The latest Go release, version 1.4, arrives as scheduled six months after 1.3
and contains only one tiny language change,
a possibly breaking change to the compiler,
a backwards-compatible simple form of for-range loop.
The release focuses primarily on implementation work, improving the garbage collector
and preparing the ground for a fully concurrent collector to be rolled out in the
next few releases.
Stacks are now contiguous, reallocated when necessary rather than linking on new
"segments";
this release therefore eliminates the notorious "hot stack split" problem.
There are some new tools available including support in the go command
for build-time source code generation.
The release also adds support for ARM processors on Android and Native Client (NaCl)
and AMD64 on Plan 9.
As always, Go 1.4 keeps the promise
of compatibility,
and almost everything
will continue to compile and run without change when moved to 1.4.
Up until Go 1.3, for-range loop had two forms
for k, v := range x {
...
}
and
for k := range x {
...
}
If one was not interested in the loop values, only the iteration itself, it was still
necessary to mention a variable (probably the blank identifier, as in
for _ = range x), because
the form
for range x {
...
}
was not syntactically permitted.
This situation seemed awkward, so as of Go 1.4 the variable-free form is now legal. The pattern arises rarely but the code can be cleaner when it does.
Updating: The change is strictly backwards compatible to existing Go
programs, but tools that analyze Go parse trees may need to be modified to accept
this new form as the
Key field of RangeStmt
may now be nil.
Given these declarations,
type T int
func (T) M() {}
var x **T
both gc and gccgo accepted the method call
x.M()
which is a double dereference of the pointer-to-pointer x.
The Go specification allows a single dereference to be inserted automatically,
but not two, so this call is erroneous according to the language definition.
It has therefore been disallowed in Go 1.4, which is a breaking change,
although very few programs will be affected.
Updating: Code that depends on the old, erroneous behavior will no longer compile but is easy to fix by adding an explicit dereference.
Go 1.4 can build binaries for ARM processors running the Android operating system.
It can also build a .so library that can be loaded by an Android application
using the supporting packages in the go.mobile repository.
A brief description of the plans for this experimental port are available
here.
The previous release introduced Native Client (NaCl) support for the 32-bit x86
(GOARCH=386)
and 64-bit x86 using 32-bit pointers (GOARCH=amd64p32).
The 1.4 release adds NaCl support for ARM (GOARCH=arm).
This release adds support for the Plan 9 operating system on AMD64 processors,
provided the kernel supports the nsec system call and uses 4K pages.
The unsafe package allows one
to defeat Go's type system by exploiting internal details of the implementation
or machine representation of data.
It was never explicitly specified what use of unsafe meant
with respect to compatibility as specified in the
Go compatibility guidelines.
The answer, of course, is that we can make no promise of compatibility
for code that does unsafe things.
We have clarified this situation in the documentation included in the release.
The Go compatibility guidelines and the
docs for the unsafe package
are now explicit that unsafe code is not guaranteed to remain compatible.
Updating: Nothing technical has changed; this is just a clarification of the documentation.
Up to Go 1.4, the runtime (garbage collector, concurrency support, interface management, maps, slices, strings, ...) was mostly written in C, with some assembler support. In 1.4, much of the code has been translated to Go so that the garbage collector can scan the stacks of programs in the runtime and get accurate information about what variables are active. This change was large but should have no semantic effect on programs.
This rewrite allows the garbage collector in 1.4 to be fully precise, meaning that it is aware of the location of all active pointers in the program. This means the heap will be smaller as there will be no false positives keeping non-pointers alive. Other related changes also reduce the heap size, which is smaller by 10%-30% overall relative to the previous release.
A consequence is that stacks are no longer segmented, eliminating the "hot split" problem. When a stack limit is reached, a new, larger stack is allocated, all active frames for the goroutine are copied there, and any pointers into the stack are updated. Performance can be noticeably better in some cases and is always more predictable. Details are available in the design document.
The use of contiguous stacks means that stacks can start smaller without triggering performance issues, so the default starting size for a goroutine's stack in 1.4 has been reduced to 2048 bytes from 8192 bytes. TODO: It may be bumped to 4096 for the release.
As preparation for the concurrent garbage collector scheduled for the 1.5 release, writes to pointer values in the heap are now done by a function call, called a write barrier, rather than directly from the function updating the value. In this next release, this will permit the garbage collector to mediate writes to the heap while it is running. This change has no semantic effect on programs in 1.4, but was included in the release to test the compiler and the resulting performance.
The implementation of interface values has been modified. In earlier releases, the interface contained a word that was either a pointer or a one-word scalar value, depending on the type of the concrete object stored. This implementation was problematical for the garbage collector, so as of 1.4 interface values always hold a pointer. In running programs, most interface values were pointers anyway, so the effect is minimal, but programs that store integers (for example) in interfaces will see more allocations.
