Use a profiler to improve linker performance (#291)

jhump · web-flow · commit 016b00994f15 · 2024-04-22T19:24:21.000Z
After measuring the impact of #286, #287, and #290 and seeing it to be too modest. I decided to use a memory profiler, and it found "the good stuff". These changes had the largest impact on allocations and performance. When linking inputs that come from descriptor protos (as opposed to inputs that are compiled from sources and have ASTs), this resulted in a 23% reduction in latency and 70% reduction in allocations. This change features the following improvements: 1. `ast.NoSourceNode` now has a pointer receiver, so wrapping one in an `ast.Node` interface value doesn't incur an allocation to put the value on the heap. This also updates `parser.ParseResult` to refer to a single `*ast.NoSourceNode` when it has no AST, instead of allocating one in each call to get a node value. The `NoSourceNode`'s underlying type is now `ast.FileInfo` so that it can be allocation-free, even for the `NodeInfo` method (which previously was allocating a new `FileInfo` each time). 3. Don't allocate a slice to hold the set of checked files for each element being resolved. Instead, we allocate a single slice up front, and re-use that throughout. 4. Don't pro-actively allocate strings that only are used for error messages; instead defer construction of the change to the construction of the error.
diff --git a/ast/enum.go b/ast/enum.go
@@ -108,7 +108,7 @@ type EnumValueDeclNode interface {
 }
 
 var _ EnumValueDeclNode = (*EnumValueNode)(nil)
-var _ EnumValueDeclNode = NoSourceNode{}
+var _ EnumValueDeclNode = (*NoSourceNode)(nil)
 
 // EnumValueNode represents an enum declaration. Example:
 //
diff --git a/ast/field.go b/ast/field.go
@@ -41,7 +41,7 @@ var _ FieldDeclNode = (*FieldNode)(nil)
 var _ FieldDeclNode = (*GroupNode)(nil)
 var _ FieldDeclNode = (*MapFieldNode)(nil)
 var _ FieldDeclNode = (*SyntheticMapField)(nil)
-var _ FieldDeclNode = NoSourceNode{}
+var _ FieldDeclNode = (*NoSourceNode)(nil)
 
 // FieldNode represents a normal field declaration (not groups or maps). It
 // can represent extension fields as well as non-extension fields (both inside
@@ -394,7 +394,7 @@ type OneofDeclNode interface {
 
 var _ OneofDeclNode = (*OneofNode)(nil)
 var _ OneofDeclNode = (*SyntheticOneof)(nil)
-var _ OneofDeclNode = NoSourceNode{}
+var _ OneofDeclNode = (*NoSourceNode)(nil)
 
 // OneofNode represents a one-of declaration. Example:
 //
diff --git a/ast/file.go b/ast/file.go
@@ -25,7 +25,7 @@ type FileDeclNode interface {
 }
 
 var _ FileDeclNode = (*FileNode)(nil)
-var _ FileDeclNode = NoSourceNode{}
+var _ FileDeclNode = (*NoSourceNode)(nil)
 
 // FileNode is the root of the AST hierarchy. It represents an entire
 // protobuf source file.
diff --git a/ast/message.go b/ast/message.go
@@ -32,7 +32,7 @@ type MessageDeclNode interface {
 var _ MessageDeclNode = (*MessageNode)(nil)
 var _ MessageDeclNode = (*SyntheticGroupMessageNode)(nil)
 var _ MessageDeclNode = (*SyntheticMapEntryNode)(nil)
-var _ MessageDeclNode = NoSourceNode{}
+var _ MessageDeclNode = (*NoSourceNode)(nil)
 
 // MessageNode represents a message declaration. Example:
 //
diff --git a/ast/no_source.go b/ast/no_source.go
@@ -41,104 +41,102 @@ func (s unknownSpan) End() SourcePos {
 // NoSourceNode is a placeholder AST node that implements numerous
 // interfaces in this package. It can be used to represent an AST
 // element for a file whose source is not available.
-type NoSourceNode struct {
-	filename string
-}
+type NoSourceNode FileInfo
 
 // NewNoSourceNode creates a new NoSourceNode for the given filename.
-func NewNoSourceNode(filename string) NoSourceNode {
-	return NoSourceNode{filename: filename}
+func NewNoSourceNode(filename string) *NoSourceNode {
+	return &NoSourceNode{name: filename}
 }
 
-func (n NoSourceNode) Name() string {
-	return n.filename
+func (n *NoSourceNode) Name() string {
+	return n.name
 }
 
-func (n NoSourceNode) Start() Token {
+func (n *NoSourceNode) Start() Token {
 	return 0
 }
 
-func (n NoSourceNode) End() Token {
+func (n *NoSourceNode) End() Token {
 	return 0
 }
 
