summaryrefslogtreecommitdiff
path: root/vendor/github.com/ugorji/go/codec/helper_unsafe.go
diff options
context:
space:
mode:
authorLibravatar kim <grufwub@gmail.com>2025-06-30 15:19:09 +0200
committerLibravatar kim <gruf@noreply.codeberg.org>2025-06-30 15:19:09 +0200
commit8b0ea560279a5bf4479555d3924c763ddeecfcad (patch)
tree005e26d4a658e565594fb259cc17948659195822 /vendor/github.com/ugorji/go/codec/helper_unsafe.go
parent[chore] bumps ncruces/go-sqlite3 v0.26.1 => v0.26.3 (#4302) (diff)
downloadgotosocial-8b0ea560279a5bf4479555d3924c763ddeecfcad.tar.xz
[chore] update go dependencies (#4304)
- github.com/KimMachineGun/automemlimit v0.7.2 => v0.7.3 - github.com/gin-contrib/cors v1.7.5 => v1.7.6 - github.com/minio/minio-go/v7 v7.0.92 => v7.0.94 - github.com/spf13/cast v1.8.0 => v1.9.2 - github.com/uptrace/bun{,/*} v1.2.11 => v1.2.14 - golang.org/x/image v0.27.0 => v0.28.0 - golang.org/x/net v0.40.0 => v0.41.0 - code.superseriousbusiness.org/go-swagger v0.31.0-gts-go1.23-fix => v0.32.3-gts-go1.23-fix Reviewed-on: https://codeberg.org/superseriousbusiness/gotosocial/pulls/4304 Co-authored-by: kim <grufwub@gmail.com> Co-committed-by: kim <grufwub@gmail.com>
Diffstat (limited to 'vendor/github.com/ugorji/go/codec/helper_unsafe.go')
-rw-r--r--vendor/github.com/ugorji/go/codec/helper_unsafe.go568
1 files changed, 238 insertions, 330 deletions
diff --git a/vendor/github.com/ugorji/go/codec/helper_unsafe.go b/vendor/github.com/ugorji/go/codec/helper_unsafe.go
index 4e29b030b..31f3266a4 100644
--- a/vendor/github.com/ugorji/go/codec/helper_unsafe.go
+++ b/vendor/github.com/ugorji/go/codec/helper_unsafe.go
@@ -1,12 +1,15 @@
// Copyright (c) 2012-2020 Ugorji Nwoke. All rights reserved.
// Use of this source code is governed by a MIT license found in the LICENSE file.
-//go:build !safe && !codec.safe && !appengine && go1.9
-// +build !safe,!codec.safe,!appengine,go1.9
-
-// minimum of go 1.9 is needed, as that is the minimum for all features and linked functions we need
-// - typedmemclr was introduced in go 1.8
-// - mapassign_fastXXX was introduced in go 1.9
+//go:build !safe && !codec.safe && !appengine && go1.21
+
+// minimum of go 1.21 is needed, as that is the minimum for all features and linked functions we need
+// - typedmemclr : go1.8
+// - mapassign_fastXXX: go1.9
+// - clear was added in go1.21
+// - unsafe.String(Data): go1.20
+// - unsafe.Add: go1.17
+// - generics/any: go1.18
// etc
package codec
@@ -21,7 +24,7 @@ import (
// This file has unsafe variants of some helper functions.
// MARKER: See helper_unsafe.go for the usage documentation.
-
+//
// There are a number of helper_*unsafe*.go files.
//
// - helper_unsafe
@@ -41,19 +44,32 @@ import (
// As of March 2021, we cannot differentiate whether running with gccgo or gollvm
// using a build constraint, as both satisfy 'gccgo' build tag.
// Consequently, we must use the lowest common denominator to support both.
-
+//
// For reflect.Value code, we decided to do the following:
// - if we know the kind, we can elide conditional checks for
// - SetXXX (Int, Uint, String, Bool, etc)
// - SetLen
//
-// We can also optimize
-// - IsNil
-
+// We can also optimize many others, incl IsNil, etc
+//
// MARKER: Some functions here will not be hit during code coverage runs due to optimizations, e.g.
// - rvCopySlice: called by decode if rvGrowSlice did not set new slice into pointer to orig slice.
// however, helper_unsafe sets it, so no need to call rvCopySlice later
// - rvSlice: same as above
+//
+// MARKER: Handling flagIndir ----
+//
+// flagIndir means that the reflect.Value holds a pointer to the data itself.
+//
+// flagIndir can be set for:
+// - references
+// Here, type.IfaceIndir() --> false
+// flagIndir is usually false (except when the value is addressable, where in flagIndir may be true)
+// - everything else (numbers, bools, string, slice, struct, etc).
+// Here, type.IfaceIndir() --> true
+// flagIndir is always true
+//
+// This knowlege is used across this file, e.g. in rv2i and rvRefPtr
const safeMode = false
@@ -88,7 +104,9 @@ const (
const transientSizeMax = 64
// should struct/array support internal strings and slices?