As of Go 1.3, the runtime crashes if it finds a memory word that should contain
a valid pointer but instead contains an obviously invalid pointer (for example, the value 3).
Programs that store integers in pointer values may run afoul of this check and crash.
In Go 1.4, setting the GODEBUG variable
invalidptr=0 disables
the crash as a workaround, but we cannot guarantee that future releases will be
able to avoid the crash; the correct fix is to rewrite code not to alias integers and pointers.
The language accepted by the assemblers cmd/5a, cmd/6a
and cmd/8a has had several changes,
mostly to make it easier to deliver type information to the runtime.
First, the textflag.h file that defines flags for TEXT directives
has been copied from the linker source directory to a standard location so it can be
included with the simple directive
#include "textflag.h"
The more important changes are in how assembler source can define the necessary
type information.
For most programs it will suffice to move data
definitions (DATA and GLOBL directives)
out of assembly into Go files
and to write a Go declaration for each assembly function.
The assembly document describes what to do.
Updating:
Assembly files that include textflag.h from its old
location will still work, but should be updated.
For the type information, most assembly routines will need no change,
but all should be examined.
Assembly source files that define data,
functions with non-empty stack frames, or functions that return pointers
need particular attention.
A description of the necessary (but simple) changes
is in the assembly document.
More information about these changes is in the assembly document.
The release schedules for the GCC and Go projects do not coincide. GCC release 4.9 contains the Go 1.2 version of gccgo. The next release, GCC 5, will likely have the Go 1.4 version of gccgo.
Go's package system makes it easy to structure programs into components with clean boundaries, but there are only two forms of access: local (unexported) and global (exported). Sometimes one wishes to have components that are not exported, for instance to avoid acquiring clients of interfaces to code that is part of a public repository but not intended for use outside the program to which it belongs.
The Go language does not have the power to enforce this distinction, but as of Go 1.4 the
go command introduces
a mechanism to define "internal" packages that may not be imported by packages outside
the source subtree in which they reside.
To create such a package, place it in a directory named internal or in a subdirectory of a directory
named internal.
When the go command sees an import of a package with internal in its path,
it verifies that the package doing the import
is within the tree rooted at the parent of the internal directory.
For example, a package .../a/b/c/internal/d/e/f
can be imported only by code in the directory tree rooted at .../a/b/c.
It cannot be imported by code in .../a/b/g or in any other repository.
For Go 1.4, the internal package mechanism is enforced for the main Go repository; from 1.5 and onward it will be enforced for any repository.
Full details of the mechanism are in the design document.
Code often lives in repositories hosted by public services such as github.com,
meaning that the import paths for packages begin with the name of the hosting service,
github.com/rsc/pdf for example.
One can use
an existing mechanism
to provide a "custom" or "vanity" import path such as
rsc.io/pdf, but
that creates two valid import paths for the package.
That is a problem: one may inadvertently import the package through the two
distinct paths in a single program, which is wasteful;
miss an update to a package because the path being used is not recognized to be
out of date;
or break clients using the old path by moving the package to a different hosting service.
Go 1.4 introduces an annotation for package clauses in Go source that identify a canonical
import path for the package.
If an import is attempted using a path that is not canonical,
the go command
will refuse to compile the importing package.
The syntax is simple: put an identifying comment on the package line. For our example, the package clause would read:
package pdf // import "rsc.io/pdf"
With this in place,
the go command will
refuse to compile a package that imports github.com/rsc/pdf,
ensuring that the code can be moved without breaking users.
The check is at build time, not download time, so if go get
fails because of this check, the mis-imported package has been copied to the local machine
and should be removed manually.
Further information is in the design document.
The go command has a new subcommand,
go generate,
to automate the running of tools to generate source code before compilation.
For example, it can be used to run the yacc
compiler-compiler on a .y file to produce the Go source file implementing the grammar,
or to automate the generation of String methods for typed constants using the new
stringer
tool in the go.tools repository.
For more information, see the design document.
Build constraints, also known as build tags, control compilation by including or excluding files
(see the documentation /go/build).
Compilation can also be controlled by the name of the file itself by "tagging" the file with
a suffix (before the .go or .s extension) with an underscore
and the name of the architecture or operating system.
For instance, the file gopher_arm.go will only be compiled if the target
processor is an ARM.
Before Go 1.4, a file called just arm.go was similarly tagged, but this behavior
can break sources when new architectures are added, causing files to suddenly become tagged.
In 1.4, therefore, a file will be tagged in this manner only if the tag (architecture or operating
system name) is preceded by an underscore.
Updating: Packages that depend on the old behavior will no longer compile correctly.
Files with names like windows.go or amd64.go should either
have explicit build tags added to the source or be renamed to something like
os_windows.go or support_amd64.go.
