]> Cypherpunks repositories - gostls13.git/log
gostls13.git
9 years agotest: fix build on GOARCH=ppc64/ppc64le
Shenghou Ma [Sat, 6 Jun 2015 19:24:18 +0000 (15:24 -0400)]
test: fix build on GOARCH=ppc64/ppc64le

Change-Id: Ibf2879c0034250c5699e21ecea0eb76340597a2a
Reviewed-on: https://go-review.googlesource.com/10810
Reviewed-by: Austin Clements <austin@google.com>
9 years agodatabase/sql: refer correct issue in TestErrBadConnReconnect comment
Alex Brainman [Fri, 5 Jun 2015 07:02:09 +0000 (17:02 +1000)]
database/sql: refer correct issue in TestErrBadConnReconnect comment

Change-Id: I283ab238b60d3a47e86296e1fbfc73ba121bef19
Reviewed-on: https://go-review.googlesource.com/10745
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/asm: drop legacy RETURN mnemonic on ppc64
Austin Clements [Wed, 3 Jun 2015 19:13:10 +0000 (15:13 -0400)]
cmd/asm: drop legacy RETURN mnemonic on ppc64

Change-Id: I999b57ef5535c18e02cc27c9bc9f896d73126b50
Reviewed-on: https://go-review.googlesource.com/10674
Reviewed-by: Russ Cox <rsc@golang.org>
Reviewed-by: Minux Ma <minux@golang.org>
9 years agocmd/compile: use obj.ARET instead of ppc64.ARETURN
Austin Clements [Wed, 3 Jun 2015 19:10:53 +0000 (15:10 -0400)]
cmd/compile: use obj.ARET instead of ppc64.ARETURN

obj.ARET is the portable return mnemonic. ppc64.ARETURN is a legacy
alias.

This was done with
    sed -i s/ppc64\.ARETURN/obj.ARET/ cmd/compile/**/*.go
    sed -i s/ARETURN/obj.ARET/ cmd/internal/obj/ppc64/obj9.go

Change-Id: I4d8e83ff411cee764774a40ef4c7c34dcbca4e43
Reviewed-on: https://go-review.googlesource.com/10673
Reviewed-by: Russ Cox <rsc@golang.org>
Reviewed-by: Minux Ma <minux@golang.org>
9 years agoall: use RET instead of RETURN on ppc64
Austin Clements [Wed, 3 Jun 2015 18:59:27 +0000 (14:59 -0400)]
all: use RET instead of RETURN on ppc64

All of the architectures except ppc64 have only "RET" for the return
mnemonic. ppc64 used to have only "RETURN", but commit cf06ea6
introduced RET as a synonym for RETURN to make ppc64 consistent with
the other architectures. However, that commit was never followed up to
make the code itself consistent by eliminating uses of RETURN.

This commit replaces all uses of RETURN in the ppc64 assembly with
RET.

This was done with
    sed -i 's/\<RETURN\>/RET/' **/*_ppc64x.s
plus one manual change to syscall/asm.s.

Change-Id: I3f6c8d2be157df8841d48de988ee43f3e3087995
Reviewed-on: https://go-review.googlesource.com/10672
Reviewed-by: Rob Pike <r@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
Reviewed-by: Minux Ma <minux@golang.org>
9 years agocmd/go: update documentation: use 'go doc' rather than 'godoc'
Robert Griesemer [Fri, 5 Jun 2015 20:14:06 +0000 (13:14 -0700)]
cmd/go: update documentation: use 'go doc' rather than 'godoc'

Change-Id: I318c1ef75b18d4687f13499ac225dde2d053505e
Reviewed-on: https://go-review.googlesource.com/10776
Reviewed-by: Rob Pike <r@golang.org>
9 years agocmd/go: use new vet location
Robert Griesemer [Thu, 4 Jun 2015 23:09:15 +0000 (16:09 -0700)]
cmd/go: use new vet location

Change-Id: I7d96ebcca5954152edb821bb41b6047a1c622949
Reviewed-on: https://go-review.googlesource.com/10731
Reviewed-by: Andrew Gerrand <adg@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
9 years agogo/types: remove unused return value
Robert Griesemer [Fri, 5 Jun 2015 18:44:31 +0000 (11:44 -0700)]
go/types: remove unused return value

Port of https://go-review.googlesource.com/10773 from x/tools.

