The latest Go release, version 1.14, arrives six months after Go 1.13. 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. We expect almost all Go programs to continue to compile and run as before.
Module support in the go
command is now ready for production use,
and we encourage all users to migrate to Go
modules for dependency management. If you are unable to migrate due to a problem in the Go
toolchain, please ensure that the problem has an
open issue
filed. (If the issue is not on the Go1.15
milestone, please let us
know why it prevents you from migrating so that we can prioritize it
appropriately.)
Per the overlapping interfaces proposal, Go 1.14 now permits embedding of interfaces with overlapping method sets: methods from an embedded interface may have the same names and identical signatures as methods already present in the (embedding) interface. This solves problems that typically (but not exclusively) occur with diamond-shaped embedding graphs. Explicitly declared methods in an interface must remain unique, as before.
Go 1.14 is the last release that will run on macOS 10.11 El Capitan. Go 1.15 will require macOS 10.12 Sierra or later.
Go 1.14 is the last Go release to support 32-bit binaries on
macOS (the darwin/386
port). They are no longer
supported by macOS, starting with macOS 10.15 (Catalina).
Go continues to support the 64-bit darwin/amd64
port.
Go 1.14 will likely be the last Go release to support 32-bit
binaries on iOS, iPadOS, watchOS, and tvOS
(the darwin/arm
port). Go continues to support the
64-bit darwin/arm64
port.
Go binaries on Windows now have DEP (Data Execution Prevention) enabled.
JavaScript values referenced from Go via js.Value
objects can now be garbage collected.
js.Value
values can no longer be compared using
the ==
operator, and instead must be compared using
their Equal
method.
js.Value
now
has IsUndefined
, IsNull
,
and IsNaN
methods.
Go 1.14 contains experimental support for 64-bit RISC-V on Linux
(GOOS=linux
, GOARCH=riscv64
). Be aware
that performance, assembly syntax stability, and possibly
correctness are a work in progress.
Go now supports the 64-bit ARM architecture on FreeBSD 12.0 or later (the
freebsd/arm64
port).
As announced in the Go 1.13 release notes,
Go 1.14 drops support for the Native Client platform (GOOS=nacl
).
The runtime now respects zone CPU caps
(the zone.cpu-cap
resource control)
for runtime.NumCPU
and the default value
of GOMAXPROCS
.
When the main module contains a top-level vendor
directory and
its go.mod
file specifies go
1.14
or
higher, the go
command now defaults to -mod=vendor
for operations that accept that flag. A new value for that flag,
-mod=mod
, causes the go
command to instead load
modules from the module cache (as when no vendor
directory is
present).
When -mod=vendor
is set (explicitly or by default), the
go
command now verifies that the main module's
vendor/modules.txt
file is consistent with its
go.mod
file.
go
list
-m
no longer silently omits
transitive dependencies that do not provide packages in
the vendor
directory. It now fails explicitly if
-mod=vendor
is set and information is requested for a module not
mentioned in vendor/modules.txt
.
The go
get
command no longer accepts
the -mod
flag. Previously, the flag's setting either
was ignored or
caused the build to fail.
-mod=readonly
is now set by default when the go.mod
file is read-only and no top-level vendor
directory is present.
-modcacherw
is a new flag that instructs the go
command to leave newly-created directories in the module cache at their
default permissions rather than making them read-only.
The use of this flag makes it more likely that tests or other tools will
accidentally add files not included in the module's verified checksum.
However, it allows the use of rm
-rf
(instead of go
clean
-modcache
)
to remove the module cache.
-modfile=file
is a new flag that instructs the go
command to read (and possibly write) an alternate go.mod
file
instead of the one in the module root directory. A file
named go.mod
must still be present in order to determine the
module root directory, but it is not accessed. When -modfile
is
specified, an alternate go.sum
file is also used: its path is
derived from the -modfile
flag by trimming the .mod
extension and appending .sum
.
