @param offset the relative offset for which the PSI element is requested. * @return the element at the offset, or null if none is found. */ @Nullable @Contract(pure=true) PsiElement findElementAt(int offset); /** * Finds a reference at the specified offset from the start of the text range of this node. * * @param offset the relative offset for which the reference is requested. * @return the reference at the offset, or null if none is found. */ @Nullable @Contract(pure=true) PsiReference findReferenceAt(int offset); /** * Returns the offset in the file to which the caret should be placed * when performing the navigation to the element. (For classes implementing * {@link PsiNamedElement}, this should return the offset in the file of the * name identifier.) * * @return the offset of the PSI element. */ @Contract(pure=true) int getTextOffset(); /** * Returns the text of the PSI element. * <p></p> * Note: This call requires traversing whole subtree, so it can be expensive for composite elements, and should be avoided if possible. * * @return the element text. * @see #textMatches * @see #textContains */ @Contract(pure=true) @NlsSafe String getText(); /** * Returns the text of the PSI element as a character array. * * @return the element text as a character array. */ @Contract(pure=true) char @NotNull [] textToCharArray(); /** * Returns the PSI element which should be used as a navigation target * when navigation to this PSI element is requested. The method can either * return {@code this} or substitute a different element if this element * does not have an associated file and offset. (For example, if the source code * of a library is attached to a project, the navigation element for a compiled * library class is its source class.) * * @return the navigation target element. */ @Contract(pure=true) PsiElement getNavigationElement(); /** * Returns the PSI element which corresponds to this element and belongs to * either the project source path or class path. The method can either return * {@code this} or substitute a different element if this element does * not belong to the source path or class path. (For example, the original * element for a library source file is the corresponding compiled class file.) * * @return the original element. */ @Contract(pure=true) PsiElement getOriginalElement(); //Q: get rid of these methods? /** * Checks if the text of this PSI element is equal to the specified character sequence. * * @param text the character sequence to compare with. * @return true if the text is equal, false otherwise. */ @Contract(pure=true) boolean textMatches(@NotNull @NonNls CharSequence text); /** * Checks if the text of this PSI element is equal to the text of the specified PSI element. * * @param element the element to compare the text with. * @return true if the text is equal, false otherwise. */ @Contract(pure=true) boolean textMatches(@NotNull PsiElement element); /** * Checks if the text of this element contains the specified character. * * @param c the character to search for. * @return true if the character is found, false otherwise. */ @Contract(pure=true) boolean textContains(char c); /** * Passes the element to the specified visitor. * * @param visitor the visitor to pass the element to. */ void accept(@NotNull PsiElementVisitor visitor); /** * Passes the children of the element to the specified visitor. * * @param visitor the visitor to pass the children to. */ void acceptChildren(@NotNull PsiElementVisitor visitor); /** * Creates a copy of the file containing the PSI element and returns the corresponding * element in the created copy. Resolve operations performed on elements in the copy * of the file will resolve to elements in the copy, not in the original file. * <p> * For light elements, may return {@code null}. * * @return the element in the file copy corresponding to this element. */ PsiElement copy(); /** * Adds a child to this PSI element. * * @param element the child element to add. * @return the element which was actually added (either {@code element} or its copy). * @throws IncorrectOperationException if the modification is not supported or not possible for some reason. */ PsiElement add(@NotNull PsiElement element) throws IncorrectOperationException; /** * Adds a child to this PSI element, before the specified anchor element. * * @param element the child element to add. * @param anchor the anchor before which the child element is inserted (must be a child of this PSI element) * @return the element which was actually added (either {@code element} or its copy). * @throws IncorrectOperationException if the modification is not supported or not possible for some reason. */ PsiElement addBefore(@NotNull PsiElement element, @Nullable PsiElement anchor) throws