Change-Id: I6aba6a63a5448b8fcbcc7f072c627c27965dbe20
Reviewed-on: https://go-review.googlesource.com/10774
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agoruntime: add blank assignment to defeat "declared but not used" error from go/types
Alan Donovan [Fri, 5 Jun 2015 17:27:49 +0000 (13:27 -0400)]
runtime: add blank assignment to defeat "declared but not used" error from go/types

gc should ideally consider this an error too; see golang/go#8560.

Change-Id: Ieee71c4ecaff493d7f83e15ba8c8a04ee90a4cf1
Reviewed-on: https://go-review.googlesource.com/10757
Reviewed-by: Robert Griesemer <gri@golang.org>
9 years agogo/types: remove MethodSetCache - not needed
Robert Griesemer [Fri, 5 Jun 2015 17:35:29 +0000 (10:35 -0700)]
go/types: remove MethodSetCache - not needed

In x/tools, MethodSetCache was moved from x/tools/go/types to
x/tools/go/types/typeutil. Mirror that change.

Change-Id: Ib838a9518371473c83fa4abc2778d42f33947c98
Reviewed-on: https://go-review.googlesource.com/10771
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agoruntime: use correct SP when installing stack barriers
Austin Clements [Fri, 5 Jun 2015 15:07:21 +0000 (11:07 -0400)]
runtime: use correct SP when installing stack barriers

Currently the stack barriers are installed at the next frame boundary
after gp.sched.sp + 1024*2^n for n=0,1,2,... However, when a G is in a
system call, we set gp.sched.sp to 0, which causes stack barriers to
be installed at *every* frame. This easily overflows the slice we've
reserved for storing the stack barrier information, and causes a
"slice bounds out of range" panic in gcInstallStackBarrier.

Fix this by using gp.syscallsp instead of gp.sched.sp if it's
non-zero. This is the same logic that gentraceback uses to determine
the current SP.

Fixes #11049.

Change-Id: Ie40eeee5bec59b7c1aa715a7c17aa63b1f1cf4e8
Reviewed-on: https://go-review.googlesource.com/10755
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/dist: add more logging details when go list std cmd fails
Brad Fitzpatrick [Fri, 5 Jun 2015 15:11:34 +0000 (08:11 -0700)]
cmd/dist: add more logging details when go list std cmd fails

Change-Id: I12e6990b46ea9c733a5718dc5ca67f1fcd2dec66
Reviewed-on: https://go-review.googlesource.com/10754
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/vet: skip exec tests on systems that can't exec
Russ Cox [Fri, 5 Jun 2015 14:34:17 +0000 (10:34 -0400)]
cmd/vet: skip exec tests on systems that can't exec

Change-Id: I09257b8f5482cba10b5f4d3813c778d6e9e74d40
Reviewed-on: https://go-review.googlesource.com/10752
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agodebug/gosym: avoid calling the shell in test
Alexis Imperial-Legrand [Tue, 2 Jun 2015 15:27:43 +0000 (17:27 +0200)]
debug/gosym: avoid calling the shell in test

Change-Id: I95bf62c0f2d77dd67515921e6aefa511cce8d95d
Reviewed-on: https://go-review.googlesource.com/10633
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/go: fix detection of ELF binaries
Russ Cox [Fri, 5 Jun 2015 13:58:47 +0000 (09:58 -0400)]
cmd/go: fix detection of ELF binaries

Change-Id: I0e6f1a39b3d6b15d05891c8b25ab6644356bde5f
Reviewed-on: https://go-review.googlesource.com/10751
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoflag: Describe the ErrorHandling enum values.
Aaron Jacobs [Fri, 5 Jun 2015 00:12:34 +0000 (10:12 +1000)]
flag: Describe the ErrorHandling enum values.

ContinueOnError is particularly confusing, because it causes
FlagSet.Parse to return as soon as it sees an error. I gather that the
intent is "continue the program" rather than "continue parsing",
compared to exiting or panicking.

Change-Id: I27370ce1f321ea4debcee5b03faff3532495c71a
Reviewed-on: https://go-review.googlesource.com/10740
Reviewed-by: Rob Pike <r@golang.org>
9 years agocmd/go: add -tags option to go vet
Rob Pike [Thu, 4 Jun 2015 21:28:14 +0000 (14:28 -0700)]
cmd/go: add -tags option to go vet

Actually add all build flags, so we also get things like -race.

