Documentation ¶
Index ¶
- Variables
- func FollowAllSymbols(symbols SymbolMap)
- func GenerateNonUniqueNameFromPath(path string) string
- func IsBooleanValue(a Expr) bool
- func IsSuperCall(stmt Stmt) bool
- type AST
- type Arg
- type ArrayBinding
- type AssignTarget
- type B
- type BArray
- type BIdentifier
- type BMissing
- type BObject
- type Binding
- type Case
- type Catch
- type CharFreq
- type Class
- type ClauseItem
- type Comment
- type Decl
- type DeclaredSymbol
- type E
- type EArray
- type EArrow
- type EAwait
- type EBigInt
- type EBinary
- type EBoolean
- type ECall
- type EClass
- type EDot
- type EFunction
- type EIdentifier
- type EIf
- type EImport
- type EImportIdentifier
- type EImportMeta
- type EIndex
- type EJSXElement
- type EMissing
- type ENew
- type ENewTarget
- type ENull
- type ENumber
- type EObject
- type EPrivateIdentifier
- type ERegExp
- type ERequire
- type ERequireResolve
- type ESpread
- type EString
- type ESuper
- type ETemplate
- type EThis
- type EUnary
- type EUndefined
- type EYield
- type EnumValue
- type ExportStarAlias
- type Expr
- type ExprOrStmt
- type Finally
- type Fn
- type FnBody
- type ImportItemStatus
- type L
- type LocRef
- type LocalKind
- type NameMinifier
- type NamedImport
- type NamespaceAlias
- type OpCode
- type OptionalChain
- type Part
- type Property
- type PropertyBinding
- type PropertyKind
- type Ref
- type S
- type SBlock
- type SBreak
- type SClass
- type SComment
- type SContinue
- type SDebugger
- type SDirective
- type SDoWhile
- type SEmpty
- type SEnum
- type SExportClause
- type SExportDefault
- type SExportEquals
- type SExportFrom
- type SExportStar
- type SExpr
- type SFor
- type SForIn
- type SForOf
- type SFunction
- type SIf
- type SImport
- type SLabel
- type SLazyExport
- type SLocal
- type SNamespace
- type SReturn
- type SSwitch
- type SThrow
- type STry
- type STypeScript
- type SWhile
- type SWith
- type Scope
- type ScopeKind
- type ScopeMember
- type SlotCounts
- type SlotNamespace
- type Span
- type Stmt
- type Symbol
- type SymbolKind
- type SymbolMap
- type SymbolUse
- type TemplatePart
Constants ¶
This section is empty.
Variables ¶
var DefaultNameMinifier = NameMinifier{ // contains filtered or unexported fields }
var OpTable = []opTableEntry{ {"+", LPrefix, false}, {"-", LPrefix, false}, {"~", LPrefix, false}, {"!", LPrefix, false}, {"void", LPrefix, true}, {"typeof", LPrefix, true}, {"delete", LPrefix, true}, {"--", LPrefix, false}, {"++", LPrefix, false}, {"--", LPostfix, false}, {"++", LPostfix, false}, {"+", LAdd, false}, {"-", LAdd, false}, {"*", LMultiply, false}, {"/", LMultiply, false}, {"%", LMultiply, false}, {"**", LExponentiation, false}, {"<", LCompare, false}, {"<=", LCompare, false}, {">", LCompare, false}, {">=", LCompare, false}, {"in", LCompare, true}, {"instanceof", LCompare, true}, {"<<", LShift, false}, {">>", LShift, false}, {">>>", LShift, false}, {"==", LEquals, false}, {"!=", LEquals, false}, {"===", LEquals, false}, {"!==", LEquals, false}, {"??", LNullishCoalescing, false}, {"||", LLogicalOr, false}, {"&&", LLogicalAnd, false}, {"|", LBitwiseOr, false}, {"&", LBitwiseAnd, false}, {"^", LBitwiseXor, false}, {",", LComma, false}, {"=", LAssign, false}, {"+=", LAssign, false}, {"-=", LAssign, false}, {"*=", LAssign, false}, {"/=", LAssign, false}, {"%=", LAssign, false}, {"**=", LAssign, false}, {"<<=", LAssign, false}, {">>=", LAssign, false}, {">>>=", LAssign, false}, {"|=", LAssign, false}, {"&=", LAssign, false}, {"^=", LAssign, false}, {"??=", LAssign, false}, {"||=", LAssign, false}, {"&&=", LAssign, false}, }
Functions ¶
func FollowAllSymbols ¶
func FollowAllSymbols(symbols SymbolMap)
Use this before calling "FollowSymbols" from separate threads to avoid concurrent map update hazards. In Go, mutating a map is not threadsafe but reading from a map is. Calling "FollowAllSymbols" first ensures that all mutation is done up front.