-const transientValueHasStringSlice = false
+// const transientValueHasStringSlice = false
+
+func isTransientType4Size(size uint32) bool { return size <= transientSizeMax }
type unsafeString struct {
Data unsafe.Pointer
@@ -144,7 +162,8 @@ func (x *unsafePerTypeElem) addrFor(k reflect.Kind) unsafe.Pointer {
x.slice = unsafeSlice{} // memclr
return unsafe.Pointer(&x.slice)
}
- x.arr = [transientSizeMax]byte{} // memclr
+ clear(x.arr[:])
+ // x.arr = [transientSizeMax]byte{} // memclr
return unsafe.Pointer(&x.arr)
}
@@ -152,9 +171,7 @@ type perType struct {
elems [2]unsafePerTypeElem
}
-type decPerType struct {
- perType
-}
+type decPerType = perType
type encPerType struct{}
@@ -183,19 +200,6 @@ func byteAt(b []byte, index uint) byte {
return *(*byte)(unsafe.Pointer(uintptr((*unsafeSlice)(unsafe.Pointer(&b)).Data) + uintptr(index)))
}
-func byteSliceOf(b []byte, start, end uint) []byte {
- s := (*unsafeSlice)(unsafe.Pointer(&b))
- s.Data = unsafe.Pointer(uintptr(s.Data) + uintptr(start))
- s.Len = int(end - start)
- s.Cap -= int(start)
- return b
-}
-
-// func byteSliceWithLen(b []byte, length uint) []byte {
-// (*unsafeSlice)(unsafe.Pointer(&b)).Len = int(length)
-// return b
-// }
-
func setByteAt(b []byte, index uint, val byte) {
// b[index] = val
*(*byte)(unsafe.Pointer(uintptr((*unsafeSlice)(unsafe.Pointer(&b)).Data) + uintptr(index))) = val
@@ -222,49 +226,26 @@ func byteSliceSameData(v1 []byte, v2 []byte) bool {
return (*unsafeSlice)(unsafe.Pointer(&v1)).Data == (*unsafeSlice)(unsafe.Pointer(&v2)).Data
}
-// MARKER: okBytesN functions will copy N bytes into the top slots of the return array.
-// These functions expect that the bound check already occured and are are valid.
-// copy(...) does a number of checks which are unnecessary in this situation when in bounds.
-
-func okBytes2(b []byte) [2]byte {
- return *((*[2]byte)(((*unsafeSlice)(unsafe.Pointer(&b))).Data))
-}
-
-func okBytes3(b []byte) [3]byte {
- return *((*[3]byte)(((*unsafeSlice)(unsafe.Pointer(&b))).Data))
-}
-
-func okBytes4(b []byte) [4]byte {
- return *((*[4]byte)(((*unsafeSlice)(unsafe.Pointer(&b))).Data))
+// isNil checks - without much effort - if an interface is nil.
+//
+// returned rv is not guaranteed to be valid (e.g. if v == nil).
+//
+// Note that this will handle all pointer-sized types e.g.
+// pointer, map, chan, func, etc.
+func isNil(v interface{}, checkPtr bool) (rv reflect.Value, b bool) {
+ b = ((*unsafeIntf)(unsafe.Pointer(&v))).ptr == nil
+ return
}
-func okBytes8(b []byte) [8]byte {
- return *((*[8]byte)(((*unsafeSlice)(unsafe.Pointer(&b))).Data))
+func ptrToLowLevel[T any](ptr *T) unsafe.Pointer {
+ return unsafe.Pointer(ptr)
}
-// isNil says whether the value v is nil.
-// This applies to references like map/ptr/unsafepointer/chan/func,
-// and non-reference values like interface/slice.
-func isNil(v interface{}) (rv reflect.Value, isnil bool) {
- var ui = (*unsafeIntf)(unsafe.Pointer(&v))
- isnil = ui.ptr == nil
- if !isnil {
- rv, isnil = unsafeIsNilIntfOrSlice(ui, v)
- }
- return
+func lowLevelToPtr[T any](v unsafe.Pointer) *T {
+ return (*T)(v)
}
-func unsafeIsNilIntfOrSlice(ui *unsafeIntf, v interface{}) (rv reflect.Value, isnil bool) {
- rv = reflect.ValueOf(v) // reflect.ValueOf is currently not inline'able - so call it directly
- tk := rv.Kind()
- isnil = (tk == reflect.Interface || tk == reflect.Slice) && *(*unsafe.Pointer)(ui.ptr) == nil
- return
-}
-
-// return the pointer for a reference (map/chan/func/pointer/unsafe.Pointer).
-// true references (map, func, chan, ptr - NOT slice) may be double-referenced? as flagIndir
-//
-// Assumes that v is a reference (map/func/chan/ptr/func)
+// Given that v is a reference (map/func/chan/ptr/unsafepointer) kind, return the pointer
func rvRefPtr(v *unsafeReflectValue) unsafe.Pointer {
if v.flag&unsafeFlagIndir != 0 {
return *(*unsafe.Pointer)(v.ptr)
@@ -295,13 +276,6 @@ func rv4istr(i interface{}) (v reflect.Value) {
}
func rv2i(rv reflect.Value) (i interface{}) {
- // We tap into implememtation details from
- // the source go stdlib reflect/value.go, and trims the implementation.
- //
- // e.g.
