Go 1.10 is not yet released. These are work-in-progress release notes. Go 1.10 is expected to be released in February 2018.
The latest Go release, version 1.10, arrives six months after go1.9. 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.
OVERVIEW HERE
There are no substantive changes to the language.
A corner case involving shifts by untyped constants has been clarified,
and as a result the compilers have been updated to allow the index expression
x[1.0 << s] where s is an untyped constant;
the go/types package already did.
The grammar for method expressions has been updated to relax the
syntax to allow any type expression as a receiver;
this matches what the compilers were already implementing.
For example, struct{io.Reader}.Read is a valid, if unusual,
method expression that the compilers already accepted and is
now permitted by the language grammar.
There are no new supported operating systems or processor architectures in this release. Most of the work has focused on strengthening the support for existing ports, in particular new instructions in the assembler and improvements to the code generated by the compilers.
TODO: default GOROOT changes in cmd/go TODO: computed GOROOT change
By default, the go tool creates its temporary files and directories
in the system temporary directory (for example, $TMPDIR on Unix).
If the new environment variable $GOTMPDIR is set,
the go tool will creates its temporary files and directories in that directory instead.
The go build command now detects out-of-date packages
purely based on the content of source files, specified build flags, and metadata stored in the compiled packages.
Modification times are no longer consulted or relevant.
The old advice to add -a to force a rebuild in cases where
the modification times were misleading for one reason or another
(for example, changes in build flags) is no longer necessary:
builds now always detect when packages must be rebuilt.
(If you observe otherwise, please file a bug.)
The go build -asmflags, -gcflags, -gccgoflags, and -ldflags options
now apply by default only to the packages listed directly on the command line.
For example, go build -gcflags=-m mypkg
passes the compiler the -m flag when building mypkg
but not its dependencies.
The new, more general form -asmflags=pattern=flags (and similarly for the others)
applies the flags only to the packages matching the pattern.
For example: go install -ldflags=cmd/gofmt=-X=main.version=1.2.3 cmd/...
installs all the commands matching cmd/... but only applies the -X option
to the linker flags for cmd/gofmt.
For more details, see go help build.
The go build command now maintains a cache of
recently built packages, separate from the installed packages in $GOROOT/pkg or $GOPATH/pkg.
The effect of the cache should be to speed builds that do not explicitly install packages
or when switching between different copies of source code (for example, when changing
back and forth between different branches in a version control system).
The old advice to add the -i flag for speed, as in go build -i
or go test -i,
is no longer necessary: builds run just as fast without -i.
For more details, see go help cache.
The go install command now installs only the
packages and commands listed directly on the command line.
For example, go install cmd/gofmt
installs the gofmt program but not any of the packages on which it depends.
The new build cache makes future commands still run as quickly as if the
dependencies had been installed.
To force the installation of dependencies, use the new
go install -i flag.
Installing dependencies should not be necessary in general,
and the very concept or installed packages may disappear in a future release.
Many details of the go build implementation have changed to support these improvements.
One new requirement implied by these changes is that
binary-only packages must now declare accurate import blocks in their
stub source code, so that those imports can be made available when
linking a program using the binary-only package.
For more details, see go help filetype.
The go test command now caches test results:
if the test executable and command line match a previous run
and the files and environment variables consulted by that run
have not changed either, go test will print
the previous test output, replacing the elapsed time with the string “(cached).”
Test caching applies only to successful test results;
only to go test
commands with an explicit list of packages; and
only to command lines using a subset of the
-cpu, -list, -parallel,
-run, -short, and -v test flags.
The idiomatic way to bypass test caching is to use -count=1.
The go test command now automatically runs
go vet on the package being tested,
to identify significant problems before running the test.
Any such problems are treated like build errors and prevent execution of the test.
Only a high-confidence subset of the available go vet
checks are enabled for this automatic check.
To disable the running of go vet, use
go test -vet=off.
The go test -coverpkg flag now
interprets its argument as a comma-separated list of patterns to match against
the dependencies of each test, not as a list of packages to load anew.
For example, go test -coverpkg=all
is now a meaningful way to run a test with coverage enabled for the test package
and all its dependencies.
Also, the go test -coverprofile option is now
supported when running multiple tests.
In case of failure due to timeout, tests are now more likely to write their profiles before exiting.