Fixes #10228.

Change-Id: I5f77dda9d1ee3208e1833702f12f68c2731c4b22
Reviewed-on: https://go-review.googlesource.com/10697
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agodoc/go1.5.txt: mention go test -count
Russ Cox [Fri, 5 Jun 2015 05:31:52 +0000 (01:31 -0400)]
doc/go1.5.txt: mention go test -count

Change-Id: I1d7b728bd161da7bd6dd460862d8be072921e8b9
Reviewed-on: https://go-review.googlesource.com/10763
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/compile: document Node fields used by each Op
Russ Cox [Thu, 4 Jun 2015 03:57:59 +0000 (23:57 -0400)]
cmd/compile: document Node fields used by each Op

Change-Id: If969d7a06c83447ee38da30f1477a6cf4bfa1a03
Reviewed-on: https://go-review.googlesource.com/10691
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/link: delete dead flags
Russ Cox [Thu, 21 May 2015 18:42:14 +0000 (14:42 -0400)]
cmd/link: delete dead flags

Also fix the interaction between -buildmode and -shared.
It's okay for -shared to change the default build mode,
but it's not okay for it to silently override an explicit -buildmode=exe.

Change-Id: Id40f93d140cddf75b19e262b3ba4856ee09a07ba
Reviewed-on: https://go-review.googlesource.com/10315
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agocmd/link: deprecate -X name value in favor of -X name=value
Russ Cox [Thu, 21 May 2015 18:35:02 +0000 (14:35 -0400)]
cmd/link: deprecate -X name value in favor of -X name=value

People invoking the linker directly already have to change their scripts
to use the new "go tool link", so this is a good time to make the -X flag
behave like all other Go flags and take just a single argument.

The old syntax will continue to be accepted (it is rewritten into the new
syntax before flag parsing). Maybe some day we will be able to retire it.

Even if we never retire the old syntax, having the new syntax at least
makes the rewriting much less of a kludge.

Change-Id: I91e8df94f4c22b2186e81d7f1016b8767d777eac
Reviewed-on: https://go-review.googlesource.com/10310
Reviewed-by: Rob Pike <r@golang.org>
9 years agocmd/compile, cmd/link: add docs
Russ Cox [Thu, 21 May 2015 18:11:33 +0000 (14:11 -0400)]
cmd/compile, cmd/link: add docs

These are the Go 1.4 docs but refreshed for Go 1.5.
The most sigificant change is that all references to the Plan 9 toolchain are gone.
The tools no longer bear any meaningful resemblance.

Change-Id: I44f5cadb832a982323d7fee0b77673e55d761b35
Reviewed-on: https://go-review.googlesource.com/10298
Reviewed-by: Rob Pike <r@golang.org>
9 years agodoc/go1.5.txt: GOMAXPROCS change
Russ Cox [Fri, 5 Jun 2015 04:37:29 +0000 (00:37 -0400)]
doc/go1.5.txt: GOMAXPROCS change

Change-Id: I5c991cad38c9e2c839314a56d3415a2aa09c1561
Reviewed-on: https://go-review.googlesource.com/10762
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoruntime: default GOMAXPROCS to NumCPU(), not 1
Russ Cox [Thu, 28 May 2015 16:35:35 +0000 (12:35 -0400)]
runtime: default GOMAXPROCS to NumCPU(), not 1

See golang.org/s/go15gomaxprocs for details.

Change-Id: I8de5df34fa01d31d78f0194ec78a2474c281243c
Reviewed-on: https://go-review.googlesource.com/10668
Reviewed-by: Rob Pike <r@golang.org>
9 years agocmd/pprof: use copy of svgpan library instead of link to remote site
Russ Cox [Mon, 27 Apr 2015 18:40:34 +0000 (14:40 -0400)]
cmd/pprof: use copy of svgpan library instead of link to remote site

Fixes #10375.