- // - a map/ptr is a reference, thus flagIndir is not set on it
- // - an int/slice is not a reference, thus flagIndir is set on it
-
urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
if refBitset.isset(byte(rv.Kind())) && urv.flag&unsafeFlagIndir != 0 {
urv.ptr = *(*unsafe.Pointer)(urv.ptr)
@@ -316,12 +290,22 @@ func rvAddr(rv reflect.Value, ptrType reflect.Type) reflect.Value {
return rv
}
+// return true if this rv - got from a pointer kind - is nil.
+// For now, only use for struct fields of pointer types, as we're guaranteed
+// that flagIndir will never be set.
+func rvPtrIsNil(rv reflect.Value) bool {
+ return rvIsNil(rv)
+}
+
+// checks if a nil'able value is nil
func rvIsNil(rv reflect.Value) bool {
urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- if urv.flag&unsafeFlagIndir != 0 {
- return *(*unsafe.Pointer)(urv.ptr) == nil
+ if urv.flag&unsafeFlagIndir == 0 {
+ return urv.ptr == nil
}
- return urv.ptr == nil
+ // flagIndir is set for a reference (ptr/map/func/unsafepointer/chan)
+ // OR kind is slice/interface
+ return *(*unsafe.Pointer)(urv.ptr) == nil
}
func rvSetSliceLen(rv reflect.Value, length int) {
@@ -499,29 +483,62 @@ func isEmptyValueFallbackRecur(urv *unsafeReflectValue, v reflect.Value, tinfos
return false
}
+// is this an empty interface/ptr/struct/map/slice/chan/array
+func isEmptyContainerValue(v reflect.Value, tinfos *TypeInfos, recursive bool) bool {
+ urv := (*unsafeReflectValue)(unsafe.Pointer(&v))
+ switch v.Kind() {
+ case reflect.Slice:
+ return (*unsafeSlice)(urv.ptr).Len == 0
+ case reflect.Struct:
+ if tinfos == nil {
+ tinfos = defTypeInfos
+ }
+ ti := tinfos.find(uintptr(urv.typ))
+ if ti == nil {
+ ti = tinfos.load(v.Type())
+ }
+ return unsafeCmpZero(urv.ptr, int(ti.size))
+ case reflect.Interface, reflect.Ptr:
+ // isnil := urv.ptr == nil // (not sufficient, as a pointer value encodes the type)
+ isnil := urv.ptr == nil || *(*unsafe.Pointer)(urv.ptr) == nil
+ if recursive && !isnil {
+ return isEmptyValue(v.Elem(), tinfos, recursive)
+ }
+ return isnil
+ case reflect.Chan:
+ return urv.ptr == nil || len_chan(rvRefPtr(urv)) == 0
+ case reflect.Map:
+ return urv.ptr == nil || len_map(rvRefPtr(urv)) == 0
+ case reflect.Array:
+ return v.Len() == 0 ||
+ urv.ptr == nil ||
+ urv.typ == nil ||
+ rtsize2(urv.typ) == 0 ||
+ unsafeCmpZero(urv.ptr, int(rtsize2(urv.typ)))
+ }
+ return false
+}
+
// --------------------------
type structFieldInfos struct {
- c unsafe.Pointer // source
- s unsafe.Pointer // sorted
+ c unsafe.Pointer // source
+ s unsafe.Pointer // sorted
+ t uint8To32TrieNode
+
length int
+
+ // byName map[string]*structFieldInfo // find sfi given a name
}
+// func (x *structFieldInfos) load(source, sorted []*structFieldInfo, sourceNames, sortedNames []string) {
func (x *structFieldInfos) load(source, sorted []*structFieldInfo) {
- s := (*unsafeSlice)(unsafe.Pointer(&sorted))
- x.s = s.Data
- x.length = s.Len
+ var s *unsafeSlice
s = (*unsafeSlice)(unsafe.Pointer(&source))
x.c = s.Data
-}
-
-func (x *structFieldInfos) sorted() (v []*structFieldInfo) {
- *(*unsafeSlice)(unsafe.Pointer(&v)) = unsafeSlice{x.s, x.length, x.length}
- // s := (*unsafeSlice)(unsafe.Pointer(&v))
- // s.Data = x.sorted0
- // s.Len = x.length
- // s.Cap = s.Len
- return
+ x.length = s.Len
+ s = (*unsafeSlice)(unsafe.Pointer(&sorted))
+ x.s = s.Data
}
func (x *structFieldInfos) source() (v []*structFieldInfo) {
@@ -529,66 +546,48 @@ func (x *structFieldInfos) source() (v []*structFieldInfo) {
return
}
-// atomicXXX is expected to be 2 words (for symmetry with atomic.Value)
-//
-// Note that we do not atomically load/store length and data pointer separately,
-// as this could lead to some races. Instead, we atomically load/store cappedSlice.
-//
-// Note: with atomic.(Load|Store)Pointer, we MUST work with an unsafe.Pointer directly.
