Go 1.21 Release Notes
Introduction to Go 1.21
The latest Go release, version 1.21, arrives six months after Go 1.20. Most of its changes are in the implementation of the toolchain, runtime, and libraries. As always, the release maintains the Go 1 promise of compatibility; in fact, Go 1.21 improves upon that promise. We expect almost all Go programs to continue to compile and run as before.
Go 1.21 introduces a small change to the numbering of releases.
In the past, we used Go 1.N to refer to both the overall Go language version and release family
as well as the first release in that family.
Starting in Go 1.21, the first release is now Go 1.N.0.
Today we are releasing both the Go 1.21 language and its initial implementation, the Go 1.21.0 release.
These notes refer to “Go 1.21”; tools like go
version
will report “go1.21.0
”
(until you upgrade to Go 1.21.1).
See “Go versions” in the “Go Toolchains” documentation for details
about the new version numbering.
Changes to the language
Go 1.21 adds three new built-ins to the language.
-
The new functions
min
andmax
compute the smallest (or largest, formax
) value of a fixed number of given arguments. See the language spec for details. -
The new function
clear
deletes all elements from a map or zeroes all elements of a slice. See the language spec for details.
Package initialization order is now specified more precisely. The new algorithm is:
- Sort all packages by import path.
- Repeat until the list of packages is empty:
- Find the first package in the list for which all imports are already initialized.
- Initialize that package and remove it from the list.
Multiple improvements that increase the power and precision of type inference have been made.
- A (possibly partially instantiated generic) function may now be called with arguments that are themselves (possibly partially instantiated) generic functions. The compiler will attempt to infer the missing type arguments of the callee (as before) and, for each argument that is a generic function that is not fully instantiated, its missing type arguments (new). Typical use cases are calls to generic functions operating on containers (such as slices.IndexFunc) where a function argument may also be generic, and where the type argument of the called function and its arguments are inferred from the container type. More generally, a generic function may now be used without explicit instantiation when it is assigned to a variable or returned as a result value if the type arguments can be inferred from the assignment.
- Type inference now also considers methods when a value is assigned to an interface: type arguments for type parameters used in method signatures may be inferred from the corresponding parameter types of matching methods.
- Similarly, since a type argument must implement all the methods of its corresponding constraint, the methods of the type argument and constraint are matched which may lead to the inference of additional type arguments.
- If multiple untyped constant arguments of different kinds (such as an untyped int and an untyped floating-point constant) are passed to parameters with the same (not otherwise specified) type parameter type, instead of an error, now type inference determines the type using the same approach as an operator with untyped constant operands. This change brings the types inferred from untyped constant arguments in line with the types of constant expressions.
- Type inference is now precise when matching corresponding types in assignments: component types (such as the elements of slices, or the parameter types in function signatures) must be identical (given suitable type arguments) to match, otherwise inference fails. This change produces more accurate error messages: where in the past type inference may have succeeded incorrectly and lead to an invalid assignment, the compiler now reports an inference error if two types can't possibly match.
More generally, the description of type inference in the language spec has been clarified. Together, all these changes make type inference more powerful and inference failures less surprising.
Go 1.21 includes a preview of a language change we are considering for a future version of Go: making for loop variables per-iteration instead of per-loop, to avoid accidental sharing bugs. For details about how to try that language change, see the LoopvarExperiment wiki page.
Go 1.21 now defines that if a goroutine is panicking and recover was called directly by a deferred
function, the return value of recover is guaranteed not to be nil. To ensure this, calling panic
with a nil interface value (or an untyped nil) causes a run-time panic of type
*runtime.PanicNilError
.
To support programs written for older versions of Go, nil panics can be re-enabled by setting
GODEBUG=panicnil=1
.
This setting is enabled automatically when compiling a program whose main package
is in a module that declares go
1.20
or earlier.
Tools
Go 1.21 adds improved support for backwards compatibility and forwards compatibility in the Go toolchain.
To improve backwards compatibility, Go 1.21 formalizes
Go's use of the GODEBUG environment variable to control
the default behavior for changes that are non-breaking according to the
compatibility policy
but nonetheless may cause existing programs to break.
(For example, programs that depend on buggy behavior may break
when a bug is fixed, but bug fixes are not considered breaking changes.)
When Go must make this kind of behavior change,
it now chooses between the old and new behavior based on the
go
line in the workspace's go.work
file
or else the main module's go.mod
file.