Change-Id: I78dc3e12035d130c405bdb284b0cceea19f084f6
Reviewed-on: https://go-review.googlesource.com/10690
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agotesting: add -test.count flag to run tests and benchmarks multiple times
Russ Cox [Thu, 4 Jun 2015 02:21:07 +0000 (22:21 -0400)]
testing: add -test.count flag to run tests and benchmarks multiple times

The flag is available from the go test command as -count:

% go test -run XXX -bench . -count 3
PASS
BenchmarkSprintfEmpty       30000000         54.0 ns/op
BenchmarkSprintfEmpty       30000000         51.9 ns/op
BenchmarkSprintfEmpty       30000000         53.8 ns/op
BenchmarkSprintfString      10000000        238 ns/op
BenchmarkSprintfString      10000000        239 ns/op
BenchmarkSprintfString      10000000        234 ns/op
BenchmarkSprintfInt         10000000        232 ns/op
BenchmarkSprintfInt         10000000        226 ns/op
BenchmarkSprintfInt         10000000        225 ns/op
...

If -cpu is set, each test is run n times for each cpu value.

Original by r (CL 10663).

Change-Id: If3dfbdf21698952daac9249b5dbca66f5301e91b
Reviewed-on: https://go-review.googlesource.com/10669
Reviewed-by: Rob Pike <r@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agocmd/go: read new non-ELF build ID in binaries
Russ Cox [Thu, 4 Jun 2015 19:20:41 +0000 (15:20 -0400)]
cmd/go: read new non-ELF build ID in binaries

Fixes #11048.
Fixes #11075.

Change-Id: I81f5ef1e1944056ce5494c91aa4a4a63c758f566
Reviewed-on: https://go-review.googlesource.com/10709
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/link: implement -buildid for non-ELF binaries
Russ Cox [Thu, 4 Jun 2015 19:15:48 +0000 (15:15 -0400)]
cmd/link: implement -buildid for non-ELF binaries

Non-ELF binary formats are much less flexible and typically do not
have a good place to store the build ID.

We store it as raw bytes at the beginning of the text segment.

The only system I know of that will be upset about this is NaCl,
and NaCl is an ELF system and does not use this.

For #11048.

Change-Id: Iaa7ace703c4cf36392e752eea9b55e2ce49e9826
Reviewed-on: https://go-review.googlesource.com/10708
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/go: use ELF note instead of binary stamp on ELF systems
Russ Cox [Thu, 4 Jun 2015 18:41:02 +0000 (14:41 -0400)]
cmd/go: use ELF note instead of binary stamp on ELF systems

Other binary formats to follow.

For #11048.

Change-Id: Ia2d8b47c99c99d171c014b7cfd23c1c7ada5231c
Reviewed-on: https://go-review.googlesource.com/10707
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/link: add -buildid flag to write Go build ID to ELF output, same as cmd/compile
Russ Cox [Thu, 4 Jun 2015 18:31:05 +0000 (14:31 -0400)]
cmd/link: add -buildid flag to write Go build ID to ELF output, same as cmd/compile

Other binary formats to follow.

Using our own note instead of the GNU build ID note because
we are not the GNU project, and I can't guarantee that the semantics
of our note and the semantics of the GNU note will match forever.
(Also they don't match today.)

For #11048.

Change-Id: Iec7e5a2e49d52b6d3a51b0aface2de7c77a45491
Reviewed-on: https://go-review.googlesource.com/10706
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/link: add -h flag, for debugging, same as cmd/compile
Russ Cox [Thu, 4 Jun 2015 18:30:30 +0000 (14:30 -0400)]
cmd/link: add -h flag, for debugging, same as cmd/compile

Change-Id: I3c9b05879fe0b6e94b63e9b65e4411ba2a917134
Reviewed-on: https://go-review.googlesource.com/10705
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/go, cmd/link: use "Go" not "GO" as ELF note identifier
Russ Cox [Thu, 4 Jun 2015 18:27:39 +0000 (14:27 -0400)]
cmd/go, cmd/link: use "Go" not "GO" as ELF note identifier

Change-Id: I038e892725836ab7718f7638e8ad5712953f2cb5
Reviewed-on: https://go-review.googlesource.com/10704
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/go: simplify ELF note reading and enable during bootstrap
Russ Cox [Thu, 4 Jun 2015 18:17:51 +0000 (14:17 -0400)]
cmd/go: simplify ELF note reading and enable during bootstrap

The bootstrap restriction is to avoid needing cgo for package net.
There's no problem with building debug/elf and debug/dwarf,
so do that.