The go test command now always
merges the standard output and standard error from a given test binary execution
and writes both to go test's standard output.
In past releases, go test only applied this
merging most of the time.
The go test -v output
now includes PAUSE and CONT status update
lines to make clearer when parallel tests pause and continue.
Finally, the new go test -json flag
filters test output through the new command
go tool test2json
to produce a machine-readable JSON-formatted description of test execution.
This should allow the creation of rich presentations of test execution
in IDEs and other tools.
For more details about all these changes,
see go help test
and the test2json documentation.
Cgo now implements a C typedef like “typedef X Y;” using a Go type alias,
so that Go code may use the types C.X and C.Y interchangeably.
It also now supports the use of niladic function-like macros.
Also, the documentation has been updated to clarify that
Go structs and Go arrays are not supported in the type signatures of cgo-exported functions.
During toolchain bootstrap, the environment variables CC and CC_FOR_TARGET specify
the default C compiler that the resulting toolchain will use for host and target builds, respectively.
However, if the toolchain will be used with multiple targets, it may be necessary to specify a different C compiler for each
(for example, a different compiler for darwin/arm64 versus linux/ppc64le).
The new set of environment variables CC_FOR_goos_goarch
allows specifying a different default C compiler for each target.
Note that these variables only apply during toolchain bootstrap,
to set the defaults used by the resulting toolchain.
Later go build commands refer to the CC environment
variable or else the built-in default.
For more details, see the cgo documentation.
The go doc tool now adds functions returning slices of T or *T
to the display of type T, similar to the existing behavior for functions returning single T or *T results.
For example:
$ go doc mail.Address
package mail // import "net/mail"
type Address struct {
Name string
Address string
}
Address represents a single mail address.
func ParseAddress(address string) (*Address, error)
func ParseAddressList(list string) ([]*Address, error)
func (a *Address) String() string
$
Previously, ParseAddressList was only shown in the package overview (go doc mail).
The go fix tool now replaces imports of "golang.org/x/net/context"
with "context".
(Forwarding aliases in the former make it completely equivalent to the latter when using Go 1.9 or later.)
The go get command now supports Fossil source code repositories.
The blocking and mutex profiles produced by the runtime/pprof package
now include symbol information, so they can be viewed
in go tool pprof
without the binary that produced the profile.
(All other profile types were changed to include symbol information in Go 1.9.)
The go tool pprof profile visualizer has been updated to
the latest version from github.com/google/pprof.
The go vet command now always has access to
complete, up-to-date type information when checking packages, even for packages using cgo or vendored imports.
The reports should be more accurate as a result.
Note that only go vet has access to this information;
the more low-level go tool vet does not
and should be avoided except when working on vet itself.
(As of Go 1.9, go vet provides access to all the same flags as
go tool vet.)
This release includes a new overview of available Go program diagnostic tools.
A few minor details of the default formatting of Go source code have changed.
First, some complex three-index slice expressions previously formatted like
x[i+1 : j:k] and now
format with more consistent spacing: x[i+1 : j : k].
Second, single-method interface literals written on a single line,
which are sometimes used in type assertions,
are no longer split onto multiple lines.
Third, blank lines following an opening brace are now always elided.
Note that these kinds of minor updates to gofmt are expected from time to time. In general, we recommend against building systems that check that source code matches the output of a specific version of gofmt. For example, a continuous integration test that fails if any code already checked into a repository is not “properly formatted” is inherently fragile and not recommended.
If multiple programs must agree about which version of gofmt is used to format a source file,
we recommend that they do this by arranging to invoke the same gofmt binary.
For example, in the Go open source repository, we arrange for goimports and
our Git pre-commit hook to agree about source code formatting by having both
invoke the gofmt binary found in the current path.
TODO: Make goimports actually do that. #22695.
As another example, inside Google we arrange that source code presubmit
checks run a gofmt binary maintained at a fixed path in a shared, distributed file system;
that on engineering workstations /usr/bin/gofmt
is a symbolic link to that same path;
and that all editor integrations used for Google development
explicitly invoke /usr/bin/gofmt.
TODO: TMI?
The compiler includes many improvements to the performance of generated code, spread fairly evenly across the supported architectures.