IncorrectOperationException; /** * Adds a child to this PSI element, after the specified anchor element. * * @param element the child element to add. * @param anchor the anchor after which the child element is inserted (must be a child of this PSI element) * @return the element which was actually added (either {@code element} or its copy). * @throws IncorrectOperationException if the modification is not supported or not possible for some reason. */ PsiElement addAfter(@NotNull PsiElement element, @Nullable PsiElement anchor) throws IncorrectOperationException; /** * Checks if it is possible to add the specified element as a child to this element, * and throws an exception if the add is not possible. Does not actually modify anything. * * @param element the child element to check the addition possibility. * @throws IncorrectOperationException if the modification is not supported or not possible for some reason. * @deprecated not all PSI implementations implement this method correctly. */ @Deprecated void checkAdd(@NotNull PsiElement element) throws IncorrectOperationException; /** * Adds a range of elements as children to this PSI element. * * @param first the first child element to add. * @param last the last child element to add (must have the same parent as {@code first}) * @return the first child element which was actually added (either {@code first} or its copy). * @throws IncorrectOperationException if the modification is not supported or not possible for some reason. */ PsiElement addRange(PsiElement first, PsiElement last) throws IncorrectOperationException; /** * Adds a range of elements as children to this PSI element, before the specified anchor element. * * @param first the first child element to add. * @param last the last child element to add (must have the same parent as {@code first}) * @param anchor the anchor before which the child element is inserted (must be a child of this PSI element) * @return the first child element which was actually added (either {@code first} or its copy). * @throws IncorrectOperationException if the modification is not supported or not possible for some reason. */ PsiElement addRangeBefore(@NotNull PsiElement first, @NotNull PsiElement last, PsiElement anchor) throws IncorrectOperationException; /** * Adds a range of elements as children to this PSI element, after the specified anchor element. * * @param first the first child element to add. * @param last the last child element to add (must have the same parent as {@code first}) * @param anchor the anchor after which the child element is inserted (must be a child of this PSI element) * @return the first child element which was actually added (either {@code first} or its copy). * @throws IncorrectOperationException if the modification is not supported or not possible for some reason. */ PsiElement addRangeAfter(PsiElement first, PsiElement last, PsiElement anchor) throws IncorrectOperationException; /** * Deletes this PSI element from the tree. * * @throws IncorrectOperationException if the modification is not supported * or not possible for some reason (for example, the file containing the element is read-only). */ void delete() throws IncorrectOperationException; /** * Checks if it is possible to delete the specified element from the tree, * and throws an exception if the addition is not possible. Does not actually modify anything. * * @throws IncorrectOperationException if the modification is not supported or not possible for some reason. * @deprecated not all PSI implementations implement this method correctly. */ @Deprecated void checkDelete() throws IncorrectOperationException; /** * Deletes a range of children of this PSI element from the tree. * * @param first the first child to delete (must be a child of this PSI element) * @param last the last child to delete (must be a child of this PSI element) * @throws IncorrectOperationException if the modification is not supported or not possible for some reason. */ void deleteChildRange(PsiElement first, PsiElement last) throws IncorrectOperationException; /** * Replaces this PSI element (along with all its children) with another element * (along with the children). * * @param newElement the element to replace this element with. * @return the element which was actually inserted in the tree (either {@code newElement} or its copy) * @throws IncorrectOperationException if the modification is not supported or not possible for some reason. */ PsiElement replace(@NotNull PsiElement newElement) throws IncorrectOperationException; /** * Checks if this PSI element is valid. Valid elements and their hierarchy members * can be accessed for reading and writing. Valid elements can still correspond to * underlying documents whose text is different, when those documents have been changed * and not yet committed ({@link PsiDocumentManager#commitDocument(com.intellij.openapi.editor.Document)}). * (In this case an attempt to change PSI will result in an exception).