func GenerateNonUniqueNameFromPath ¶
For readability, the names of certain automatically-generated symbols are derived from the file name. For example, instead of the CommonJS wrapper for a file being called something like "require273" it can be called something like "require_react" instead. This function generates the part of these identifiers that's specific to the file path. It can take both an absolute path (OS-specific) and a path in the source code (OS-independent).
Note that these generated names do not at all relate to the correctness of the code as far as avoiding symbol name collisions. These names still go through the renaming logic that all other symbols go through to avoid name collisions.
func IsBooleanValue ¶
func IsSuperCall ¶
Types ¶
type AST ¶
type AST struct { ApproximateLineCount int32 NestedScopeSlotCounts SlotCounts HasLazyExport bool // This is a list of CommonJS features. When a file uses CommonJS features, // it's not a candidate for "flat bundling" and must be wrapped in its own // closure. HasTopLevelReturn bool UsesExportsRef bool UsesModuleRef bool // This is a list of ES6 features HasES6Imports bool HasES6Exports bool Hashbang string Directive string URLForCSS string Parts []Part Symbols []Symbol ModuleScope *Scope CharFreq *CharFreq ExportsRef Ref ModuleRef Ref WrapperRef Ref // These are stored at the AST level instead of on individual AST nodes so // they can be manipulated efficiently without a full AST traversal ImportRecords []ast.ImportRecord // These are used when bundling. They are filled in during the parser pass // since we already have to traverse the AST then anyway and the parser pass // is conveniently fully parallelized. NamedImports map[Ref]NamedImport NamedExports map[string]Ref TopLevelSymbolToParts map[Ref][]uint32 ExportStarImportRecords []uint32 SourceMapComment Span }
func (*AST) HasCommonJSFeatures ¶
func (*AST) HasES6Syntax ¶
func (*AST) UsesCommonJSExports ¶
type ArrayBinding ¶
type AssignTarget ¶
type AssignTarget uint8
const ( AssignTargetNone AssignTarget = iota AssignTargetReplace // "a = b" AssignTargetUpdate // "a += b" )
type B ¶
type B interface {
// contains filtered or unexported methods
}
This interface is never called. Its purpose is to encode a variant type in Go's type system.
type BArray ¶
type BArray struct { Items []ArrayBinding HasSpread bool IsSingleLine bool }
type BIdentifier ¶
type BIdentifier struct{ Ref Ref }
type BObject ¶
type BObject struct { Properties []PropertyBinding IsSingleLine bool }
type CharFreq ¶
type CharFreq [64]int32
This is a histogram of character frequencies for minification
func (*CharFreq) Compile ¶
func (freq *CharFreq) Compile() NameMinifier
type ClauseItem ¶
type DeclaredSymbol ¶
type E ¶
type E interface {
// contains filtered or unexported methods
}
This interface is never called. Its purpose is to encode a variant type in Go's type system.