TODO: What to say about DWARF work, if anything? Global constants (CL 61019), variable decomposition (CL 50878), variable liveness and location lists (CL 41770), more? What is enabled by default?
TODO: What to say about FMA, if anything? The spec change was mentioned in Go 1.9 but I am not sure whether any new architectures turned it on in Go 1.10.
The various build modes
has been ported to more systems.
Specifically, c-shared now works on linux/ppc64le, windows/386, and windows/amd64;
pie now works on darwin/amd64 and also forces the use of external linking on all systems;
and plugin now works on linux/ppc64le.
The linux/ppc64le port now requires the use of external linking
with any programs that use cgo, even uses by the standard library.
For the ARM 32-bit port, the assembler now supports the instructions
BFC,
BFI,
BFX,
BFXU,
FMULAD,
FMULAF,
FMULSD,
FMULSF,
FNMULAD,
FNMULAF,
FNMULSD,
FNMULSF,
MULAD,
MULAF,
MULSD,
MULSF,
NMULAD,
NMULAF,
NMULD,
NMULF,
NMULSD,
NMULSF,
XTAB,
XTABU,
XTAH,
and
XTAHU.
For the ARM 64-bit port, the assembler now supports the
VADD,
VADDP,
VADDV,
VAND,
VCMEQ,
VDUP,
VEOR,
VLD1,
VMOV,
VMOVI,
VMOVS,
VORR,
VREV32,
and
VST1
instructions.
For the PowerPC 64-bit port, the assembler now supports the POWER9 instructions
ADDEX,
CMPEQB,
COPY,
DARN,
LDMX,
MADDHD,
MADDHDU,
MADDLD,
MFVSRLD,
MTVSRDD,
MTVSRWS,
PASTECC,
VCMPNEZB,
VCMPNEZBCC,
and
VMSUMUDM.
For the S390X port, the assembler now supports the
TMHH,
TMHL,
TMLH,
and
TMLL
instructions.
For the X86 64-bit port, the assembler now supports 359 new instructions
and is believed to be complete up to and including the Intel AVX-256 extensions.
The assembler also no longer implements MOVL $0, AX
as an XORL instruction,
to avoid clearing the condition flags unexpectedly.
TODO: Words about GCC 8 and Go 1.10.
TODO: Don't start new threads from locked threads or threads that Go did not create. LockOSThread/UnlockOSThread now nest. LockOSThread + return kills the thread
Stack traces no longer include implicit wrapper functions (previously marked <autogenerated>),
unless a fault or panic happens in the wrapper itself.
There is no longer a limit on the GOMAXPROCS setting.
(In Go 1.9 the limit was 1024.)
As always, the changes are so general and varied that precise statements about performance are difficult to make. Most programs should run a bit faster, due to speedups in the garbage collector, better generated code, and optimizations in the core library.
TODO: Anything?
All of the changes to the standard library are minor. The changes in bytes and net/url are the most likely to require updating of existing programs.
As always, there are various minor changes and updates to the library, made with the Go 1 promise of compatibility in mind.
In general, the handling of special header formats is significantly improved and expanded.
FileInfoHeader has always
recorded the Unix UID and GID numbers from its os.FileInfo argument
(specifically, from the system-dependent information returned by the FileInfo's Sys method)
in the returned Header.
Now it also records the user and group names corresponding to those IDs,
as well as the major and minor device numbers for device files.
Errors created by the package now begin with a consistent “tar:” prefix. (Previously they almost all began with a consistent “archive/tar:” prefix.) TODO: Why are we changing these? (#22740)
The new Header.Format field
of type Format
controls which tar header format the Writer uses.
The default, as before, is to select the most widely-supported header type
that can encoding the fields needed by the header (USTAR if possible, or else PAX if possible, or else GNU).
The Reader sets Header.Format for each header it reads.
Reader and the Writer now support PAX records,
using the new Header.PAXRecords field.
The Reader no longer insists that the file name or link name in GNU headers
be valid UTF-8.
When writing PAX- or GNU-format headers, the Writer now includes
the Header.AccessTime and Header.ChangeTime fields (if set).
When writing PAX-format headers, the times include sub-second precision.
The Writer.Flush method,
which has had no real effect since Go 1.1, is now marked deprecated.
Go 1.10 adds more complete support for times and character set encodings in ZIP archives.