An upcoming CL is going to add more note processing code,
and it simplifies things not to have to think about the code being
missing half the time.

Change-Id: I0e2f120ac23f14db6ecfcec7bfe254a69abcf7b6
Reviewed-on: https://go-review.googlesource.com/10703
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/go: fix fmt.Errorf call sites (latent bug)
Robert Griesemer [Thu, 4 Jun 2015 22:49:43 +0000 (15:49 -0700)]
cmd/go: fix fmt.Errorf call sites (latent bug)

Change-Id: I4ff42113c5dda553d6f3d58f0e4836dce7c84710
Reviewed-on: https://go-review.googlesource.com/10730
Reviewed-by: Rob Pike <r@golang.org>
9 years agocmd/doc: do not show unexported constants
Rob Pike [Tue, 2 Jun 2015 21:43:38 +0000 (14:43 -0700)]
cmd/doc: do not show unexported constants

The go/doc package doesn't remove unexported entries from const
and var blocks, so we must trim them ourselves.

Fixes #11008

Change-Id: Ibd60d87e09333964e2588340a2ca2b8804bbaa28
Reviewed-on: https://go-review.googlesource.com/10643
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/vet: adjust vet to use go/types and friends from std repo
Robert Griesemer [Thu, 4 Jun 2015 20:07:22 +0000 (13:07 -0700)]
cmd/vet: adjust vet to use go/types and friends from std repo

- s|"golang.org/x/tools/go/exact"|"go/constant"|
- s|"golang.org/x/tools/go/types"|"go/types"|
- removed import of gcimporter
- import "go/importer" instead
- trivial adjustments to make use of go/importer
- adjusted import paths for whitelist.go

Change-Id: I43488ff44c329cd869c92dcc31193fb31bebfd29
Reviewed-on: https://go-review.googlesource.com/10695
Reviewed-by: Rob Pike <r@golang.org>
9 years agocmd/vet: move cmd/vet into std repo
Robert Griesemer [Thu, 4 Jun 2015 19:54:58 +0000 (12:54 -0700)]
cmd/vet: move cmd/vet into std repo

cp -r x/tools/cmd/vet cmd/vet without any changes.
The next change will adjust the source to use std
repo go/types and friends.

This may (temporarily) break the build; the next
commit (immediately following) will fix it. We do
it in two commits so that we can see the manual
changes.

Change-Id: Ic45dab7066f13923e21f8c61200c8c3fd447b171
Reviewed-on: https://go-review.googlesource.com/10694
Reviewed-by: Rob Pike <r@golang.org>
9 years agocmd/dist: use -tags=lldb for iOS tests
David Crawshaw [Thu, 4 Jun 2015 19:03:38 +0000 (15:03 -0400)]
cmd/dist: use -tags=lldb for iOS tests

As of golang.org/cl/9154, running go test will override a previous
go install -a -tags=lldb std with the tag-less version of stdlib. So
we pass -tags=lldb into the relevant go test commands.

Change-Id: I1c718289d7212373a9383eff53a643f06598f5ed
Reviewed-on: https://go-review.googlesource.com/10701
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>

9 years agonet/http: the Set-Cookie Expiration time zone should be GMT, not UTC
Brad Fitzpatrick [Thu, 4 Jun 2015 19:57:53 +0000 (12:57 -0700)]
net/http: the Set-Cookie Expiration time zone should be GMT, not UTC

Per RFC 6265.

Change-Id: I2b6b145f5d057f96509332509d722602ed9e2bbd
Reviewed-on: https://go-review.googlesource.com/10711
Reviewed-by: Brett Slatkin <bslatkin@google.com>
Reviewed-by: Andrew Gerrand <adg@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

9 years agotest: make test/run.go support sharding
Brad Fitzpatrick [Thu, 4 Jun 2015 06:21:30 +0000 (23:21 -0700)]
test: make test/run.go support sharding

Also modifies 'dist test' to use that sharding, and removes some old
temporary stuff from dist test which are no longer required.