type ECall ¶
type ECall struct { Target Expr Args []Expr OptionalChain OptionalChain IsDirectEval bool // True if there is a comment containing "@__PURE__" or "#__PURE__" preceding // this call expression. This is an annotation used for tree shaking, and // means that the call can be removed if it's unused. It does not mean the // call is pure (e.g. it may still return something different if called twice). // // Note that the arguments are not considered to be part of the call. If the // call itself is removed due to this annotation, the arguments must remain // if they have side effects. CanBeUnwrappedIfUnused bool }
func (*ECall) HasSameFlagsAs ¶
type EDot ¶
type EDot struct { Target Expr Name string NameLoc logger.Loc OptionalChain OptionalChain // If true, this property access is known to be free of side-effects. That // means it can be removed if the resulting value isn't used. CanBeRemovedIfUnused bool // If true, this property access is a function that, when called, can be // unwrapped if the resulting value is unused. Unwrapping means discarding // the call target but keeping any arguments with side effects. CallCanBeUnwrappedIfUnused bool }
func (*EDot) HasSameFlagsAs ¶
type EIdentifier ¶
type EIdentifier struct { Ref Ref // If true, this identifier is known to not have a side effect (i.e. to not // throw an exception) when referenced. If false, this identifier may or may // not have side effects when referenced. This is used to allow the removal // of known globals such as "Object" if they aren't used. CanBeRemovedIfUnused bool // If true, this identifier represents a function that, when called, can be // unwrapped if the resulting value is unused. Unwrapping means discarding // the call target but keeping any arguments with side effects. CallCanBeUnwrappedIfUnused bool }
type EImport ¶
type EImport struct { Expr Expr ImportRecordIndex *uint32 // Comments inside "import()" expressions have special meaning for Webpack. // Preserving comments inside these expressions makes it possible to use // esbuild as a TypeScript-to-JavaScript frontend for Webpack to improve // performance. We intentionally do not interpret these comments in esbuild // because esbuild is not Webpack. But we do preserve them since doing so is // harmless, easy to maintain, and useful to people. See the Webpack docs for // more info: https://webpack.js.org/api/module-methods/#magic-comments. LeadingInteriorComments []Comment }
type EImportIdentifier ¶
type EImportIdentifier struct {
Ref Ref
}
This is similar to an EIdentifier but it represents a reference to an ES6 import item.
Depending on how the code is linked, the file containing this EImportIdentifier may or may not be in the same module group as the file it was imported from.
If it's the same module group than we can just merge the import item symbol with the corresponding symbol that was imported, effectively renaming them to be the same thing and statically binding them together.
But if it's a different module group, then the import must be dynamically evaluated using a property access off the corresponding namespace symbol, which represents the result of a require() call.
It's stored as a separate type so it's not easy to confuse with a plain identifier. For example, it'd be bad if code trying to convert "{x: x}" into "{x}" shorthand syntax wasn't aware that the "x" in this case is actually "{x: importedNamespace.x}". This separate type forces code to opt-in to doing this instead of opt-out.
type EImportMeta ¶
type EImportMeta struct{}
type EIndex ¶
type EIndex struct { Target Expr Index Expr OptionalChain OptionalChain }
func (*EIndex) HasSameFlagsAs ¶
type EJSXElement ¶
type ENewTarget ¶
type ENewTarget struct{}
type EPrivateIdentifier ¶
type EPrivateIdentifier struct {
Ref Ref
}
This is similar to EIdentifier but it represents class-private fields and methods. It can be used where computed properties can be used, such as EIndex and Property.
type ERequireResolve ¶ added in v0.7.6
type ERequireResolve struct {
ImportRecordIndex uint32
}
type ETemplate ¶
type ETemplate struct { Tag *Expr Head []uint16 HeadRaw string // This is only filled out for tagged template literals Parts []TemplatePart }
type EUndefined ¶
type EUndefined struct{}
type ExportStarAlias ¶
type ExprOrStmt ¶
type ImportItemStatus ¶
type ImportItemStatus uint8
const ( ImportItemNone ImportItemStatus = iota // The linker doesn't report import/export mismatch errors ImportItemGenerated // The printer will replace this import with "undefined" ImportItemMissing )
type L ¶
type L int