The original ZIP format used the standard MS-DOS encoding of year, month, day, hour, minute, and second into fields in two 16-bit values.
That encoding cannot represent time zones or odd seconds, so multiple extensions have been
introduced to allow richer encodings.
In Go 1.10, the Reader and Writer
now support the widely-understood Info-Zip extension that encodes the time separately in the 32-bit Unix “seconds since epoch” form.
The FileHeader's new Modified field of type time.Time
obsoletes the ModifiedTime and ModifiedDate fields, which continue to hold the MS-DOS encoding.
The ModTime and
SetModTime methods
now simply read and write the new Modified field.
The Reader and Writer now adopt the common
convention that ZIP archive storing the Unix time encoding store the local time
in the MS-DOS field, so that the time zone offset can be inferred.
TODO: These last bits are not true but probably should be (#22738)
The header for each file in a ZIP archive has a flag bit indicating whether
the name and comment fields are encoded as UTF-8, as opposed to a system-specific default encoding.
In Go 1.8 and earlier, the Writer never set the UTF-8 bit.
In Go 1.9, the Writer changed to set the UTF-8 bit almost always.
This broke the creation of ZIP archives containing Shift-JIS file names.
In Go 1.10, the Writer now sets the UTF-8 bit only when
both the name and the comment field are valid UTF-8 and at least one is non-ASCII.
Because non-ASCII encodings very rarely look like valid UTF-8, the new
heuristic should be correct nearly all the time.
Setting a FileHeader's new NonUTF8 field to true
disables the heuristic entirely for that file.
The Writer also now support setting the end-of-central-directory record's comment field,
by setting the Writer's new Comment field
before calling the Close method.
TODO: May change (#22737).
The new Reader.Size
and Writer.Size
methods report the Reader or Writer's underlying buffer size.
The
Fields,
FieldsFunc,
Split,
and
SplitAfter
each already returned slices pointing into the same underlying array as its input.
Go 1.10 changes each of the returned subslices to have capacity equal to its length,
so that appending to a subslice will not overwrite adjacent data in the original input.
NewOFB now panics if given
an initialization vector of incorrect length, like the other constructors in the
package always have.
(Previously it returned a nil Stream implementation.)
The TLS server now advertises support for SHA-512 signatures when using TLS 1.2. The server already supported the signatures, but some clients would not select them unless explicitly advertised.
Leaf certificate validation now enforces the name constraints for all names contained in the certificate, not just the one name that a client has asked about. Extended key usage restrictions are similarly now checked all at once. As a result, after a certificate has been validated, now it can be trusted in its entirety. It is no longer necessary to revalidate the certificate for each additional name or key usage. TODO: Link to docs that may not exist yet.
Parsed certificates also now report URI names and IP, email, and URI constraints, using the new
Certificate fields
URIs, PermittedIPRanges, ExcludedIPRanges,
PermittedEmailAddresses, ExcludedEmailAddresses,
PermittedURIDomains, and ExcludedURIDomains.
The new MarshalPKCS8PrivateKey
function converts a private key to PKCS#8 encoded form.
Name now implements a
String method that
formats the X.509 distinguished name in the standard RFC 2253 format.
Drivers that want to construct a sql.DB for
their clients can now implement the Connector interface
and call the new sql.OpenDB function,
instead of needing to encode all configuration into a string
passed to sql.Open.
Drivers that implement ExecerContext
no longer need to implement Execer;
similarly, drivers that implement QueryerContext
no longer need to implement Queryer.
Previously, even if the context-based interfaces were implemented they were ignored
unless the non-context-based interfaces were also implemented.
To allow drivers to better isolate different clients using a cached driver connection in succession,
if a Conn implements the new
SessionResetter interface,
database/sql will now call ResetSession before
reusing the Conn for a new client.
This release adds 348 new relocation constants divided between the relocation types
R_386,
R_AARCH64,
R_ARM,
R_PPC64,
and
R_X86_64.
Go 1.10 adds support for reading relocations from Mach-O sections,
using the Section struct's new Relocs field
and the newReloc,
RelocTypeARM,
RelocTypeARM64,
RelocTypeGeneric,
and
RelocTypeX86_64
types and associated constants.
Go 1.10 also adds support for the LC_RPATH load command,
represented by the types
RpathCmd and
Rpath,
and new named constants
for the various flag bits found in headers.