-func (n NoSourceNode) NodeInfo(Node) NodeInfo {
+func (n *NoSourceNode) NodeInfo(Node) NodeInfo {
 	return NodeInfo{
-		fileInfo: &FileInfo{name: n.filename},
+		fileInfo: (*FileInfo)(n),
 	}
 }
 
-func (n NoSourceNode) GetSyntax() Node {
+func (n *NoSourceNode) GetSyntax() Node {
 	return n
 }
 
-func (n NoSourceNode) GetName() Node {
+func (n *NoSourceNode) GetName() Node {
 	return n
 }
 
-func (n NoSourceNode) GetValue() ValueNode {
+func (n *NoSourceNode) GetValue() ValueNode {
 	return n
 }
 
-func (n NoSourceNode) FieldLabel() Node {
+func (n *NoSourceNode) FieldLabel() Node {
 	return n
 }
 
-func (n NoSourceNode) FieldName() Node {
+func (n *NoSourceNode) FieldName() Node {
 	return n
 }
 
-func (n NoSourceNode) FieldType() Node {
+func (n *NoSourceNode) FieldType() Node {
 	return n
 }
 
-func (n NoSourceNode) FieldTag() Node {
+func (n *NoSourceNode) FieldTag() Node {
 	return n
 }
 
-func (n NoSourceNode) FieldExtendee() Node {
+func (n *NoSourceNode) FieldExtendee() Node {
 	return n
 }
 
-func (n NoSourceNode) GetGroupKeyword() Node {
+func (n *NoSourceNode) GetGroupKeyword() Node {
 	return n
 }
 
-func (n NoSourceNode) GetOptions() *CompactOptionsNode {
+func (n *NoSourceNode) GetOptions() *CompactOptionsNode {
 	return nil
 }
 
-func (n NoSourceNode) RangeStart() Node {
+func (n *NoSourceNode) RangeStart() Node {
 	return n
 }
 
-func (n NoSourceNode) RangeEnd() Node {
+func (n *NoSourceNode) RangeEnd() Node {
 	return n
 }
 
-func (n NoSourceNode) GetNumber() Node {
+func (n *NoSourceNode) GetNumber() Node {
 	return n
 }
 
-func (n NoSourceNode) MessageName() Node {
+func (n *NoSourceNode) MessageName() Node {
 	return n
 }
 
-func (n NoSourceNode) OneofName() Node {
+func (n *NoSourceNode) OneofName() Node {
 	return n
 }
 
-func (n NoSourceNode) GetInputType() Node {
+func (n *NoSourceNode) GetInputType() Node {
 	return n
 }
 
-func (n NoSourceNode) GetOutputType() Node {
+func (n *NoSourceNode) GetOutputType() Node {
 	return n
 }
 
-func (n NoSourceNode) Value() interface{} {
+func (n *NoSourceNode) Value() interface{} {
 	return nil
 }
 
-func (n NoSourceNode) RangeOptions(func(*OptionNode) bool) {
+func (n *NoSourceNode) RangeOptions(func(*OptionNode) bool) {
 }
diff --git a/ast/options.go b/ast/options.go
@@ -26,7 +26,7 @@ type OptionDeclNode interface {
 }
 
 var _ OptionDeclNode = (*OptionNode)(nil)
-var _ OptionDeclNode = NoSourceNode{}
+var _ OptionDeclNode = (*NoSourceNode)(nil)
 
 // OptionNode represents the declaration of a single option for an element.
 // It is used both for normal option declarations (start with "option" keyword
@@ -410,4 +410,4 @@ var _ NodeWithOptions = RPCDeclNode(nil)
 var _ NodeWithOptions = FieldDeclNode(nil)
 var _ NodeWithOptions = EnumValueDeclNode(nil)
 var _ NodeWithOptions = (*ExtensionRangeNode)(nil)
-var _ NodeWithOptions = NoSourceNode{}
+var _ NodeWithOptions = (*NoSourceNode)(nil)
diff --git a/ast/ranges.go b/ast/ranges.go
@@ -108,7 +108,7 @@ type RangeDeclNode interface {
 }
 
 var _ RangeDeclNode = (*RangeNode)(nil)
-var _ RangeDeclNode = NoSourceNode{}
+var _ RangeDeclNode = (*NoSourceNode)(nil)
 
 // RangeNode represents a range expression, used in both extension ranges and
 // reserved ranges. Example:
diff --git a/ast/service.go b/ast/service.go
@@ -108,7 +108,7 @@ type RPCDeclNode interface {
 }
 
 var _ RPCDeclNode = (*RPCNode)(nil)
-var _ RPCDeclNode = NoSourceNode{}
+var _ RPCDeclNode = (*NoSourceNode)(nil)
 