GOINSECURE
is a new environment variable that instructs
the go
command to not require an HTTPS connection, and to skip
certificate validation, when fetching certain modules directly from their
origins. Like the existing GOPRIVATE
variable, the value
of GOINSECURE
is a comma-separated list of glob patterns.
When module-aware mode is enabled explicitly (by setting
GO111MODULE=on
), most module commands have more
limited functionality if no go.mod
file is present. For
example, go
build
,
go
run
, and other build commands can only build
packages in the standard library and packages specified as .go
files on the command line.
Previously, the go
command would resolve each package path
to the latest version of a module but would not record the module path
or version. This resulted in slow,
non-reproducible builds.
go
get
continues to work as before, as do
go
mod
download
and
go
list
-m
with explicit versions.
+incompatible
versions
If the latest version of a module contains a go.mod
file,
go
get
will no longer upgrade to an
incompatible
major version of that module unless such a version is requested explicitly
or is already required.
go
list
also omits incompatible major versions
for such a module when fetching directly from version control, but may
include them if reported by a proxy.
go.mod
file maintenance
go
commands other than
go
mod
tidy
no longer
remove a require
directive that specifies a version of an indirect dependency
that is already implied by other (transitive) dependencies of the main
module.
go
commands other than
go
mod
tidy
no longer
edit the go.mod
file if the changes are only cosmetic.
When -mod=readonly
is set, go
commands will no
longer fail due to a missing go
directive or an erroneous
// indirect
comment.
The go
command now supports Subversion repositories in module mode.
The go
command now includes snippets of plain-text error messages
from module proxies and other HTTP servers.
An error message will only be shown if it is valid UTF-8 and consists of only
graphic characters and spaces.
go test -v
now streams t.Log
output as it happens,
rather than at the end of all tests.
This release improves the performance of most uses
of defer
to incur almost zero overhead compared to
calling the deferred function directly.
As a result, defer
can now be used in
performance-critical code without overhead concerns.
Goroutines are now asynchronously preemptible.
As a result, loops without function calls no longer potentially
deadlock the scheduler or significantly delay garbage collection.
This is supported on all platforms except windows/arm
,
darwin/arm
, js/wasm
, and
plan9/*
.
A consequence of the implementation of preemption is that on Unix
systems, including Linux and macOS systems, programs built with Go
1.14 will receive more signals than programs built with earlier
releases.
This means that programs that use packages
like syscall
or golang.org/x/sys/unix
will see more slow system calls fail with EINTR
errors.
Those programs will have to handle those errors in some way, most
likely looping to try the system call again. For more
information about this
see man
7 signal
for Linux systems or similar documentation for
other systems.
The page allocator is more efficient and incurs significantly less
lock contention at high values of GOMAXPROCS
.
This is most noticeable as lower latency and higher throughput for
large allocations being done in parallel and at a high rate.
Internal timers, used by
time.After
,
time.Tick
,
net.Conn.SetDeadline
,
and friends, are more efficient, with less lock contention and fewer
context switches.
This is a performance improvement that should not cause any user
visible changes.
This release adds -d=checkptr
as a compile-time option
for adding instrumentation to check that Go code is following
unsafe.Pointer
safety rules dynamically.
This option is enabled by default (except on Windows) with
the -race
or -msan
flags, and can be
disabled with -gcflags=all=-d=checkptr=0
.
Specifically, -d=checkptr
checks the following:
unsafe.Pointer
to *T
,
the resulting pointer must be aligned appropriately
for T
.
unsafe.Pointer
-typed operands must point
into the same object.
Using -d=checkptr
is not currently recommended on
Windows because it causes false alerts in the standard library.
The compiler can now emit machine-readable logs of key optimizations
using the -json
flag, including inlining, escape
analysis, bounds-check elimination, and nil-check elimination.
Detailed escape analysis diagnostics (-m=2
) now work again.
This had been dropped from the new escape analysis implementation in
the previous release.
All Go symbols in macOS binaries now begin with an underscore, following platform conventions.