const ( LLowest L = iota LComma LSpread LYield LAssign LConditional LNullishCoalescing LLogicalOr LLogicalAnd LBitwiseOr LBitwiseXor LBitwiseAnd LEquals LCompare LShift LAdd LMultiply LExponentiation LPrefix LPostfix LNew LCall LMember )
https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Operators/Operator_Precedence
type NameMinifier ¶
type NameMinifier struct {
// contains filtered or unexported fields
}
func (*NameMinifier) NumberToMinifiedName ¶
func (minifier *NameMinifier) NumberToMinifiedName(i int) string
type NamedImport ¶
type NamedImport struct { // Parts within this file that use this import LocalPartsWithUses []uint32 Alias string AliasLoc logger.Loc NamespaceRef Ref ImportRecordIndex uint32 // It's useful to flag exported imports because if they are in a TypeScript // file, we can't tell if they are a type or a value. IsExported bool }
type NamespaceAlias ¶
type OpCode ¶
type OpCode int
const ( // Prefix UnOpPos OpCode = iota UnOpNeg UnOpCpl UnOpNot UnOpVoid UnOpTypeof UnOpDelete // Prefix update UnOpPreDec UnOpPreInc // Postfix update UnOpPostDec UnOpPostInc // Left-associative BinOpAdd BinOpSub BinOpMul BinOpDiv BinOpRem BinOpPow BinOpLt BinOpLe BinOpGt BinOpGe BinOpIn BinOpInstanceof BinOpShl BinOpShr BinOpUShr BinOpLooseEq BinOpLooseNe BinOpStrictEq BinOpStrictNe BinOpNullishCoalescing BinOpLogicalOr BinOpLogicalAnd BinOpBitwiseOr BinOpBitwiseAnd BinOpBitwiseXor // Non-associative BinOpComma // Right-associative BinOpAssign BinOpAddAssign BinOpSubAssign BinOpMulAssign BinOpDivAssign BinOpRemAssign BinOpPowAssign BinOpShlAssign BinOpShrAssign BinOpUShrAssign BinOpBitwiseOrAssign BinOpBitwiseAndAssign BinOpBitwiseXorAssign BinOpNullishCoalescingAssign BinOpLogicalOrAssign BinOpLogicalAndAssign )
If you add a new token, remember to add it to "OpTable" too
func (OpCode) BinaryAssignTarget ¶
func (op OpCode) BinaryAssignTarget() AssignTarget
func (OpCode) IsLeftAssociative ¶
func (OpCode) IsRightAssociative ¶
func (OpCode) UnaryAssignTarget ¶
func (op OpCode) UnaryAssignTarget() AssignTarget
type OptionalChain ¶
type OptionalChain uint8
const ( // "a.b" OptionalChainNone OptionalChain = iota // "a?.b" OptionalChainStart // "a?.b.c" => ".c" is OptionalChainContinue // "(a?.b).c" => ".c" is OptionalChainNone OptionalChainContinue )
type Part ¶
type Part struct { Stmts []Stmt Scopes []*Scope // Each is an index into the file-level import record list ImportRecordIndices []uint32 // All symbols that are declared in this part. Note that a given symbol may // have multiple declarations, and so may end up being declared in multiple // parts (e.g. multiple "var" declarations with the same name). Also note // that this list isn't deduplicated and may contain duplicates. DeclaredSymbols []DeclaredSymbol // An estimate of the number of uses of all symbols used within this part. SymbolUses map[Ref]SymbolUse // The indices of the other parts in this file that are needed if this part // is needed. LocalDependencies map[uint32]bool // If true, this part can be removed if none of the declared symbols are // used. If the file containing this part is imported, then all parts that // don't have this flag enabled must be included. CanBeRemovedIfUnused bool // If true, this is the automatically-generated part for this file's ES6 // exports. It may hold the "const exports = {};" statement and also the // "__export(exports, { ... })" call to initialize the getters. IsNamespaceExport bool // This is used for generated parts that we don't want to be present if they // aren't needed. This enables tree shaking for these parts even if global // tree shaking isn't enabled. ForceTreeShaking bool }
Each file is made up of multiple parts, and each part consists of one or more top-level statements. Parts are used for tree shaking and code splitting analysis. Individual parts of a file can be discarded by tree shaking and can be assigned to separate chunks (i.e. output files) by code splitting.
type Property ¶
type Property struct { TSDecorators []Expr Key Expr // This is omitted for class fields Value *Expr // This is used when parsing a pattern that uses default values: // // [a = 1] = []; // ({a = 1} = {}); // // It's also used for class fields: // // class Foo { a = 1 } // Initializer *Expr Kind PropertyKind IsComputed bool IsMethod bool IsStatic bool WasShorthand bool }
type PropertyBinding ¶
type PropertyKind ¶
type PropertyKind int
const ( PropertyNormal PropertyKind = iota PropertyGet PropertySet PropertySpread )
type Ref ¶
Files are parsed in parallel for speed. We want to allow each parser to generate symbol IDs that won't conflict with each other. We also want to be able to quickly merge symbol tables from all files into one giant symbol table.