Upgrading to a new Go toolchain but leaving the go
line
set to its original (older) Go version preserves the behavior of the older
toolchain.
With this compatibility support, the latest Go toolchain should always
be the best, most secure, implementation of an older version of Go.
See “Go, Backwards Compatibility, and GODEBUG” for details.
To improve forwards compatibility, Go 1.21 now reads the go
line
in a go.work
or go.mod
file as a strict
minimum requirement: go
1.21.0
means
that the workspace or module cannot be used with Go 1.20 or with Go 1.21rc1.
This allows projects that depend on fixes made in later versions of Go
to ensure that they are not used with earlier versions.
It also gives better error reporting for projects that make use of new Go features:
when the problem is that a newer Go version is needed,
that problem is reported clearly, instead of attempting to build the code
and printing errors about unresolved imports or syntax errors.
To make these new stricter version requirements easier to manage,
the go
command can now invoke not just the toolchain
bundled in its own release but also other Go toolchain versions found in the PATH
or downloaded on demand.
If a go.mod
or go.work
go
line
declares a minimum requirement on a newer version of Go, the go
command will find and run that version automatically.
The new toolchain
directive sets a suggested minimum toolchain to use,
which may be newer than the strict go
minimum.
See “Go Toolchains” for details.
Go command
The -pgo
build flag now defaults to -pgo=auto
,
and the restriction of specifying a single main package on the command
line is now removed. If a file named default.pgo
is present
in the main package's directory, the go
command will use
it to enable profile-guided optimization for building the corresponding
program.
The -C
dir
flag must now be the first
flag on the command-line when used.
The new go
test
option
-fullpath
prints full path names in test log messages,
rather than just base names.
The go
test
-c
flag now
supports writing test binaries for multiple packages, each to
pkg.test
where pkg
is the package name.
It is an error if more than one test package being compiled has a given package name.
The go
test
-o
flag now
accepts a directory argument, in which case test binaries are written to that
directory instead of the current directory.
Cgo
In files that import "C"
, the Go toolchain now
correctly reports errors for attempts to declare Go methods on C types.
Runtime
When printing very deep stacks, the runtime now prints the first 50 (innermost) frames followed by the bottom 50 (outermost) frames, rather than just printing the first 100 frames. This makes it easier to see how deeply recursive stacks started, and is especially valuable for debugging stack overflows.
On Linux platforms that support transparent huge pages, the Go runtime now manages which parts of the heap may be backed by huge pages more explicitly. This leads to better utilization of memory: small heaps should see less memory used (up to 50% in pathological cases) while large heaps should see fewer broken huge pages for dense parts of the heap, improving CPU usage and latency by up to 1%.
As a result of runtime-internal garbage collection tuning,
applications may see up to a 40% reduction in application tail latency
and a small decrease in memory use. Some applications may also observe
a small loss in throughput.
The memory use decrease should be proportional to the loss in
throughput, such that the previous release's throughput/memory
tradeoff may be recovered (with little change to latency) by
increasing GOGC
and/or GOMEMLIMIT
slightly.
Calls from C to Go on threads created in C require some setup to prepare for Go execution. On Unix platforms, this setup is now preserved across multiple calls from the same thread. This significantly reduces the overhead of subsequent C to Go calls from ~1-3 microseconds per call to ~100-200 nanoseconds per call.
Compiler
Profile-guide optimization (PGO), added as a preview in Go 1.20, is now ready
for general use. PGO enables additional optimizations on code identified as
hot by profiles of production workloads. As mentioned in the
Go command section, PGO is enabled by default for
binaries that contain a default.pgo
profile in the main
package directory. Performance improvements vary depending on application
behavior, with most programs from a representative set of Go programs seeing
between 2 and 7% improvement from enabling PGO. See the
PGO user guide for detailed documentation.
PGO builds can now devirtualize some interface method calls, adding a concrete call to the most common callee. This enables further optimization, such as inlining the callee.
Go 1.21 improves build speed by up to 6%, largely thanks to building the compiler itself with PGO.
Assembler
On amd64, frameless nosplit assembly functions are no longer automatically marked as NOFRAME
.
Instead, the NOFRAME
attribute must be explicitly specified if desired,
which is already the behavior on other architectures supporting frame pointers.
With this, the runtime now maintains the frame pointers for stack transitions.
The verifier that checks for incorrect uses of R15
when dynamic linking on amd64 has been improved.