-
-// ----------------------
-type atomicTypeInfoSlice struct {
- v unsafe.Pointer // *[]rtid2ti
-}
-
-func (x *atomicTypeInfoSlice) load() (s []rtid2ti) {
- x2 := atomic.LoadPointer(&x.v)
- if x2 != nil {
- s = *(*[]rtid2ti)(x2)
- }
+func (x *structFieldInfos) sorted() (v []*structFieldInfo) {
+ *(*unsafeSlice)(unsafe.Pointer(&v)) = unsafeSlice{x.s, x.length, x.length}
return
}
-func (x *atomicTypeInfoSlice) store(p []rtid2ti) {
- atomic.StorePointer(&x.v, unsafe.Pointer(&p))
+// --------------------------
+
+type uint8To32TrieNodeNoKids struct {
+ key uint8
+ valid bool // the value marks the end of a full stored string
+ numkids uint8
+ _ byte // padding
+ value uint32
}
-// MARKER: in safe mode, atomicXXX are atomic.Value, which contains an interface{}.
-// This is 2 words.
-// consider padding atomicXXX here with a uintptr, so they fit into 2 words also.
+type uint8To32TrieNodeKids = *uint8To32TrieNode
-// --------------------------
-type atomicRtidFnSlice struct {
- v unsafe.Pointer // *[]codecRtidFn
+func (x *uint8To32TrieNode) setKids(kids []uint8To32TrieNode) {
+ x.numkids = uint8(len(kids))
+ x.kids = &kids[0]
}
-
-func (x *atomicRtidFnSlice) load() (s []codecRtidFn) {
- x2 := atomic.LoadPointer(&x.v)
- if x2 != nil {
- s = *(*[]codecRtidFn)(x2)
- }
+func (x *uint8To32TrieNode) getKids() (v []uint8To32TrieNode) {
+ *(*unsafeSlice)(unsafe.Pointer(&v)) = unsafeSlice{unsafe.Pointer(x.kids), int(x.numkids), int(x.numkids)}
return
}
-
-func (x *atomicRtidFnSlice) store(p []codecRtidFn) {
- atomic.StorePointer(&x.v, unsafe.Pointer(&p))
-}
+func (x *uint8To32TrieNode) truncKids() { x.numkids = 0 }
// --------------------------
-type atomicClsErr struct {
- v unsafe.Pointer // *clsErr
+
+// Note that we do not atomically load/store length and data pointer separately,
+// as this could lead to some races. Instead, we atomically load/store cappedSlice.
+
+type atomicRtidFnSlice struct {
+ v unsafe.Pointer // *[]codecRtidFn
}
-func (x *atomicClsErr) load() (e clsErr) {
- x2 := (*clsErr)(atomic.LoadPointer(&x.v))
- if x2 != nil {
- e = *x2
- }
- return
+func (x *atomicRtidFnSlice) load() (s unsafe.Pointer) {
+ return atomic.LoadPointer(&x.v)
}
-func (x *atomicClsErr) store(p clsErr) {
- atomic.StorePointer(&x.v, unsafe.Pointer(&p))
+func (x *atomicRtidFnSlice) store(p unsafe.Pointer) {
+ atomic.StorePointer(&x.v, p)
}
// --------------------------
@@ -660,98 +659,79 @@ func (n *fauxUnion) rb() (v reflect.Value) {
// --------------------------
func rvSetBytes(rv reflect.Value, v []byte) {
- urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- *(*[]byte)(urv.ptr) = v
+ *(*[]byte)(rvPtr(rv)) = v
}
func rvSetString(rv reflect.Value, v string) {
- urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- *(*string)(urv.ptr) = v
+ *(*string)(rvPtr(rv)) = v
}
func rvSetBool(rv reflect.Value, v bool) {
- urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- *(*bool)(urv.ptr) = v
+ *(*bool)(rvPtr(rv)) = v
}
func rvSetTime(rv reflect.Value, v time.Time) {
- urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- *(*time.Time)(urv.ptr) = v
+ *(*time.Time)(rvPtr(rv)) = v
}
func rvSetFloat32(rv reflect.Value, v float32) {
- urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- *(*float32)(urv.ptr) = v
+ *(*float32)(rvPtr(rv)) = v
}
func rvSetFloat64(rv reflect.Value, v float64) {
- urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- *(*float64)(urv.ptr) = v
+ *(*float64)(rvPtr(rv)) = v
}
func rvSetComplex64(rv reflect.Value, v complex64) {
- urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- *(*complex64)(urv.ptr) = v
+ *(*complex64)(rvPtr(rv)) = v
}
func rvSetComplex128(rv reflect.Value, v complex128) {
- urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- *(*complex128)(urv.ptr) = v
+ *(*complex128)(rvPtr(rv)) = v
}
func rvSetInt(rv reflect.Value, v int) {
- urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- *(*int)(urv.ptr) = v
+ *(*int)(rvPtr(rv)) = v
}
func rvSetInt8(rv reflect.Value, v int8) {
- urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- *(*int8)(urv.ptr) = v
+ *(*int8)(rvPtr(rv)) = v
}
func rvSetInt16(rv reflect.Value, v int16) {
- urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- *(*int16)(urv.ptr) = v