We can accomplish both goals by giving each symbol ID two parts: an outer index that is unique to the parser goroutine, and an inner index that increments as the parser generates new symbol IDs. Then a symbol map can be an array of arrays indexed first by outer index, then by inner index. The maps can be merged quickly by creating a single outer array containing all inner arrays from all parsed files.
func FollowSymbols ¶
Returns the canonical ref that represents the ref for the provided symbol. This may not be the provided ref if the symbol has been merged with another symbol.
type S ¶
type S interface {
// contains filtered or unexported methods
}
This interface is never called. Its purpose is to encode a variant type in Go's type system.
type SDirective ¶
type SDirective struct {
Value []uint16
}
type SExportClause ¶
type SExportClause struct { Items []ClauseItem IsSingleLine bool }
type SExportDefault ¶
type SExportDefault struct { DefaultName LocRef Value ExprOrStmt // May be a SFunction or SClass }
type SExportEquals ¶
type SExportEquals struct {
Value Expr
}
This is an "export = value;" statement in TypeScript
type SExportFrom ¶
type SExportFrom struct { Items []ClauseItem NamespaceRef Ref ImportRecordIndex uint32 IsSingleLine bool }
type SExportStar ¶
type SExportStar struct { NamespaceRef Ref Alias *ExportStarAlias ImportRecordIndex uint32 }
type SImport ¶
type SImport struct { // If this is a star import: This is a Ref for the namespace symbol. The Loc // for the symbol is StarLoc. // // Otherwise: This is an auto-generated Ref for the namespace representing // the imported file. In this case StarLoc is nil. The NamespaceRef is used // when converting this module to a CommonJS module. NamespaceRef Ref DefaultName *LocRef Items *[]ClauseItem StarNameLoc *logger.Loc ImportRecordIndex uint32 IsSingleLine bool }
This object represents all of these types of import statements:
import 'path' import {item1, item2} from 'path' import * as ns from 'path' import defaultItem, {item1, item2} from 'path' import defaultItem, * as ns from 'path'
Many parts are optional and can be combined in different ways. The only restriction is that you cannot have both a clause and a star namespace.
type SLazyExport ¶
type SLazyExport struct {
Value Expr
}
The decision of whether to export an expression using "module.exports" or "export default" is deferred until linking using this statement kind
type Scope ¶
type Scope struct { Kind ScopeKind Parent *Scope Children []*Scope Members map[string]ScopeMember Generated []Ref // This is used to store the ref of the label symbol for ScopeLabel scopes. LabelRef Ref LabelStmtIsLoop bool // If a scope contains a direct eval() expression, then none of the symbols // inside that scope can be renamed. We conservatively assume that the // evaluated code might reference anything that it has access to. ContainsDirectEval bool }
type ScopeMember ¶
type SlotCounts ¶
type SlotCounts [3]uint32
func (*SlotCounts) UnionMax ¶
func (a *SlotCounts) UnionMax(b SlotCounts)
type SlotNamespace ¶
type SlotNamespace uint8
const ( SlotDefault SlotNamespace = iota SlotLabel SlotPrivateName SlotMustNotBeRenamed )
type Stmt ¶
func AssignStmt ¶
type Symbol ¶
type Symbol struct { // This is the name that came from the parser. Printed names may be renamed // during minification or to avoid name collisions. Do not use the original // name during printing. OriginalName string // This is used for symbols that represent items in the import clause of an // ES6 import statement. These should always be referenced by EImportIdentifier // instead of an EIdentifier. When this is present, the expression should // be printed as a property access off the namespace instead of as a bare // identifier. // // For correctness, this must be stored on the symbol instead of indirectly // associated with the Ref for the symbol somehow. In ES6 "flat bundling" // mode, re-exported symbols are collapsed using MergeSymbols() and renamed // symbols from other files that end up at this symbol must be able to tell // if it has a namespace alias. NamespaceAlias *NamespaceAlias // Used by the parser for single pass parsing. Symbols that have been merged // form a linked-list where the last link is the symbol to use. This link is // an invalid ref if it's the last link. If this isn't invalid, you need to // FollowSymbols to get the real one. Link Ref // An estimate of the number of uses of this symbol. This is used to detect // whether a symbol is used or not. For example, TypeScript imports that are // unused must be removed because they are probably type-only imports. This // is an estimate and may not be completely accurate due to oversights in the // code. But it should always be non-zero when the symbol is used. UseCountEstimate uint32 // This is for generating cross-chunk imports and exports for code splitting. // It's stored as one's complement so the zero value is invalid. ChunkIndex uint32 // This is used for minification. Symbols that are declared in sibling scopes // can share a name. A good heuristic (from Google Closure Compiler) is to // assign names to symbols from sibling scopes in declaration order. That way // local variable names are reused in each global function like this, which // improves gzip compression: // // function x(a, b) { ... } // function y(a, b, c) { ... } // // The parser fills this in for symbols inside nested scopes. There are three // slot namespaces: regular symbols, label symbols, and private symbols. This // is stored as one's complement so the zero value is invalid. NestedScopeSlot uint32 Kind SymbolKind // Certain symbols must not be renamed or minified. For example, the // "arguments" variable is declared by the runtime for every function. // Renaming can also break any identifier used inside a "with" statement. MustNotBeRenamed bool // We automatically generate import items for property accesses off of // namespace imports. This lets us remove the expensive namespace imports // while bundling in many cases, replacing them with a cheap import item // instead: // // import * as ns from 'path' // ns.foo() // // That can often be replaced by this, which avoids needing the namespace: // // import {foo} from 'path' // foo() // // However, if the import is actually missing then we don't want to report a // compile-time error like we do for real import items. This status lets us // avoid this. We also need to be able to replace such import items with // undefined, which this status is also used for. ImportItemStatus ImportItemStatus }
Note: the order of values in this struct matters to reduce struct size.