Linker
On windows/amd64, the linker (with help from the compiler) now emits SEH unwinding data by default, which improves the integration of Go applications with Windows debuggers and other tools.
In Go 1.21 the linker (with help from the compiler) is now capable of deleting dead (unreferenced) global map variables, if the number of entries in the variable initializer is sufficiently large, and if the initializer expressions are side-effect free.
Core library
New log/slog package
The new log/slog package provides structured logging with levels. Structured logging emits key-value pairs to enable fast, accurate processing of large amounts of log data. The package supports integration with popular log analysis tools and services.
New testing/slogtest package
The new testing/slogtest package can help to validate slog.Handler implementations.
New slices package
The new slices package provides many common operations on slices, using generic functions that work with slices of any element type.
New maps package
The new maps package provides several common operations on maps, using generic functions that work with maps of any key or element type.
New cmp package
The new cmp package defines the type
constraint Ordered
and
two new generic functions
Less
and Compare
that are
useful with ordered
types.
Minor changes to the library
As always, there are various minor changes and updates to the library, made with the Go 1 promise of compatibility in mind. There are also various performance improvements, not enumerated here.
- archive/tar
-
The implementation of the
io/fs.FileInfo
interface returned byHeader.FileInfo
now implements aString
method that callsio/fs.FormatFileInfo
.
- archive/zip
-
The implementation of the
io/fs.FileInfo
interface returned byFileHeader.FileInfo
now implements aString
method that callsio/fs.FormatFileInfo
.The implementation of the
io/fs.DirEntry
interface returned by theio/fs.ReadDirFile.ReadDir
method of theio/fs.File
returned byReader.Open
now implements aString
method that callsio/fs.FormatDirEntry
.
- bytes
-
The
Buffer
type has two new methods:Available
andAvailableBuffer
. These may be used along with theWrite
method to append directly to theBuffer
.
- context
-
The new
WithoutCancel
function returns a copy of a context that is not canceled when the original context is canceled.The new
WithDeadlineCause
andWithTimeoutCause
functions provide a way to set a context cancellation cause when a deadline or timer expires. The cause may be retrieved with theCause
function.The new
AfterFunc
function registers a function to run after a context has been cancelled.An optimization means that the results of calling
Background
andTODO
and converting them to a shared type can be considered equal. In previous releases they were always different. ComparingContext
values for equality has never been well-defined, so this is not considered to be an incompatible change.
- crypto/ecdsa
-
PublicKey.Equal
andPrivateKey.Equal
now execute in constant time.
- crypto/elliptic
-
All of the
Curve
methods have been deprecated, along withGenerateKey
,Marshal
, andUnmarshal
. For ECDH operations, the newcrypto/ecdh
package should be used instead. For lower-level operations, use third-party modules such as filippo.io/nistec.
- crypto/rand
-
The
crypto/rand
package now uses thegetrandom
system call on NetBSD 10.0 and later.
- crypto/rsa
-
The performance of private RSA operations (decryption and signing) is now better than Go 1.19 for
GOARCH=amd64
andGOARCH=arm64
. It had regressed in Go 1.20.Due to the addition of private fields to
PrecomputedValues
,PrivateKey.Precompute
must be called for optimal performance even if deserializing (for example from JSON) a previously-precomputed private key.PublicKey.Equal
andPrivateKey.Equal
now execute in constant time.The
GenerateMultiPrimeKey
function and thePrecomputedValues.CRTValues
field have been deprecated.PrecomputedValues.CRTValues
will still be populated whenPrivateKey.Precompute
is called, but the values will not be used during decryption operations.
- crypto/sha256
-
SHA-224 and SHA-256 operations now use native instructions when available when
GOARCH=amd64
, providing a performance improvement on the order of 3-4x.
- crypto/tls
-
Servers now skip verifying client certificates (including not running
Config.VerifyPeerCertificate
) for resumed connections, besides checking the expiration time. This makes session tickets larger when client certificates are in use. Clients were already skipping verification on resumption, but now check the expiration time even ifConfig.InsecureSkipVerify
is set.Applications can now control the content of session tickets.
-
The new
SessionState
type describes a resumable session. -
The
SessionState.Bytes
method andParseSessionState
function serialize and deserialize aSessionState
. -
The
Config.WrapSession
andConfig.UnwrapSession
hooks convert aSessionState
to and from a ticket on the server side. -
The
Config.EncryptTicket
andConfig.DecryptTicket
methods provide a default implementation ofWrapSession
andUnwrapSession
. -
The
ClientSessionState.ResumptionState
method andNewResumptionState
function may be used by aClientSessionCache
implementation to store and resume sessions on the client side.