Marshal now correctly encodes
strings containing asterisks as type UTF8String instead of PrintableString,
unless the string is in a struct field with a tag forcing the use of PrintableString.
Marshal also now respects struct tags containing application directives.
Unmarshal now respects
struct field tags using the explicit and tag
directives.
Reader now disallows the use of
nonsensical Comma and Comment settings,
such as NUL, carriage return, newline, invalid runes, and the Unicode replacement character,
or setting Comma and Comment equal to each other.
In the case of a syntax error in a CSV record that spans multiple input lines, Reader
now reports the line on which the record started in the ParseError's new StartLine field.
Reader also no longer strips carriage return characters
appearing before newline characters in multiline quoted strings.
TODO: Maybe not (#22746).
The new functions
NewEncoder
and
NewDecoder
provide streaming conversions to and from hexadecimal,
analogous to equivalent functions already in
encoding/base32
and
encoding/base64.
When the functions
Decode
and
DecodeString
encounter malformed input,
they each now return the number of bytes already converted
along with the error.
Previously they always returned a count of 0 with any error.
The Decoder
adds a new method
DisallowUnknownFields
that causes it to report inputs with unknown JSON fields as a decoding error.
(The default behavior has always been to discard unknown fields.)
The new function
NewTokenDecoder
is like
NewDecoder
but creates a decoder reading from a TokenReader
instead of an XML-formatted byte stream.
This is meant to enable the construction of XML stream transformers in client libraries.
The default
Usage function now prints
its first line of output to
CommandLine.Output()
instead of assuming os.Stderr,
so that the usage message is properly redirected for
clients using CommandLine.SetOutput.
PrintDefaults now
adds appropriate indentation after newlines in flag usage strings,
so that multi-line usage strings display nicely.
FlagSet adds new methods
ErrorHandling,
Name,
and
Output,
to retrieve the settings passed to
NewFlagSet
and
FlagSet.SetOutput.
To support the doc change described above,
functions returning slices of T, *T, **T, and so on
are now reported in T's Type's Funcs list,
instead of in the Package's Funcs list.
The For function now accepts a non-nil lookup argument.
The changes to the default formatting of Go source code discussed in the gofmt section above are implemented in the go/printer package and also affect the output of the higher-level go/format package.
Implementations of the Hash interface are now
encouraged to implement encoding.BinaryMarshaler
and encoding.BinaryUnmarshaler
to allow saving and recreating their internal state,
and all implementations in the standard library
(hash/crc32, crypto/sha256, and so on)
now implement those interfaces.
The new actions {{"{{break}}"}} and {{"{{continue}}"}}
break out of the innermost {{"{{range"}} ...}} loop,
like the corresponding Go statements.
TODO: something about the AddParseTree problem (#21844).
Int now supports conversions to and from bases 2 through 62
in its SetString and Text methods.
(Previously it only allowed bases 2 through 36.)
The value of the constant MaxBase has been updated.
Int adds a new
CmpAbs method
that is like Cmp but
compares only the absolute values (not the signs) of its arguments.
The new function and corresponding
Rand.Shuffle method
shuffle an input sequence.
The existing function and corresponding
Rand.Perm method
have been updated to use a more efficient algorithm, with the result
that the specific permutations they return have changed.
TODO: Remove? (#22744)
The new functions
Round
and
RoundToEven
round their arguments to the nearest integer;
Round rounds a half-integer to its larger integer neighbor (away from zero)
while RoundToEven rounds a half-integer its even integer neighbor.
The new functions
Erfinv
and
Erfcinv
compute the inverse error function and the
inverse complementary error function.
Reader
now accepts parts with empty filename attributes.
ParseMediaType now discards
invalid attribute values; previously it returned those values as empty strings.
The Conn and
Listener implementations
in this package now guarantee that when Close returns,
the underlying file descriptor has been closed.
(In earlier releases, if the Close stopped pending I/O
in other goroutines, the closing of the file descriptor could happen in one of those
goroutines shortly after Close returned.)
TCPListener and
UnixListener
now implement
syscall.Conn,
to allow setting options on the underlying file descriptor
using syscall.RawConn.Control.
The Conn implementations returned by Pipe
now support setting read and write deadlines.