+ *(*int16)(rvPtr(rv)) = v
}
func rvSetInt32(rv reflect.Value, v int32) {
- urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- *(*int32)(urv.ptr) = v
+ *(*int32)(rvPtr(rv)) = v
}
func rvSetInt64(rv reflect.Value, v int64) {
- urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- *(*int64)(urv.ptr) = v
+ *(*int64)(rvPtr(rv)) = v
}
func rvSetUint(rv reflect.Value, v uint) {
- urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- *(*uint)(urv.ptr) = v
+ *(*uint)(rvPtr(rv)) = v
}
func rvSetUintptr(rv reflect.Value, v uintptr) {
- urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- *(*uintptr)(urv.ptr) = v
+ *(*uintptr)(rvPtr(rv)) = v
}
func rvSetUint8(rv reflect.Value, v uint8) {
- urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- *(*uint8)(urv.ptr) = v
+ *(*uint8)(rvPtr(rv)) = v
}
func rvSetUint16(rv reflect.Value, v uint16) {
- urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- *(*uint16)(urv.ptr) = v
+ *(*uint16)(rvPtr(rv)) = v
}
func rvSetUint32(rv reflect.Value, v uint32) {
- urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- *(*uint32)(urv.ptr) = v
+ *(*uint32)(rvPtr(rv)) = v
}
func rvSetUint64(rv reflect.Value, v uint64) {
- urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- *(*uint64)(urv.ptr) = v
+ *(*uint64)(rvPtr(rv)) = v
}
// ----------------
@@ -775,12 +755,10 @@ func rvSetDirect(rv reflect.Value, v reflect.Value) {
uv := (*unsafeReflectValue)(unsafe.Pointer(&v))
if uv.flag&unsafeFlagIndir == 0 {
*(*unsafe.Pointer)(urv.ptr) = uv.ptr
- } else if uv.ptr == unsafeZeroAddr {
- if urv.ptr != unsafeZeroAddr {
- typedmemclr(urv.typ, urv.ptr)
- }
- } else {
+ } else if uv.ptr != unsafeZeroAddr {
typedmemmove(urv.typ, urv.ptr, uv.ptr)
+ } else if urv.ptr != unsafeZeroAddr {
+ typedmemclr(urv.typ, urv.ptr)
}
}
@@ -812,11 +790,9 @@ func rvMakeSlice(rv reflect.Value, ti *typeInfo, xlen, xcap int) (_ reflect.Valu
// It is typically called when we know that SetLen(...) cannot be done.
func rvSlice(rv reflect.Value, length int) reflect.Value {
urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- var x []struct{}
- ux := (*unsafeSlice)(unsafe.Pointer(&x))
- *ux = *(*unsafeSlice)(urv.ptr)
+ ux := *(*unsafeSlice)(urv.ptr) // copy slice header
ux.Len = length
- urv.ptr = unsafe.Pointer(ux)
+ urv.ptr = unsafe.Pointer(&ux)
return rv
}
@@ -834,10 +810,16 @@ func rvGrowSlice(rv reflect.Value, ti *typeInfo, cap, incr int) (v reflect.Value
// ------------
-func rvSliceIndex(rv reflect.Value, i int, ti *typeInfo) (v reflect.Value) {
+func rvArrayIndex(rv reflect.Value, i int, ti *typeInfo, isSlice bool) (v reflect.Value) {
urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
uv := (*unsafeReflectValue)(unsafe.Pointer(&v))
- uv.ptr = unsafe.Pointer(uintptr(((*unsafeSlice)(urv.ptr)).Data) + uintptr(int(ti.elemsize)*i))
+ if isSlice {
+ uv.ptr = unsafe.Pointer(uintptr(((*unsafeSlice)(urv.ptr)).Data))
+ } else {
+ uv.ptr = unsafe.Pointer(uintptr(urv.ptr))
+ }
+ uv.ptr = unsafe.Add(uv.ptr, ti.elemsize*uint32(i))
+ // uv.ptr = unsafe.Pointer(ptr + uintptr(int(ti.elemsize)*i))
uv.typ = ((*unsafeIntf)(unsafe.Pointer(&ti.elem))).ptr
uv.flag = uintptr(ti.elemkind) | unsafeFlagIndir | unsafeFlagAddr
return
@@ -861,19 +843,11 @@ func rvCapSlice(rv reflect.Value) int {
return (*unsafeSlice)(urv.ptr).Cap
}
-func rvArrayIndex(rv reflect.Value, i int, ti *typeInfo) (v reflect.Value) {
- urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- uv := (*unsafeReflectValue)(unsafe.Pointer(&v))
- uv.ptr = unsafe.Pointer(uintptr(urv.ptr) + uintptr(int(ti.elemsize)*i))
- uv.typ = ((*unsafeIntf)(unsafe.Pointer(&ti.elem))).ptr
- uv.flag = uintptr(ti.elemkind) | unsafeFlagIndir | unsafeFlagAddr
- return
-}
-
// if scratch is nil, then return a writable view (assuming canAddr=true)
-func rvGetArrayBytes(rv reflect.Value, scratch []byte) (bs []byte) {
+func rvGetArrayBytes(rv reflect.Value, _ []byte) (bs []byte) {
urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
bx := (*unsafeSlice)(unsafe.Pointer(&bs))
+ // bx.Data, bx.Len, bx.Cap = urv.ptr, rv.Len(), bx.Len
bx.Data = urv.ptr
bx.Len = rv.Len()
bx.Cap = bx.Len
@@ -889,7 +863,7 @@ func rvGetArray4Slice(rv reflect.Value) (v reflect.Value) {
//
// Consequently, we use rvLenSlice, not rvCapSlice.