To reduce the potential for session tickets to be used as a tracking mechanism across connections, the server now issues new tickets on every resumption (if they are supported and not disabled) and tickets don't bear an identifier for the key that encrypted them anymore. If passing a large number of keys to
Conn.SetSessionTicketKeys
, this might lead to a noticeable performance cost.Both clients and servers now implement the Extended Master Secret extension (RFC 7627). The deprecation of
ConnectionState.TLSUnique
has been reverted, and is now set for resumed connections that support Extended Master Secret.The new
QUICConn
type provides support for QUIC implementations, including 0-RTT support. Note that this is not itself a QUIC implementation, and 0-RTT is still not supported in TLS.The new
VersionName
function returns the name for a TLS version number.The TLS alert codes sent from the server for client authentication failures have been improved. Previously, these failures always resulted in a "bad certificate" alert. Now, certain failures will result in more appropriate alert codes, as defined by RFC 5246 and RFC 8446:
- For TLS 1.3 connections, if the server is configured to require client authentication using RequireAnyClientCert or RequireAndVerifyClientCert, and the client does not provide any certificate, the server will now return the "certificate required" alert.
- If the client provides a certificate that is not signed by the set of trusted certificate authorities configured on the server, the server will return the "unknown certificate authority" alert.
- If the client provides a certificate that is either expired or not yet valid, the server will return the "expired certificate" alert.
- In all other scenarios related to client authentication failures, the server still returns "bad certificate".
-
The new
- crypto/x509
-
RevocationList.RevokedCertificates
has been deprecated and replaced with the newRevokedCertificateEntries
field, which is a slice ofRevocationListEntry
.RevocationListEntry
contains all of the fields inpkix.RevokedCertificate
, as well as the revocation reason code.Name constraints are now correctly enforced on non-leaf certificates, and not on the certificates where they are expressed.
- debug/elf
-
The new
File.DynValue
method may be used to retrieve the numeric values listed with a given dynamic tag.The constant flags permitted in a
DT_FLAGS_1
dynamic tag are now defined with typeDynFlag1
. These tags have names starting withDF_1
.The package now defines the constant
COMPRESS_ZSTD
.The package now defines the constant
R_PPC64_REL24_P9NOTOC
.
- debug/pe
-
Attempts to read from a section containing uninitialized data using
Section.Data
or the reader returned bySection.Open
now return an error.
- embed
-
The
io/fs.File
returned byFS.Open
now has aReadAt
method that implementsio.ReaderAt
.Calling
FS.Open.Stat
will return a type that now implements aString
method that callsio/fs.FormatFileInfo
.
- encoding/binary
-
The new
NativeEndian
variable may be used to convert between byte slices and integers using the current machine's native endianness.
- errors
-
The new
ErrUnsupported
error provides a standardized way to indicate that a requested operation may not be performed because it is unsupported. For example, a call toos.Link
when using a file system that does not support hard links.
- flag
-
The new
BoolFunc
function andFlagSet.BoolFunc
method define a flag that does not require an argument and calls a function when the flag is used. This is similar toFunc
but for a boolean flag.A flag definition (via
Bool
,BoolVar
,Int
,IntVar
, etc.) will panic ifSet
has already been called on a flag with the same name. This change is intended to detect cases where changes in initialization order cause flag operations to occur in a different order than expected. In many cases the fix to this problem is to introduce a explicit package dependence to correctly order the definition before anySet
operations.
- go/ast
-
The new
IsGenerated
predicate reports whether a file syntax tree contains the special comment that conventionally indicates that the file was generated by a tool. -
The new
File.GoVersion
field records the minimum Go version required by any//go:build
or// +build
directives.
- go/build
-
The package now parses build directives (comments that start with
//go:
) in file headers (before thepackage
declaration). These directives are available in the newPackage
fieldsDirectives
,TestDirectives
, andXTestDirectives
.
- go/build/constraint
-
The new
GoVersion
function returns the minimum Go version implied by a build expression.
- go/token
-
The new
File.Lines
method returns the file's line-number table in the same form as accepted byFile.SetLines
.
- go/types
-
The new
Package.GoVersion
method returns the Go language version used to check the package.
- hash/maphash
-
The
hash/maphash
package now has a pure Go implementation, selectable with thepurego
build tag.