The IPConn.ReadMsgIP,
IPConn.WriteMsgIP,
UDPConn.ReadMsgUDP,
and
UDPConn.WriteMsgUDP,
methods are now implemented on Windows.
On the client side, an HTTP proxy (most commonly configured by
ProxyFromEnvironment)
can now be specified as an https:// URL,
meaning that the client connects to the proxy over HTTPS before issuing a standard, proxied HTTP request.
(Previously, HTTP proxy URLs were required to begin with http:// or socks5://.)
On the server side, FileServer and its single-file equivalent ServeFile
now apply If-Range checks to HEAD requests.
FileServer also now reports directory read failures to the Server's ErrorLog.
Redirect now sets the Content-Type header before writing its HTTP response.
ParseAddress and
ParseAddressList and
now support a variety of obsolete address formats.
The Client adds a new
Noop method,
to test whether the server is still responding.
It also now defends against possible SMTP injection in the inputs
to the Hello
and Verify methods.
ReadMIMEHeader
now discards continuation (indented) header lines that appear before the first actual (unindented) header line.
ResolveReference
now preseves multiple leading slashes in the target URL.
Previously it rewrote multiple leading slashes to a single slash,
which resulted in the http.Client
following certain redirects incorrectly.
For example, this code's output has changed:
base, _ := url.Parse("http://host//path//to/page1")
target, _ := url.Parse("page2")
fmt.Println(base.ResolveReference(target))
Note the doubled slashes around path.
In Go 1.9 and earlier, the resolved URL was http://host/path//to/page2:
the doubled slash before path was incorrectly rewritten
to a single slash, while the doubled slash after path was
correctly preserved.
Go 1.10 preserves both doubled slashes, resolving to http://host//path//to/page2
as required by RFC 3986.
This change may break existing buggy programs that unintentionally
construct a base URL with a leading doubled slash in the path and inadvertently
depend on ResolveReference to correct that mistake.
For example, this can happen if code adds a host prefix
like http://host/ to a path like /my/api,
resulting in a URL with a doubled slash: http://host//my/api.
File adds new methods
SetDeadline,
SetReadDeadline,
and
SetWriteDeadline
that allow setting I/O deadlines when the
underlying file descriptor supports non-blocking I/O operations.
The definition of these methods matches those in net.Conn.
Also matching net.Conn,
File's
Close method
now guarantee that when Close returns,
the underlying file descriptor has been closed.
(In earlier releases, like for net.Conn's,
if the Close stopped pending I/O
in other goroutines, the closing of the file descriptor could happen in one of those
goroutines shortly after Close returned.)
On BSD, macOS, and Solaris systems,
Chtimes
now supports setting file times with nanosecond precision
(assuming the underlying file system can represent them).
The Copy function now allows copying
from a string into a byte array or byte slice, to match the
built-in copy function.
As noted above, the blocking and mutex profiles now include symbol information so that they can be viewed without needing the binary that generated them.
ParseUint now returns
the maximum magnitude integer of the appropriate size
with any ErrRange error, as it was already documented to do.
Previously it returned 0 with ErrRange errors.
A new type
Builder is a replacement for
bytes.Buffer for the use case of
accumulating text into a string result.
The Builder's API is a restricted subset of bytes.Buffer's
that allows it to safely avoid making a duplicate copy of the data
during the String method.
On Windows,
the new SysProcAttr field Token,
of type Token allows the creation of a process that
runs as another user during StartProcess
(and therefore also during os.StartProcess and
exec.Cmd.Start).
The new function CreateProcessAsUser
gives access to the underlying system call.
On BSD, macOS, and Solaris systems, UtimesNano
is now implemented.
The new actions {{"{{break}}"}} and {{"{{continue}}"}}
break out of the innermost {{"{{range"}} ...}} loop,
like the corresponding Go statements.
LoadLocation now uses the directory
or uncompressed zip file named by the $ZONEINFO
environment variable before looking in the default system-specific list of
known installation locations or in $GOROOT/lib/time/zoneinfo.zip.
TODO: Maybe CL 68890.
The unicode package and associated
support throughout the system has been upgraded from version 9.0 to
Unicode 10.0,
which adds 8,518 new characters, including four new scripts, one new property,
a Bitcoin currency symbol, and 56 new emoji.