 // RPCNode represents an RPC declaration. Example:
 //
diff --git a/ast/values.go b/ast/values.go
@@ -54,7 +54,7 @@ var _ ValueNode = (*SpecialFloatLiteralNode)(nil)
 var _ ValueNode = (*SignedFloatLiteralNode)(nil)
 var _ ValueNode = (*ArrayLiteralNode)(nil)
 var _ ValueNode = (*MessageLiteralNode)(nil)
-var _ ValueNode = NoSourceNode{}
+var _ ValueNode = (*NoSourceNode)(nil)
 
 // StringValueNode is an AST node that represents a string literal.
 // Such a node can be a single literal (*StringLiteralNode) or a
diff --git a/linker/resolve.go b/linker/resolve.go
diff --git a/parser/result.go b/parser/result.go

Original file line number	Diff line number	Diff line change
`@@ -108,7 +108,7 @@ type EnumValueDeclNode interface {`
`108`	`108`	`}`
`109`	`109`
`110`	`110`	`var _ EnumValueDeclNode = (*EnumValueNode)(nil)`
`111`		`-var _ EnumValueDeclNode = NoSourceNode{}`
	`111`	`+var _ EnumValueDeclNode = (*NoSourceNode)(nil)`
`112`	`112`
`113`	`113`	`// EnumValueNode represents an enum declaration. Example:`
`114`	`114`	`//`
Original file line number	Diff line number	Diff line change
`@@ -25,7 +25,7 @@ type FileDeclNode interface {`
`25`	`25`	`}`
`26`	`26`
`27`	`27`	`var _ FileDeclNode = (*FileNode)(nil)`
`28`		`-var _ FileDeclNode = NoSourceNode{}`
	`28`	`+var _ FileDeclNode = (*NoSourceNode)(nil)`
`29`	`29`
`30`	`30`	`// FileNode is the root of the AST hierarchy. It represents an entire`
`31`	`31`	`// protobuf source file.`
Original file line number	Diff line number	Diff line change
`@@ -41,104 +41,102 @@ func (s unknownSpan) End() SourcePos {`
`41`	`41`	`// NoSourceNode is a placeholder AST node that implements numerous`
`42`	`42`	`// interfaces in this package. It can be used to represent an AST`
`43`	`43`	`// element for a file whose source is not available.`
`44`		`-type NoSourceNode struct {`
`45`		`- filename string`
`46`		`-}`
	`44`	`+type NoSourceNode FileInfo`
`47`	`45`
`48`	`46`	`// NewNoSourceNode creates a new NoSourceNode for the given filename.`
`49`		`-func NewNoSourceNode(filename string) NoSourceNode {`
`50`		`- return NoSourceNode{filename: filename}`
	`47`	`+func NewNoSourceNode(filename string) *NoSourceNode {`
	`48`	`+ return &NoSourceNode{name: filename}`
`51`	`49`	`}`
`52`	`50`
`53`		`-func (n NoSourceNode) Name() string {`
`54`		`- return n.filename`
	`51`	`+func (n *NoSourceNode) Name() string {`
	`52`	`+ return n.name`
`55`	`53`	`}`
`56`	`54`
`57`		`-func (n NoSourceNode) Start() Token {`
	`55`	`+func (n *NoSourceNode) Start() Token {`
`58`	`56`	`return 0`
`59`	`57`	`}`
`60`	`58`
`61`		`-func (n NoSourceNode) End() Token {`
	`59`	`+func (n *NoSourceNode) End() Token {`
`62`	`60`	`return 0`
`63`	`61`	`}`
`64`	`62`
`65`		`-func (n NoSourceNode) NodeInfo(Node) NodeInfo {`
	`63`	`+func (n *NoSourceNode) NodeInfo(Node) NodeInfo {`
`66`	`64`	`return NodeInfo{`
`67`		`- fileInfo: &FileInfo{name: n.filename},`
	`65`	`+ fileInfo: (*FileInfo)(n),`
`68`	`66`	`}`
`69`	`67`	`}`
`70`	`68`
`71`		`-func (n NoSourceNode) GetSyntax() Node {`
	`69`	`+func (n *NoSourceNode) GetSyntax() Node {`
`72`	`70`	`return n`
`73`	`71`	`}`
`74`	`72`
`75`		`-func (n NoSourceNode) GetName() Node {`
	`73`	`+func (n *NoSourceNode) GetName() Node {`
`76`	`74`	`return n`
`77`	`75`	`}`
`78`	`76`
`79`		`-func (n NoSourceNode) GetValue() ValueNode {`
	`77`	`+func (n *NoSourceNode) GetValue() ValueNode {`
`80`	`78`	`return n`