func (*Symbol) SlotNamespace ¶
func (s *Symbol) SlotNamespace() SlotNamespace
type SymbolKind ¶
type SymbolKind uint8
const ( // An unbound symbol is one that isn't declared in the file it's referenced // in. For example, using "window" without declaring it will be unbound. SymbolUnbound SymbolKind = iota // This has special merging behavior. You're allowed to re-declare these // symbols more than once in the same scope. These symbols are also hoisted // out of the scope they are declared in to the closest containing function // or module scope. These are the symbols with this kind: // // - Function arguments // - Function statements // - Variables declared using "var" // SymbolHoisted SymbolHoistedFunction // There's a weird special case where catch variables declared using a simple // identifier (i.e. not a binding pattern) block hoisted variables instead of // becoming an error: // // var e = 0; // try { throw 1 } catch (e) { // print(e) // 1 // var e = 2 // print(e) // 2 // } // print(e) // 0 (since the hoisting stops at the catch block boundary) // // However, other forms are still a syntax error: // // try {} catch (e) { let e } // try {} catch ({e}) { var e } // // This symbol is for handling this weird special case. SymbolCatchIdentifier // Generator and async functions are not hoisted, but still have special // properties such as being able to overwrite previous functions with the // same name SymbolGeneratorOrAsyncFunction // This is the special "arguments" variable inside functions SymbolArguments // Classes can merge with TypeScript namespaces. SymbolClass // A class-private identifier (i.e. "#foo"). SymbolPrivateField SymbolPrivateMethod SymbolPrivateGet SymbolPrivateSet SymbolPrivateGetSetPair SymbolPrivateStaticField SymbolPrivateStaticMethod SymbolPrivateStaticGet SymbolPrivateStaticSet SymbolPrivateStaticGetSetPair // Labels are in their own namespace SymbolLabel // TypeScript enums can merge with TypeScript namespaces and other TypeScript // enums. SymbolTSEnum // TypeScript namespaces can merge with classes, functions, TypeScript enums, // and other TypeScript namespaces. SymbolTSNamespace // In TypeScript, imports are allowed to silently collide with symbols within // the module. Presumably this is because the imports may be type-only. SymbolImport // This annotates all other symbols that don't have special behavior. SymbolOther // This symbol causes a compile error when referenced SymbolError )
func (SymbolKind) Feature ¶
func (kind SymbolKind) Feature() compat.JSFeature
func (SymbolKind) IsHoisted ¶
func (kind SymbolKind) IsHoisted() bool
func (SymbolKind) IsHoistedOrFunction ¶
func (kind SymbolKind) IsHoistedOrFunction() bool
func (SymbolKind) IsPrivate ¶
func (kind SymbolKind) IsPrivate() bool
type SymbolMap ¶
type SymbolMap struct { // This could be represented as a "map[Ref]Symbol" but a two-level array was // more efficient in profiles. This appears to be because it doesn't involve // a hash. This representation also makes it trivial to quickly merge symbol // maps from multiple files together. Each file only generates symbols in a // single inner array, so you can join the maps together by just make a // single outer array containing all of the inner arrays. See the comment on // "Ref" for more detail. Outer [][]Symbol }