- html/template
-
The new error
ErrJSTemplate
is returned when an action appears in a JavaScript template literal. Previously an unexported error was returned.
- io/fs
-
The new
FormatFileInfo
function returns a formatted version of aFileInfo
. The newFormatDirEntry
function returns a formatted version of aDirEntry
. The implementation ofDirEntry
returned byReadDir
now implements aString
method that callsFormatDirEntry
, and the same is true for theDirEntry
value passed toWalkDirFunc
.
- math/big
-
The new
Int.Float64
method returns the nearest floating-point value to a multi-precision integer, along with an indication of any rounding that occurred.
- net
-
On Linux, the net package can now use Multipath TCP when the kernel supports it. It is not used by default. To use Multipath TCP when available on a client, call the
Dialer.SetMultipathTCP
method before calling theDialer.Dial
orDialer.DialContext
methods. To use Multipath TCP when available on a server, call theListenConfig.SetMultipathTCP
method before calling theListenConfig.Listen
method. Specify the network as"tcp"
or"tcp4"
or"tcp6"
as usual. If Multipath TCP is not supported by the kernel or the remote host, the connection will silently fall back to TCP. To test whether a particular connection is using Multipath TCP, use theTCPConn.MultipathTCP
method.In a future Go release we may enable Multipath TCP by default on systems that support it.
- net/http
-
The new
ResponseController.EnableFullDuplex
method allows server handlers to concurrently read from an HTTP/1 request body while writing the response. Normally, the HTTP/1 server automatically consumes any remaining request body before starting to write the response, to avoid deadlocking clients which attempt to write a complete request before reading the response. TheEnableFullDuplex
method disables this behavior.The new
ErrSchemeMismatch
error is returned byClient
andTransport
when the server responds to an HTTPS request with an HTTP response.The net/http package now supports
errors.ErrUnsupported
, in that the expressionerrors.Is(http.ErrNotSupported, errors.ErrUnsupported)
will return true.
- os
-
Programs may now pass an empty
time.Time
value to theChtimes
function to leave either the access time or the modification time unchanged.On Windows the
File.Chdir
method now changes the current directory to the file, rather than always returning an error.On Unix systems, if a non-blocking descriptor is passed to
NewFile
, calling theFile.Fd
method will now return a non-blocking descriptor. Previously the descriptor was converted to blocking mode.On Windows calling
Truncate
on a non-existent file used to create an empty file. It now returns an error indicating that the file does not exist.On Windows calling
TempDir
now uses GetTempPath2W when available, instead of GetTempPathW. The new behavior is a security hardening measure that prevents temporary files created by processes running as SYSTEM to be accessed by non-SYSTEM processes.On Windows the os package now supports working with files whose names, stored as UTF-16, can't be represented as valid UTF-8.
On Windows
Lstat
now resolves symbolic links for paths ending with a path separator, consistent with its behavior on POSIX platforms.The implementation of the
io/fs.DirEntry
interface returned by theReadDir
function and theFile.ReadDir
method now implements aString
method that callsio/fs.FormatDirEntry
.The implementation of the
io/fs.FS
interface returned by theDirFS
function now implements theio/fs.ReadFileFS
and theio/fs.ReadDirFS
interfaces.
- path/filepath
-
The implementation of the
io/fs.DirEntry
interface passed to the function argument ofWalkDir
now implements aString
method that callsio/fs.FormatDirEntry
.
- reflect
-
In Go 1.21,
ValueOf
no longer forces its argument to be allocated on the heap, allowing aValue
's content to be allocated on the stack. Most operations on aValue
also allow the underlying value to be stack allocated.The new
Value
methodValue.Clear
clears the contents of a map or zeros the contents of a slice. This corresponds to the newclear
built-in added to the language.The
SliceHeader
andStringHeader
types are now deprecated. In new code preferunsafe.Slice
,unsafe.SliceData
,unsafe.String
, orunsafe.StringData
.
- regexp
-
Regexp
now definesMarshalText
andUnmarshalText
methods. These implementencoding.TextMarshaler
andencoding.TextUnmarshaler
and will be used by packages such as encoding/json.