<br><br> * * Most method calls on invalid PSI result in {@link PsiInvalidElementAccessException}. * Once invalid, elements can't become valid again. * Elements become invalid in following cases: * <ul> * <li>They have been deleted via PSI operation (e.g. {@link #delete()})</li> * <li>They have been deleted as a result of an incremental reparse (document commit)</li> * <li>Their containing file has been changed externally, or renamed so that its PSI had to be rebuilt from scratch</li> * </ul> * * Note that calls to this method are expected to be rare and can even be considered a code smell. In general, * when you're given some PSI, you should assume it's valid. If it turns out to be invalid, it's the responsibility * of those who gave you this PSI, not yours, and they should be fixed, not your code.<br><br> * * The rare circumstances where {@code isValid} check makes sense * are those where it's obvious from the surrounding code why the PSI could become invalid. For example, right after a PSI modification * or at the start of a read action (because any write action could've invalidated the PSI between read actions, * and you should never expect PSI to survive that). And even in these circumstances, please consider alternatives * that support PSI restoration, e.g. {@link SmartPsiElementPointer}s. * * @return true if the element is valid, false otherwise. * @see com.intellij.psi.util.PsiUtilCore#ensureValid(PsiElement) */ @Contract(pure=true) boolean isValid(); /** * Checks if the contents of the element can be modified (if it belongs to a * non-read-only source file.) * * @return true if the element can be modified, false otherwise. */ @Contract(pure=true) boolean isWritable(); /** * The contents of the returned collection are copied after the method returns, * the platform doesn't store or modify the returned collection. * * @return collection of declarations in this element, or empty collection if there are no such declarations * @see com.intellij.model.psi.PsiSymbolDeclarationProvider */ @Experimental default @NotNull Collection<? extends @NotNull PsiSymbolDeclaration> getOwnDeclarations() { return Collections.emptyList(); } /** * The returned references are expected to be used by language support, * for example in Java `foo` element in `foo = 42` expression has a reference, * which is used by Java language support to compute expected type of the assignment. * <p> * On the other hand {@code "bar"} literal in {@code new File("bar")} is a string literal, * and from Java language perspective it has no references, * but the framework support "knows" that this literal contains the reference to a file. * These are external references. * <p/> * The contents of the returned collection are copied after the method returns, * the platform doesn't store or modify the returned collection. * * @return collection of references from this element, or empty collection if there are no such references * @see com.intellij.model.psi.PsiExternalReferenceHost * @see com.intellij.model.psi.PsiSymbolReferenceService#getReferences(PsiElement) */ @Experimental default @NotNull Collection<? extends @NotNull PsiSymbolReference> getOwnReferences() { return Collections.emptyList(); } /** * Returns the reference from this PSI element to another PSI element (or elements), if one exists. * If the element has multiple associated references (see {@link #getReferences()} * for an example), returns the first associated reference. * * @return the reference instance, or null if the PSI element does not have any * associated references. * @see com.intellij.psi.search.searches.ReferencesSearch */ @Nullable @Contract(pure=true) PsiReference getReference(); /** * Returns all references from this PSI element to other PSI elements. An element can * have multiple references when, for example, the element is a string literal containing * multiple sub-strings which are valid full-qualified class names. If an element * contains only one text fragment which acts as a reference but the reference has * multiple possible targets, {@link PsiPolyVariantReference} should be used instead * of returning multiple references. * <p/> * Actually, it's preferable to call {@link PsiReferenceService#getReferences} instead * as it allows adding references by plugins when the element implements {@link ContributedReferenceHost}. * * @return the array of references, or an empty array if the element has no associated * references. * @see PsiReferenceService#getReferences * @see com.intellij.psi.search.searches.ReferencesSearch */ @Contract(pure=true) PsiReference @NotNull [] getReferences(); /** * Returns a copyable user data object attached to this element. * * @param key the key for accessing the user data object. * @return the user data object, or null if no such object is found in the current element. * @see #putCopyableUserData(Key, Object) */ @Nullable @Contract(pure = true) <T> T getCopyableUserData(@NotNull Key<T> key); /** * Attaches a copyable user data object to this element. Copyable user data objects are copied * when the PSI elements are copied. * * @param key the key for accessing the user data object. * @param value the user data object to attach. * @see #getCopyableUserData(Key) */ <T> void putCopyableUserData(@NotNull Key<T> key, @Nullable T value); /** * Passes the declarations contained in this PSI element and its children * for processing to the specified scope processor. * * @param processor the processor receiving the declarations. * @param lastParent the child of this element has been processed during the previous * step of the tree up walk (declarations under this element do not need * to be processed again) * @param place the original element from which the tree walk-up was initiated. * @return true if the declaration processing should continue or false if it should be stopped. */ boolean processDeclarations(@NotNull PsiScopeProcessor processor, @NotNull ResolveState state, @Nullable PsiElement lastParent, @NotNull PsiElement place); /** * Returns the element which should be used as the parent of this element in a tree up * walk during a resolve operation. For most elements, this returns {@code getParent()}, * but the context can be overridden for some elements like code fragments (see * {@link JavaCodeFragmentFactory#createCodeBlockCodeFragment(String, PsiElement, boolean)}). * * @return the resolve context element. */ @Nullable @Contract(pure=true) PsiElement getContext(); /** * Checks if an actual source or class file corresponds to the element. Non-physical elements include, * for example, PSI elements created for the watch expressions in the debugger. * Non-physical elements do not generate tree change events. * Also, {@link PsiDocumentManager#getDocument(PsiFile)} returns null for non-physical elements. * Not to be confused with {@link FileViewProvider#isPhysical()}. * * @return true if the element is physical, false otherwise. */ @Contract(pure=true) boolean isPhysical(); /** * Returns the scope in which the declarations for the references in this PSI element are searched. * * @return the resolve scope instance. */ @NotNull @Contract(pure=true) GlobalSearchScope getResolveScope(); /** * Returns the scope in which references to this element are searched. * * @return the search scope instance. * @see com.intellij.psi.search.PsiSearchHelper#getUseScope(PsiElement) */ @NotNull @Contract(pure=true) SearchScope getUseScope(); /** * Returns the AST node corresponding to the element. * * @return the AST node instance. */ @Contract(pure=true) ASTNode getNode(); /** * toString() should never be presented to the user. */ @Override @NonNls @Contract(pure=true) String toString(); /** * This method shouldn't be called by clients directly, because there are no guarantees of it being symmetric. * It's called by {@link PsiManager#areElementsEquivalent(PsiElement, PsiElement)} internally, which clients should invoke instead.<p/> * * Implementations of this method should return {@code true} if the parameter is resolve-equivalent to {@code this}, i.e. it represents * the same entity from the language perspective. See also {@link PsiManager#areElementsEquivalent(PsiElement, PsiElement)} documentation. */ @Contract(pure=true) boolean isEquivalentTo(PsiElement another); -
StorageManager interface StorageManager { fun <K, V : Any> createMemoizedFunction(compute:
(K) -> V): MemoizedFunctionToNotNull<K, V> fun <K, V : Any> createMemoizedFunctionWithNullableValues(compute: (K) -> V?): MemoizedFunctionToNullable<K, V> fun <T : Any> createLazyValue(computable: () -> T): NotNullLazyValue<T> fun <T : Any> createNullableLazyValue(computable: () -> T?): NullableLazyValue<T> } Additional cost pro fi le Complex to 🤖 -
Frontend Kotlin Compiler Kotlin IDE plugin Love and
peace DeclarationDescriptor KotlinType BindingContext ResolvedCall DataFlowInfo LexicalScope -
Frontend Kotlin Compiler Kotlin IDE plugin Love and
peace, or not DeclarationDescriptor KotlinType BindingContext ResolvedCall DataFlowInfo LexicalScope Other IDE plugins 💧 💧 💢 🎃 fixme -
Use IR? Intermediate Representation Without IR Lexer Parser Semantic Analyzer
Tokens AST Backend AST Semantics Sources Lexer Parser Semantic Analyzer Tokens Backend IR Sources IR Converter AST IR With IR -
K2 Work fl ow Lexer & Parser .kt Backend FIR
Binary Semantic Analyzer PSI→FIR Converter PSI FIR -