This release includes experimental support for compiler-inserted coverage instrumentation for fuzzing. See issue 14565 for more details. This API may change in future releases.
Bounds check elimination now uses information from slice creation and can
eliminate checks for indexes with types smaller than int
.
Go 1.14 includes a new package,
hash/maphash
,
which provides hash functions on byte sequences.
These hash functions are intended to be used to implement hash tables or
other data structures that need to map arbitrary strings or byte
sequences to a uniform distribution on unsigned 64-bit integers.
The hash functions are collision-resistant but not cryptographically secure.
The hash value of a given byte sequence is consistent within a single process, but will be different in different processes.
As always, there are various minor changes and updates to the library, made with the Go 1 promise of compatibility in mind.
Support for SSL version 3.0 (SSLv3) has been removed. Note that SSLv3 is the cryptographically broken protocol predating TLS.
TLS 1.3 can't be disabled via the GODEBUG
environment
variable anymore. Use the
Config.MaxVersion
field to configure TLS versions.
When multiple certificate chains are provided through the
Config.Certificates
field, the first one compatible with the peer is now automatically
selected. This allows for example providing an ECDSA and an RSA
certificate, and letting the package automatically select the best one.
Note that the performance of this selection is going to be poor unless the
Certificate.Leaf
field is set.
The new CipherSuites
and InsecureCipherSuites
functions return a list of currently implemented cipher suites.
The new CipherSuiteName
function returns a name for a cipher suite ID.
The new
(*ClientHelloInfo).SupportsCertificate
and
(*CertificateRequestInfo).SupportsCertificate
methods expose whether a peer supports a certain certificate.
The tls
package no longer supports the legacy Next Protocol
Negotiation (NPN) extension and now only supports ALPN. In previous
releases it supported both. There are no API changes and applications
should function identically as before. Most other clients and servers have
already removed NPN support in favor of the standardized ALPN.
RSA-PSS signatures are now used when supported in TLS 1.2 handshakes. This
won't affect most applications, but custom
Certificate.PrivateKey
implementations that don't support RSA-PSS signatures will need to use the new
Certificate.SupportedSignatureAlgorithms
field to disable them.
Config.Certificates
and
Config.GetCertificate
can now both be nil if
Config.GetConfigForClient
is set. If the callbacks return neither certificates nor an error, the
unrecognized_name
is now sent.
The new CertificateRequestInfo.Version
field provides the TLS version to client certificates callbacks.
The new TLS_ECDHE_RSA_WITH_CHACHA20_POLY1305_SHA256
and
TLS_ECDHE_ECDSA_WITH_CHACHA20_POLY1305_SHA256
constants use
the final names for the cipher suites previously referred to as
TLS_ECDHE_RSA_WITH_CHACHA20_POLY1305
and
TLS_ECDHE_ECDSA_WITH_CHACHA20_POLY1305
.
Certificate.CreateCRL
now supports Ed25519 issuers.
The debug/dwarf
package now supports reading DWARF
version 5.
The new
method (*Data).AddSection
supports adding arbitrary new DWARF sections from the input file
to the DWARF Data
.
The new
method (*Reader).ByteOrder
returns the byte order of the current compilation unit.
This may be used to interpret attributes that are encoded in the
native ordering, such as location descriptions.
The new
method (*LineReader).Files
returns the file name table from a line reader.
This may be used to interpret the value of DWARF attributes such
as AttrDeclFile
.
Unmarshal
now supports ASN.1 string type BMPString, represented by the new
TagBMPString
constant.
The Decoder
type supports a new
method InputOffset
that returns the input stream byte offset of the current
decoder position.
Compact
no longer
escapes the U+2028
and U+2029
characters, which
was never a documented feature. For proper escaping, see HTMLEscape
.
Number
no longer
accepts invalid numbers, to follow the documented behavior more closely.
If a program needs to accept invalid numbers like the empty string,
consider wrapping the type with Unmarshaler
.