- runtime
-
Textual stack traces produced by Go programs, such as those produced when crashing, calling
runtime.Stack
, or collecting a goroutine profile withdebug=2
, now include the IDs of the goroutines that created each goroutine in the stack trace.Crashing Go applications can now opt-in to Windows Error Reporting (WER) by setting the environment variable
GOTRACEBACK=wer
or callingdebug.SetTraceback("wer")
before the crash. Other than enabling WER, the runtime will behave as withGOTRACEBACK=crash
. On non-Windows systems,GOTRACEBACK=wer
is ignored.GODEBUG=cgocheck=2
, a thorough checker of cgo pointer passing rules, is no longer available as a debug option. Instead, it is available as an experiment usingGOEXPERIMENT=cgocheck2
. In particular this means that this mode has to be selected at build time instead of startup time.GODEBUG=cgocheck=1
is still available (and is still the default).A new type
Pinner
has been added to the runtime package.Pinner
s may be used to "pin" Go memory such that it may be used more freely by non-Go code. For instance, passing Go values that reference pinned Go memory to C code is now allowed. Previously, passing any such nested reference was disallowed by the cgo pointer passing rules. See the docs for more details.
- runtime/metrics
-
A few previously-internal GC metrics, such as live heap size, are now available.
GOGC
andGOMEMLIMIT
are also now available as metrics.
- runtime/trace
-
Collecting traces on amd64 and arm64 now incurs a substantially smaller CPU cost: up to a 10x improvement over the previous release.
Traces now contain explicit stop-the-world events for every reason the Go runtime might stop-the-world, not just garbage collection.
- sync
-
The new
OnceFunc
,OnceValue
, andOnceValues
functions capture a common use of Once to lazily initialize a value on first use.
- syscall
-
On Windows the
Fchdir
function now changes the current directory to its argument, rather than always returning an error.On FreeBSD
SysProcAttr
has a new fieldJail
that may be used to put the newly created process in a jailed environment.On Windows the syscall package now supports working with files whose names, stored as UTF-16, can't be represented as valid UTF-8. The
UTF16ToString
andUTF16FromString
functions now convert between UTF-16 data and WTF-8 strings. This is backward compatible as WTF-8 is a superset of the UTF-8 format that was used in earlier releases.Several error values match the new
errors.ErrUnsupported
, such thaterrors.Is(err, errors.ErrUnsupported)
returns true.ENOSYS
ENOTSUP
EOPNOTSUPP
EPLAN9
(Plan 9 only)ERROR_CALL_NOT_IMPLEMENTED
(Windows only)ERROR_NOT_SUPPORTED
(Windows only)EWINDOWS
(Windows only)
- testing
-
The new
-test.fullpath
option will print full path names in test log messages, rather than just base names.The new
Testing
function reports whether the program is a test created bygo
test
.
- testing/fstest
-
Calling
Open.Stat
will return a type that now implements aString
method that callsio/fs.FormatFileInfo
.
- unicode
-
The
unicode
package and associated support throughout the system has been upgraded to Unicode 15.0.0.
Ports
Darwin
As announced in the Go 1.20 release notes, Go 1.21 requires macOS 10.15 Catalina or later; support for previous versions has been discontinued.
Windows
As announced in the Go 1.20 release notes, Go 1.21 requires at least Windows 10 or Windows Server 2016; support for previous versions has been discontinued.
WebAssembly
The new go:wasmimport
directive can now be used in Go programs
to import functions from the WebAssembly host.
The Go scheduler now interacts much more efficiently with the JavaScript event loop, especially in applications that block frequently on asynchronous events.
WebAssembly System Interface
Go 1.21 adds an experimental port to the
WebAssembly System Interface (WASI), Preview 1
(GOOS=wasip1
, GOARCH=wasm
).
As a result of the addition of the new GOOS
value
"wasip1
", Go files named *_wasip1.go
will now be ignored
by Go tools except when that GOOS
value is being
used.
If you have existing filenames matching that pattern, you will
need to rename them.
ppc64/ppc64le
On Linux, GOPPC64=power10
now generates PC-relative instructions, prefixed
instructions, and other new Power10 instructions. On AIX, GOPPC64=power10
generates Power10 instructions, but does not generate PC-relative instructions.
When building position-independent binaries for GOPPC64=power10
GOOS=linux
GOARCH=ppc64le
, users can expect reduced binary
sizes in most cases, in some cases 3.5%. Position-independent binaries are built for
ppc64le with the following -buildmode
values:
c-archive
, c-shared
, shared
, pie
, plugin
.
loong64
The linux/loong64
port now supports -buildmode=c-archive
,
-buildmode=c-shared
and -buildmode=pie
.