- t := reflectArrayOf(rvLenSlice(rv), rv.Type().Elem())
+ t := reflect.ArrayOf(rvLenSlice(rv), rv.Type().Elem())
// v = rvZeroAddrK(t, reflect.Array)
uv := (*unsafeReflectValue)(unsafe.Pointer(&v))
@@ -921,99 +895,84 @@ func rvCopySlice(dest, src reflect.Value, elemType reflect.Type) {
// ------------
+func rvPtr(rv reflect.Value) unsafe.Pointer {
+ return (*unsafeReflectValue)(unsafe.Pointer(&rv)).ptr
+}
+
func rvGetBool(rv reflect.Value) bool {
- v := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- return *(*bool)(v.ptr)
+ return *(*bool)(rvPtr(rv))
}
func rvGetBytes(rv reflect.Value) []byte {
- v := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- return *(*[]byte)(v.ptr)
+ return *(*[]byte)(rvPtr(rv))
}
func rvGetTime(rv reflect.Value) time.Time {
- v := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- return *(*time.Time)(v.ptr)
+ return *(*time.Time)(rvPtr(rv))
}
func rvGetString(rv reflect.Value) string {
- v := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- return *(*string)(v.ptr)
+ return *(*string)(rvPtr(rv))
}
func rvGetFloat64(rv reflect.Value) float64 {
- v := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- return *(*float64)(v.ptr)
+ return *(*float64)(rvPtr(rv))
}
func rvGetFloat32(rv reflect.Value) float32 {
- v := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- return *(*float32)(v.ptr)
+ return *(*float32)(rvPtr(rv))
}
func rvGetComplex64(rv reflect.Value) complex64 {
- v := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- return *(*complex64)(v.ptr)
+ return *(*complex64)(rvPtr(rv))
}
func rvGetComplex128(rv reflect.Value) complex128 {
- v := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- return *(*complex128)(v.ptr)
+ return *(*complex128)(rvPtr(rv))
}
func rvGetInt(rv reflect.Value) int {
- v := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- return *(*int)(v.ptr)
+ return *(*int)(rvPtr(rv))
}
func rvGetInt8(rv reflect.Value) int8 {
- v := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- return *(*int8)(v.ptr)
+ return *(*int8)(rvPtr(rv))
}
func rvGetInt16(rv reflect.Value) int16 {
- v := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- return *(*int16)(v.ptr)
+ return *(*int16)(rvPtr(rv))
}
func rvGetInt32(rv reflect.Value) int32 {
- v := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- return *(*int32)(v.ptr)
+ return *(*int32)(rvPtr(rv))
}
func rvGetInt64(rv reflect.Value) int64 {
- v := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- return *(*int64)(v.ptr)
+ return *(*int64)(rvPtr(rv))
}
func rvGetUint(rv reflect.Value) uint {
- v := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- return *(*uint)(v.ptr)
+ return *(*uint)(rvPtr(rv))
}
func rvGetUint8(rv reflect.Value) uint8 {
- v := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- return *(*uint8)(v.ptr)
+ return *(*uint8)(rvPtr(rv))
}
func rvGetUint16(rv reflect.Value) uint16 {
- v := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- return *(*uint16)(v.ptr)
+ return *(*uint16)(rvPtr(rv))
}
func rvGetUint32(rv reflect.Value) uint32 {
- v := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- return *(*uint32)(v.ptr)
+ return *(*uint32)(rvPtr(rv))
}
func rvGetUint64(rv reflect.Value) uint64 {
- v := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- return *(*uint64)(v.ptr)
+ return *(*uint64)(rvPtr(rv))
}
func rvGetUintptr(rv reflect.Value) uintptr {
- v := (*unsafeReflectValue)(unsafe.Pointer(&rv))
- return *(*uintptr)(v.ptr)
+ return *(*uintptr)(rvPtr(rv))
}
func rvLenMap(rv reflect.Value) int {
@@ -1027,32 +986,6 @@ func rvLenMap(rv reflect.Value) int {
return len_map(rvRefPtr((*unsafeReflectValue)(unsafe.Pointer(&rv))))
}
-// copy is an intrinsic, which may use asm if length is small,
-// or make a runtime call to runtime.memmove if length is large.
-// Performance suffers when you always call runtime.memmove function.
-//
-// Consequently, there's no value in a copybytes call - just call copy() directly
-
-// func copybytes(to, from []byte) (n int) {
-// n = (*unsafeSlice)(unsafe.Pointer(&from)).Len
-// memmove(
-// (*unsafeSlice)(unsafe.Pointer(&to)).Data,
-// (*unsafeSlice)(unsafe.Pointer(&from)).Data,
-// uintptr(n),
-// )
-// return
-// }
-
-// func copybytestr(to []byte, from string) (n int) {
-// n = (*unsafeSlice)(unsafe.Pointer(&from)).Len
-// memmove(
-// (*unsafeSlice)(unsafe.Pointer(&to)).Data,
-// (*unsafeSlice)(unsafe.Pointer(&from)).Data,
-// uintptr(n),
-// )
-// return
-// }
-
// Note: it is hard to find len(...) of an array type,
// as that is a field in the arrayType representing the array, and hard to introspect.