There were a number of minor changes to the
cmd/go
command worth noting.
cgo is being used to build the package,
the go command now refuses to compile C source files,
since the relevant C compilers
(6c etc.)
are intended to be removed from the installation in some future release.
(They are used today only to build part of the runtime.)
It is difficult to use them correctly in any case, so any extant uses are likely incorrect,
so we have disabled them.
go test
subcommand has a new flag, -o, to set the name of the resulting binary,
corresponding to the same flag in other subcommands.
The non-functional -file flag has been removed.
go test
subcommand will compile and link all *_test.go files in the package,
even when there are no Test functions in them.
It previously ignored such files.
go build
subcommand's
-a flag has been changed for non-development installations.
For installations running a released distribution, the -a flag will no longer
rebuild the standard library and commands, to avoid overwriting the installation's files.
TODO godoc news
In the main Go source repository, the source code for the packages was kept in
the directory src/pkg, which made sense but differed from
other repositories, including the Go sub-repositories such as go.tools.
In Go 1.4, the pkg level of the source tree is now gone, so for example
the fmt package's source, once kept in
directory src/pkg/fmt, now lives one level higher in src/fmt.
Updating: Tools like godoc that discover source code
need to know about the new location. All tools and services maintained by the Go team
have been updated.
Due to the runtime changes in this release, Go 1.4 will require SWIG 3.0.3. At time of writing that has not yet been released, but we expect it to be by Go 1.4's release date. TODO
The standard repository's top-level misc directory used to contain
Go support for editors and IDEs: plugins, initialization scripts and so on.
Maintaining these was becoming time-consuming
and needed external help because many of the editors listed were not used by
members of the core team.
It also required us to make decisions about which plugin was best for a given
editor, even for editors we do not use.
The Go community at large is much better suited to managing this information. In Go 1.4, therefore, this support has been removed from the repository. Instead, there is a curated, informative list of what's available on a wiki page.
Most programs will run about the same speed or slightly faster in 1.4 than in 1.3; some will be slightly slower. There are many changes, making it hard to be precise about what to expect.
As mentioned above, much of the runtime was translated to Go from C, which led to some reduction in heap sizes. It also improved performance slightly because the Go compiler is better at optimization, due to things like inlining, than the C compiler used to build the runtime.
The garbage collector was sped up, leading to measurable improvements for garbage-heavy programs. On the other hand, the new write barriers slow things down again, typically by about the same amount but, depending on their behavior, some programs may be somewhat slower or faster.
Library changes that affect performance are documented below.
There are no new packages in this release.
TODO major changes
bufio: handling of empty tokens at EOF changed, may require scanner change (CL 145390043) syscall: now frozen (CL 129820043); go.sys subrepo created: http://golang.org/s/go1.4-syscall
The following list summarizes a number of minor changes to the library, mostly additions. See the relevant package documentation for more information about each change.
compress/flate,
compress/gzip,
and compress/zlib
packages now support a Reset method
for the decompressors, allowing them to reuse buffers and improve performance.
crypto/tls package
now supports APLN as defined in RFC 7301.
crypto/tls package
now supports programmatic selection of server certificates
through the new CertificateForName function
of the Config struct.
encoding/asn1 package,
optional elements with a default value will now only be omitted if they have that value.
encoding/csv package no longer
quotes empty strings but does quote the end-of-data marker \. (backslash dot).
This is permitted by the definition of CSV and allows it to work better with Postgres.
encoding/gob package has been rewritten to eliminate
the use of unsafe operations, allowing it to be used in environments that do not permit use of the
unsafe package.
For typical uses it will be 10-30% slower, but the delta is dependent on the type of the data and
in some cases, especially involving arrays, it can be faster.
There is no functional change.
fmt package,
formatting of pointers to maps has changed to be consistent with that of pointers
to structs, arrays, and so on.
For instance, &map[string]int{"one": 1} now prints by default as
&map[one: 1] rather than as a hexadecimal pointer value.
os package
now implements symbolic links on the Windows operating system
through the Symlink function.
Other operating systems already have this functionality.
reflect package's
Type interface
has a new method, Comparable,
that reports whether the type implements general comparisons.
reflect package, the
Value interface is now three instead of four words
because of changes to the implementation of interfaces in the runtime.
This saves memory but has no semantic effect.
text/scanner package's
Scanner type
has a new function,
IsIdentRune,
allowing one to control the definition of an identifier when scanning.
text/template package's boolean
functions eq, lt, and so on have been generalized to allow comparison
of signed and unsigned integers, simplifying their use in practice.
(Previously one could only compare values of the same signedness.)
All negative values compare less than all unsigned values.
time package now uses the standard symbol for the micro prefix,
the micro symbol (U+00B5 'ยต'), to print microsecond durations.
ParseDuration still accepts us
but the package no longer prints microseconds as us.