The Context
type has a new field Dir
which may be used to set
the working directory for the build.
The default is the current directory of the running process.
In module mode, this is used to locate the main module.
The new
function NewFromFiles
computes package documentation from a list
of *ast.File
's and associates examples with the
appropriate package elements.
The new information is available in a new Examples
field
in the Package
, Type
,
and Func
types, and a
new Suffix
field in
the Example
type.
TempDir
can now create directories
whose names have predictable prefixes and suffixes.
As with TempFile
, if the pattern
contains a '*', the random string replaces the last '*'.
The
new Lmsgprefix
flag may be used to tell the logging functions to emit the
optional output prefix immediately before the log message rather
than at the start of the line.
The new FMA
function
computes x*y+z
in floating point with no
intermediate rounding of the x*y
computation. Several architectures implement this computation
using dedicated hardware instructions for additional performance.
The GCD
method
now allows the inputs a
and b
to be
zero or negative.
The new functions
Rem
,
Rem32
, and
Rem64
support computing a remainder even when the quotient overflows.
The default type of .js
and .mjs
files
is now text/javascript
rather
than application/javascript
.
This is in accordance
with an
IETF draft that treats application/javascript
as obsolete.
The
new Reader
method NextRawPart
supports fetching the next MIME part without transparently
decoding quoted-printable
data.
The new Header
method Values
can be used to fetch all values associated with a
canonicalized key.
The
new Transport
field DialTLSContext
can be used to specify an optional dial function for creating
TLS connections for non-proxied HTTPS requests.
This new field can be used instead
of DialTLS
,
which is now considered deprecated; DialTLS
will
continue to work, but new code should
use DialTLSContext
, which allows the transport to
cancel dials as soon as they are no longer needed.
On Windows, ServeFile
now correctly
serves files larger than 2GB.
The
new Server
field EnableHTTP2
supports enabling HTTP/2 on the test server.
The
new MIMEHeader
method Values
can be used to fetch all values associated with a canonicalized
key.
On Windows,
the CTRL_CLOSE_EVENT
, CTRL_LOGOFF_EVENT
,
and CTRL_SHUTDOWN_EVENT
events now generate
a syscall.SIGTERM
signal, similar to how Control-C
and Control-Break generate a syscall.SIGINT
signal.
The plugin
package now supports freebsd/amd64
.
StructOf
now
supports creating struct types with unexported fields, by
setting the PkgPath
field in
a StructField
element.
runtime.Goexit
can no longer be aborted by a
recursive panic
/recover
.
On macOS, SIGPIPE
is no longer forwarded to signal
handlers installed before the Go runtime is initialized.
This is necessary because macOS delivers SIGPIPE
to the main thread
rather than the thread writing to the closed pipe.
The generated profile no longer includes the pseudo-PCs used for inline marks. Symbol information of inlined functions is encoded in the format the pprof tool expects. This is a fix for the regression introduced during recent releases.
The NumError
type now has
an Unwrap
method that may be used to retrieve the reason that a conversion
failed.
This supports using NumError
values
with errors.Is
to see
if the underlying error
is strconv.ErrRange
or strconv.ErrSyntax
.
Unlocking a highly contended Mutex
now directly
yields the CPU to the next goroutine waiting for
that Mutex
. This significantly improves the
performance of highly contended mutexes on high CPU count
machines.
The testing package now supports cleanup functions, called after
a test or benchmark has finished, by calling
T.Cleanup
or
B.Cleanup
respectively.
The text/template package now correctly reports errors when a
parenthesized argument is used as a function.
This most commonly shows up in erroneous cases like
{{if (eq .F "a") or (eq .F "b")}}
.
This should be written as {{if or (eq .F "a") (eq .F "b")}}
.
The erroneous case never worked as expected, and will now be
reported with an error can't give argument to non-function
.
The unicode
package and associated
support throughout the system has been upgraded from Unicode 11.0 to
Unicode 12.0,
which adds 554 new characters, including four new scripts, and 61 new emoji.