//
@@ -1065,24 +998,26 @@ func rvLenMap(rv reflect.Value) int {
//
// It is more performant to provide a value that the map entry is set into,
// and that elides the allocation.
-
-// go 1.4+ has runtime/hashmap.go or runtime/map.go which has a
-// hIter struct with the first 2 values being key and value
-// of the current iteration.
//
+// go 1.4 through go 1.23 (in runtime/hashmap.go or runtime/map.go) has a hIter struct
+// with the first 2 values being pointers for key and value of the current iteration.
+// The next 6 values are pointers, followed by numeric types (uintptr, uint8, bool, etc).
// This *hIter is passed to mapiterinit, mapiternext, mapiterkey, mapiterelem.
-// We bypass the reflect wrapper functions and just use the *hIter directly.
//
-// Though *hIter has many fields, we only care about the first 2.
+// In go 1.24, swissmap was introduced, and it provides a compatibility layer
+// for hIter (called linknameIter). This has only 2 pointer fields after the key and value pointers.
+//
+// Note: We bypass the reflect wrapper functions and just use the *hIter directly.
//
-// We directly embed this in unsafeMapIter below
+// When 'faking' these types with our own, we MUST ensure that the GC sees the pointers
+// appropriately. These are reflected in goversion_(no)swissmap_unsafe.go files.
+// In these files, we pad the extra spaces appropriately.
//
-// hiter is typically about 12 words, but we just fill up unsafeMapIter to 32 words,
-// so it fills multiple cache lines and can give some extra space to accomodate small growth.
+// Note: the faux hIter/linknameIter is directly embedded in unsafeMapIter below
type unsafeMapIter struct {
mtyp, mptr unsafe.Pointer
- k, v reflect.Value
+ k, v unsafeReflectValue
kisref bool
visref bool
mapvalues bool
@@ -1092,7 +1027,7 @@ type unsafeMapIter struct {
it struct {
key unsafe.Pointer
value unsafe.Pointer
- _ [20]uintptr // padding for other fields (to make up 32 words for enclosing struct)
+ _ unsafeMapIterPadding
}
}
@@ -1112,18 +1047,16 @@ func (t *unsafeMapIter) Next() (r bool) {
}
if helperUnsafeDirectAssignMapEntry || t.kisref {
- (*unsafeReflectValue)(unsafe.Pointer(&t.k)).ptr = t.it.key
+ t.k.ptr = t.it.key
} else {
- k := (*unsafeReflectValue)(unsafe.Pointer(&t.k))
- typedmemmove(k.typ, k.ptr, t.it.key)
+ typedmemmove(t.k.typ, t.k.ptr, t.it.key)
}
if t.mapvalues {
if helperUnsafeDirectAssignMapEntry || t.visref {
- (*unsafeReflectValue)(unsafe.Pointer(&t.v)).ptr = t.it.value
+ t.v.ptr = t.it.value
} else {
- v := (*unsafeReflectValue)(unsafe.Pointer(&t.v))
- typedmemmove(v.typ, v.ptr, t.it.value)
+ typedmemmove(t.v.typ, t.v.ptr, t.it.value)
}
}
@@ -1131,11 +1064,11 @@ func (t *unsafeMapIter) Next() (r bool) {
}
func (t *unsafeMapIter) Key() (r reflect.Value) {
- return t.k
+ return *(*reflect.Value)(unsafe.Pointer(&t.k))
}
func (t *unsafeMapIter) Value() (r reflect.Value) {
- return t.v
+ return *(*reflect.Value)(unsafe.Pointer(&t.v))
}
func (t *unsafeMapIter) Done() {}
@@ -1162,14 +1095,14 @@ func mapRange(t *mapIter, m, k, v reflect.Value, mapvalues bool) {
// t.it = (*unsafeMapHashIter)(reflect_mapiterinit(t.mtyp, t.mptr))
mapiterinit(t.mtyp, t.mptr, unsafe.Pointer(&t.it))
- t.k = k
+ t.k = *(*unsafeReflectValue)(unsafe.Pointer(&k))
t.kisref = refBitset.isset(byte(k.Kind()))
if mapvalues {
- t.v = v
+ t.v = *(*unsafeReflectValue)(unsafe.Pointer(&v))
t.visref = refBitset.isset(byte(v.Kind()))
} else {
- t.v = reflect.Value{}
+ t.v = unsafeReflectValue{}
}
}
@@ -1182,13 +1115,6 @@ func unsafeMapKVPtr(urv *unsafeReflectValue) unsafe.Pointer {
return urv.ptr
}
-// func mapDelete(m, k reflect.Value) {
-// var urv = (*unsafeReflectValue)(unsafe.Pointer(&k))
-// var kptr = unsafeMapKVPtr(urv)
-// urv = (*unsafeReflectValue)(unsafe.Pointer(&m))
-// mapdelete(urv.typ, rv2ptr(urv), kptr)
-// }
-
// return an addressable reflect value that can be used in mapRange and mapGet operations.
//
// all calls to mapGet or mapRange will call here to get an addressable reflect.Value.