K2 Work fl ow Lexer & Parser AST→FIR Converter
Lighter AST Backend FIR Binary Semantic Analyzer FIR PSI→FIR Converter PSI FIR In compiler In IDE .kt -
if → when if (flag) { doThis() } else {
doThat() } when { flag -> doThis() else -> doThat() } -
for → while for (item in list) { print(item) }
val iterator = list.iterator() while (iterator.hasNext()) { val item = iterator.next() print(item) } -
val (name, age) = getUser() val destruct = getUser() val
name = destruct.component1() val age = destruct.component2() Destructuring declaration -
Raw FIR Imports Semantic Analyzer in K2 Supertypes Types Body
Resolve Status FIR FIR FIR FIR FIR FIR -
FirResolvePhase enum class FirResolvePhase { RAW_FIR, IMPORTS, COMPILER_REQUIRED_ANNOTATIONS, COMPANION_GENERATION, SUPER_TYPES,
SEALED_CLASS_INHERITORS, TYPES, STATUS, ARGUMENTS_OF_ANNOTATIONS, CONTRACTS, IMPLICIT_TYPES_BODY_RESOLVE, ANNOTATIONS_ARGUMENTS_MAPPING, EXPECT_ACTUAL_MATCHING, BODY_RESOLVE } (slightly simpli fi ed) -
RAW_FIR public? final? [RAW_FIR] fun factorial([RAW_FIR] n: Int): Int {
LAZY_BLOCK } fun factorial(n: Int): Int { if (n == 0) return 1 return (1..n).reduce { acc, i -> acc * i } } -
TYPES public? final? [TYPES] fun factorial([TYPES] n: R|kotlin/Int|): R|kotlin/Int| {
LAZY_BLOCK } fun factorial(n: Int): Int { if (n == 0) return 1 return (1..n).reduce { acc, i -> acc * i } } -
STATUS public final [STATUS] fun factorial([STATUS] n: R|kotlin/Int|): R|kotlin/Int| {
LAZY_BLOCK } fun factorial(n: Int): Int { if (n == 0) return 1 return (1..n).reduce { acc, i -> acc * i } } -
CONTRACTS public final [CONTRACTS] fun factorial([CONTRACTS] n: R|kotlin/Int|): R|kotlin/Int| {
when () { ==(n#, IntegerLiteral(0)) -> { ^factorial IntegerLiteral(1) } } ^factorial IntegerLiteral(1).rangeTo#(n#).reduce#( <L> = [STATUS] reduce@fun <implicit>.<anonymous>( [RAW_FIR] acc: <implicit>, [RAW_FIR] i: <implicit> ): <implicit> <inline=Unknown> { acc#.times#(i#) } ) } fun factorial(n: Int): Int { if (n == 0) return 1 return (1..n).reduce { acc, i -> acc * i } } -
BODY_RESOLVE public final [BODY_RESOLVE] fun factorial([BODY_RESOLVE] n: R|kotlin/Int|): R|kotlin/Int| {
when () { ==(R|<local>/n|, Int(0)) -> { ^factorial Int(1) } } ^factorial Int(1).R|kotlin/Int.rangeTo|(R|<local>/n|).R|kotlin/collections/reduce|<R|kotlin/Int|, R|kotlin/Int|>( <L> = [BODY_RESOLVE] [MatchingParameterFunctionTypeKey=kotlin/Function2<S, T, S>] reduce@fun <anonymous>( [BODY_RESOLVE] acc: R|kotlin/Int|, [BODY_RESOLVE] i: R|kotlin/Int| ): R|kotlin/Int| <inline=Inline, kind=UNKNOWN> { ^ R|<local>/acc|.R|kotlin/Int.times|(R|<local>/i|) } ) } fun factorial(n: Int): Int { if (n == 0) return 1 return (1..n).reduce { acc, i -> acc * i } } -
BODY_RESOLVE public final [BODY_RESOLVE] fun factorial([BODY_RESOLVE] n: R|kotlin/Int|): R|kotlin/Int| {
when () { ==(R|<local>/n|, Int(0)) -> { ^factorial Int(1) } } ^factorial Int(1).R|kotlin/Int.rangeTo|(R|<local>/n|).R|kotlin/collections/reduce|<R|kotlin/Int|, R|kotlin/Int|>( <L> = [BODY_RESOLVE] [MatchingParameterFunctionTypeKey=kotlin/Function2<S, T, S>] reduce@fun <anonymous>( [BODY_RESOLVE] acc: R|kotlin/Int|, [BODY_RESOLVE] i: R|kotlin/Int| ): R|kotlin/Int| <inline=Inline, kind=UNKNOWN> { ^ R|<local>/acc|.R|kotlin/Int.times|(R|<local>/i|) } ) } fun factorial(n: Int): Int { if (n == 0) return 1 return (1..n).reduce { acc, i -> acc * i } } Only function body is mutated Signature is the same -
override fun isOverriddenFunction(overrideCandidate: FirSimpleFunction, baseDeclaration: FirSimpleFunction): Boolean { if (overrideCandidate.isStatic
!= baseDeclaration.isStatic) return false overrideCandidate.lazyResolveToPhase(FirResolvePhase.TYPES) baseDeclaration.lazyResolveToPhase(FirResolvePhase.TYPES) // NB: overrideCandidate is from Java and has no receiver val receiverTypeRef = baseDeclaration.receiverParameter?.typeRef val baseParameterTypes = listOfNotNull(receiverTypeRef) + baseDeclaration.valueParameters.map { it.returnTypeRef } if (overrideCandidate.valueParameters.size != baseParameterTypes.size) return false val substitutor = buildTypeParametersSubstitutorIfCompatible(overrideCandidate, baseDeclaration) if (!overrideCandidate.valueParameters.zip(baseParameterTypes).all { (paramFromJava, baseType) -> isEqualTypes(paramFromJava.returnTypeRef, baseType, substitutor) }) { return false } if (considerReturnTypeKinds && !doesReturnTypesHaveSameKind(overrideCandidate, baseDeclaration)) return false return true } lazyResolveToPhase() -
K1 Frontend K1 Compiler K1 IDE plugin DeclarationDescriptor
KotlinType BindingContext ResolvedCall DataFlowInfo LexicalScope K2 Frontend FirTransformer FirSymbolProvider FirElement FirDeclarationChecker FirTowerResolver K2 Compiler 😿 -
K1 Frontend K2 Frontend Analysis API K1 Compiler K2
Compiler K1 IDE plugin K2 IDE plugin