The State of Go - Andrew Gerrand

The State of GoWhere we are in June 2014

Andrew Gerrand

Yesterday

Initial reception

Looking back on the first months.

Complaints

The complaints started right away…

Familiar complaints

Familiar complaints

The first HN thread about Go was mostly an argument about braces and indentation.

Historic moments

The first public language change (Dec 10, 2009):

The gopher

The gopher

The gopher

A keen observation

Signs of hope

Today

Growth

User groups

Today there are 48 active user groups in 15 countries across 5 continents.

GopherCon

Community-driven. 3 days. 700 attendees. 24 talks. 1 panel. 4 hours of lightning talks!

The gopher

The gopher

The gopher

Go 1.3

Released last Wednesday, 18 June 2014.

Highlights:

Native Client support (amd64, 386)

DragonFly BSD, Plan 9, and Solaris support

Improved performance (GC, contiguous stacks, race detector, regexp)

Faster incremental builds (linker overhaul)

Precise GC of stacks

Static analysis in godoc

Tomorrow

Upcoming conferences

Go 1.4

Due December 1st, 2014.

What might be in it? (Speculative list follows.)

New language features

Nope.

Canonical import paths

A means for a repository to specify its canonical import path.

Goal: stop people from checking out repos from the wrong path.

Applications:

Stable vanity import paths.

"camlistore.org" -> "camlistore.googlesource.com/camlistore"

Make GitHub's fork model less painful.

(Proposal forthcoming.)

Internal packages

Current proposal:

"An import of a path containing the element “internal” is disallowed if the importingcode is outside the tree rooted at the parent of the “internal” directory."

Code in /a/b/c/internal/d/e/f can be imported only by code in the directorytree rooted at /a/b/c.

$GOROOT/src/pkg/internal/xxx can be imported only by other code in thestandard library ($GOROOT/src/).

And, later:

$GOPATH/src/mypkg/internal/foo can be imported only by code in$GOPATH/src/mypkg.

golang.org/s/go14internal (http://golang.org/s/go14internal)

src/pkg → src

Move the standard library from $GOROOT/src/pkg to $GOROOT/src.

Basically makes $GOROOT work just like $GOPATH.

(Not a big deal for most, but a nice clean up for the core.)

golang.org/s/go14nopkg (http://golang.org/s/go14nopkg)

File system access throttling

Throttling of access to the file system at the os package level.

Goal:

Launch a million goroutines that do file system stuff.

Don't launch a million operating system threads to block on the syscalls.

Make it work just as nicely as it does with the network.

(Proposal forthcoming.)

"go generate"

A convenience for generating Go source code. Example uses:

yacc: generating .go files from yacc grammar (.y) files

protobufs: generating .pb.go files from protocol buffer definition (.proto) files

HTML: embedding .html files into Go source code

Other ideas abound…

Proposed mechanism:

Add generator directive in Go source files:

//go:generate yacc -o gopher.go gopher.y

Use the Go tool to execute the directives:

% go generate [packagepath]

(Proposal forthcoming.)

More tool chain work

cmd/go support for gccgo

cmd/go support for embedded data files

Race detector improvements

The transition from C to Go continues:

Linker rewritten in Go (cmd/link)

Assembler rewritten in Go (cmd/asm)

Possibly: cmd/gc translated to Go

Better GC and malloc

The existing allocator is based on TCMalloc (designed for C/C++). It can be re-designed to better suit Go.

Lighter representation and colocation of GC metadata.

Various strategies for reducing GC pause time (concurrent sweep, etc).

Rough estimate: 10-20% overall speedup, 20-50% GC pause reduction, 5-10% RSSreduction.

Details:

golang.org/s/go14mallocgc (http://golang.org/s/go14mallocgc)

Other runtime changes

Concurrency improvements:

Lock-free fast path for non-blocking receive from empty channel.

Rewrite select to hold only one channel lock at a time.

C to Go:

G-stack runtime converted to Go.

Package changes

Easier structuring features in text/template

Bring reflect.Value back to 3 words

Fix embedded field visibility (reflect)

Safe version of encoding/gob

Unicode 7 support

Ports

Native Client arm

PowerPC 64

ARM64

Go support for Android

Go support for the NDK. (Like C/C++ today, but better!)

Goal: enough support for writing games.

Based on the work of Elias Naur and others in the community.

golang.org/s/go14android (http://golang.org/s/go14android)

More

golang.org/s/go14todo (http://golang.org/s/go14todo)

Questions?

Thank you

Andrew Gerrand