@@ -1205,53 +1131,39 @@ func mapAddrLoopvarRV(t reflect.Type, k reflect.Kind) (rv reflect.Value) {
return
}
-// ---------- ENCODER optimized ---------------
-
-func (e *Encoder) jsondriver() *jsonEncDriver {
- return (*jsonEncDriver)((*unsafeIntf)(unsafe.Pointer(&e.e)).ptr)
-}
-
-func (d *Decoder) zerocopystate() bool {
- return d.decByteState == decByteStateZerocopy && d.h.ZeroCopy
-}
-
-func (d *Decoder) stringZC(v []byte) (s string) {
- // MARKER: inline zerocopystate directly so genHelper forwarding function fits within inlining cost
-
- // if d.zerocopystate() {
- if d.decByteState == decByteStateZerocopy && d.h.ZeroCopy {
- return stringView(v)
- }
- return d.string(v)
+func makeMapReflect(typ reflect.Type, size int) (rv reflect.Value) {
+ t := (*unsafeIntf)(unsafe.Pointer(&typ)).ptr
+ urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
+ urv.typ = t
+ urv.flag = uintptr(reflect.Map)
+ urv.ptr = makemap(t, size, nil)
+ return
}
-func (d *Decoder) mapKeyString(callFnRvk *bool, kstrbs, kstr2bs *[]byte) string {
- if !d.zerocopystate() {
- *callFnRvk = true
- if d.decByteState == decByteStateReuseBuf {
- *kstrbs = append((*kstrbs)[:0], (*kstr2bs)...)
- *kstr2bs = *kstrbs
- }
- }
- return stringView(*kstr2bs)
-}
-
-// ---------- DECODER optimized ---------------
-
-func (d *Decoder) jsondriver() *jsonDecDriver {
- return (*jsonDecDriver)((*unsafeIntf)(unsafe.Pointer(&d.d)).ptr)
+func (d *decoderBase) bytes2Str(in []byte, state dBytesAttachState) (s string, mutable bool) {
+ return stringView(in), state <= dBytesAttachBuffer
}
// ---------- structFieldInfo optimized ---------------
-func (n *structFieldInfoPathNode) rvField(v reflect.Value) (rv reflect.Value) {
+func (n *structFieldInfoNode) rvField(v reflect.Value) (rv reflect.Value) {
// we already know this is exported, and maybe embedded (based on what si says)
uv := (*unsafeReflectValue)(unsafe.Pointer(&v))
+
urv := (*unsafeReflectValue)(unsafe.Pointer(&rv))
// clear flagEmbedRO if necessary, and inherit permission bits from v
urv.flag = uv.flag&(unsafeFlagStickyRO|unsafeFlagIndir|unsafeFlagAddr) | uintptr(n.kind)
urv.typ = ((*unsafeIntf)(unsafe.Pointer(&n.typ))).ptr
urv.ptr = unsafe.Pointer(uintptr(uv.ptr) + uintptr(n.offset))
+
+ // *(*unsafeReflectValue)(unsafe.Pointer(&rv)) = unsafeReflectValue{
+ // unsafeIntf: unsafeIntf{
+ // typ: ((*unsafeIntf)(unsafe.Pointer(&n.typ))).ptr,
+ // ptr: unsafe.Pointer(uintptr(uv.ptr) + uintptr(n.offset)),
+ // },
+ // flag: uv.flag&(unsafeFlagStickyRO|unsafeFlagIndir|unsafeFlagAddr) | uintptr(n.kind),
+ // }
+
return
}
@@ -1299,10 +1211,6 @@ func unsafeNew(typ unsafe.Pointer) unsafe.Pointer {
// failing with "error: undefined reference" error.
// however, runtime.{mallocgc, newarray} are supported, so use that instead.
-//go:linkname memmove runtime.memmove
-//go:noescape
-func memmove(to, from unsafe.Pointer, n uintptr)
-
//go:linkname mallocgc runtime.mallocgc
//go:noescape
func mallocgc(size uintptr, typ unsafe.Pointer, needzero bool) unsafe.Pointer
@@ -1319,10 +1227,6 @@ func mapiterinit(typ unsafe.Pointer, m unsafe.Pointer, it unsafe.Pointer)
//go:noescape
func mapiternext(it unsafe.Pointer) (key unsafe.Pointer)
-//go:linkname mapdelete runtime.mapdelete
-//go:noescape
-func mapdelete(typ unsafe.Pointer, m unsafe.Pointer, key unsafe.Pointer)
-
//go:linkname mapassign runtime.mapassign
//go:noescape
func mapassign(typ unsafe.Pointer, m unsafe.Pointer, key unsafe.Pointer) unsafe.Pointer
@@ -1331,6 +1235,10 @@ func mapassign(typ unsafe.Pointer, m unsafe.Pointer, key unsafe.Pointer) unsafe.
//go:noescape
func mapaccess2(typ unsafe.Pointer, m unsafe.Pointer, key unsafe.Pointer) (val unsafe.Pointer, ok bool)
+//go:linkname makemap runtime.makemap
+//go:noescape
+func makemap(typ unsafe.Pointer, size int, h unsafe.Pointer) unsafe.Pointer
+
// reflect.typed{memmove, memclr, slicecopy} will handle checking if the type has pointers or not,
// and if a writeBarrier is needed, before delegating to the right method in the runtime.
//