`81`	`79`	`}`
`82`	`80`
`83`		`-func (n NoSourceNode) FieldLabel() Node {`
	`81`	`+func (n *NoSourceNode) FieldLabel() Node {`
`84`	`82`	`return n`
`85`	`83`	`}`
`86`	`84`
`87`		`-func (n NoSourceNode) FieldName() Node {`
	`85`	`+func (n *NoSourceNode) FieldName() Node {`
`88`	`86`	`return n`
`89`	`87`	`}`
`90`	`88`
`91`		`-func (n NoSourceNode) FieldType() Node {`
	`89`	`+func (n *NoSourceNode) FieldType() Node {`
`92`	`90`	`return n`
`93`	`91`	`}`
`94`	`92`
`95`		`-func (n NoSourceNode) FieldTag() Node {`
	`93`	`+func (n *NoSourceNode) FieldTag() Node {`
`96`	`94`	`return n`
`97`	`95`	`}`
`98`	`96`
`99`		`-func (n NoSourceNode) FieldExtendee() Node {`
	`97`	`+func (n *NoSourceNode) FieldExtendee() Node {`
`100`	`98`	`return n`
`101`	`99`	`}`
`102`	`100`
`103`		`-func (n NoSourceNode) GetGroupKeyword() Node {`
	`101`	`+func (n *NoSourceNode) GetGroupKeyword() Node {`
`104`	`102`	`return n`
`105`	`103`	`}`
`106`	`104`
`107`		`-func (n NoSourceNode) GetOptions() *CompactOptionsNode {`
	`105`	`+func (n NoSourceNode) GetOptions() CompactOptionsNode {`
`108`	`106`	`return nil`
`109`	`107`	`}`
`110`	`108`
`111`		`-func (n NoSourceNode) RangeStart() Node {`
	`109`	`+func (n *NoSourceNode) RangeStart() Node {`
`112`	`110`	`return n`
`113`	`111`	`}`
`114`	`112`
`115`		`-func (n NoSourceNode) RangeEnd() Node {`
	`113`	`+func (n *NoSourceNode) RangeEnd() Node {`
`116`	`114`	`return n`
`117`	`115`	`}`
`118`	`116`
`119`		`-func (n NoSourceNode) GetNumber() Node {`
	`117`	`+func (n *NoSourceNode) GetNumber() Node {`
`120`	`118`	`return n`
`121`	`119`	`}`
`122`	`120`
`123`		`-func (n NoSourceNode) MessageName() Node {`
	`121`	`+func (n *NoSourceNode) MessageName() Node {`
`124`	`122`	`return n`
`125`	`123`	`}`
`126`	`124`
`127`		`-func (n NoSourceNode) OneofName() Node {`
	`125`	`+func (n *NoSourceNode) OneofName() Node {`
`128`	`126`	`return n`
`129`	`127`	`}`
`130`	`128`
`131`		`-func (n NoSourceNode) GetInputType() Node {`
	`129`	`+func (n *NoSourceNode) GetInputType() Node {`
`132`	`130`	`return n`
`133`	`131`	`}`
`134`	`132`
`135`		`-func (n NoSourceNode) GetOutputType() Node {`
	`133`	`+func (n *NoSourceNode) GetOutputType() Node {`
`136`	`134`	`return n`
`137`	`135`	`}`
`138`	`136`
`139`		`-func (n NoSourceNode) Value() interface{} {`
	`137`	`+func (n *NoSourceNode) Value() interface{} {`
`140`	`138`	`return nil`
`141`	`139`	`}`
`142`	`140`
`143`		`-func (n NoSourceNode) RangeOptions(func(*OptionNode) bool) {`
	`141`	`+func (n NoSourceNode) RangeOptions(func(OptionNode) bool) {`
`144`	`142`	`}`
Original file line number	Diff line number	Diff line change
`@@ -108,7 +108,7 @@ type RangeDeclNode interface {`
`108`	`108`	`}`
`109`	`109`
`110`	`110`	`var _ RangeDeclNode = (*RangeNode)(nil)`
`111`		`-var _ RangeDeclNode = NoSourceNode{}`
	`111`	`+var _ RangeDeclNode = (*NoSourceNode)(nil)`
`112`	`112`
`113`	`113`	`// RangeNode represents a range expression, used in both extension ranges and`
`114`	`114`	`// reserved ranges. Example:`
Original file line number	Diff line number	Diff line change
`@@ -108,7 +108,7 @@ type RPCDeclNode interface {`
`108`	`108`	`}`
`109`	`109`
`110`	`110`	`var _ RPCDeclNode = (*RPCNode)(nil)`
`111`		`-var _ RPCDeclNode = NoSourceNode{}`
	`111`	`+var _ RPCDeclNode = (*NoSourceNode)(nil)`
`112`	`112`
`113`	`113`	`// RPCNode represents an RPC declaration. Example:`
`114`	`114`	`//`