'dist test' now also supports running a list of tests given in
arguments, mutually exclusive with the existing -run=REGEXP flag. The
hacky fast paths for avoiding the 1 second "go list" latency are now
removed and only apply to the case where partial tests are run via
args, instead of regex.  The build coordinator will use both styles
for awhile. (the statically-sharded ARM builders on scaleway will
continue to use regexps, but the dynamically-shared builders on GCE
will use the list of tests)

Updates #10029

Change-Id: I557800a54dfa6f3b5100ef4c26fe397ba5189813
Reviewed-on: https://go-review.googlesource.com/10688
Reviewed-by: Andrew Gerrand <adg@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/internal/obj/arm64: make function prologue more predictable
Josh Bleecher Snyder [Fri, 29 May 2015 01:17:58 +0000 (18:17 -0700)]
cmd/internal/obj/arm64: make function prologue more predictable

Static branch prediction guesses that forward branches aren't taken.
Since stacks are rarely grown, make the forward branch mean grow.

While we're here, remove the debug-only instruction
saving the frame size in the temp register.

Sample disassembly for

func f() {
_ = [128]byte{}
}

Before:

0x4008248         ldr        x1, [x28, #0x10]
0x400824c         sub        x2, sp, #0x90
0x4008250         cmp        x2, x1
0x4008254         b.hi       0x4008268
0x4008258         mov        x3, x30
0x400825c         movz       x27, #0x90
0x4008260         bl         runtime.morestack_noctxt
0x4008264         b          main.f
0x4008268         sub        sp, sp, #0x90
0x400826c         add        x16, sp, #0x10
0x4008270         str        xzr, [x16]
0x4008274         str        xzr, [x16, #0x8]
0x4008278         str        xzr, [x16, #0x10]
0x400827c         str        xzr, [x16, #0x18]
0x4008280         str        xzr, [x16, #0x20]
0x4008284         str        xzr, [x16, #0x28]
0x4008288         str        xzr, [x16, #0x30]
0x400828c         str        xzr, [x16, #0x38]
0x4008290         str        xzr, [x16, #0x40]
0x4008294         str        xzr, [x16, #0x48]
0x4008298         str        xzr, [x16, #0x50]
0x400829c         str        xzr, [x16, #0x58]
0x40082a0         str        xzr, [x16, #0x60]
0x40082a4         str        xzr, [x16, #0x68]
0x40082a8         str        xzr, [x16, #0x70]
0x40082ac         str        xzr, [x16, #0x78]
0x40082b0         add        sp, sp, #0x90
0x40082b4         ret

After:

0x4004bc8         ldr        x1, [x28, #0x10]
0x4004bcc         sub        x2, sp, #0x90
0x4004bd0         cmp        x2, x1
0x4004bd4         b.ls       0x4004c28
0x4004bd8         sub        sp, sp, #0x90
0x4004bdc         add        x16, sp, #0x10
0x4004be0         str        xzr, [x16]
0x4004be4         str        xzr, [x16, #0x8]
0x4004be8         str        xzr, [x16, #0x10]
0x4004bec         str        xzr, [x16, #0x18]
0x4004bf0         str        xzr, [x16, #0x20]
0x4004bf4         str        xzr, [x16, #0x28]
0x4004bf8         str        xzr, [x16, #0x30]
0x4004bfc         str        xzr, [x16, #0x38]
0x4004c00         str        xzr, [x16, #0x40]
0x4004c04         str        xzr, [x16, #0x48]
0x4004c08         str        xzr, [x16, #0x50]
0x4004c0c         str        xzr, [x16, #0x58]
0x4004c10         str        xzr, [x16, #0x60]
0x4004c14         str        xzr, [x16, #0x68]
0x4004c18         str        xzr, [x16, #0x70]
0x4004c1c         str        xzr, [x16, #0x78]
0x4004c20         add        sp, sp, #0x90
0x4004c24         ret
0x4004c28         mov        x3, x30
0x4004c2c         bl         runtime.morestack_noctxt
0x4004c30         b          main.f

Updates #10587.

Package sort benchmarks using an iPhone 6:

name            old time/op  new time/op  delta
SearchWrappers   355ns ± 1%   328ns ± 1%  -7.57%  (p=0.000 n=25+19)
SortString1K     580µs ± 1%   577µs ± 1%  -0.48%  (p=0.000 n=25+25)
StableString1K  1.04ms ± 0%  1.04ms ± 0%    ~     (p=0.851 n=24+25)
SortInt1K        251µs ± 1%   247µs ± 1%  -1.52%  (p=0.000 n=23+25)
StableInt1K      267µs ± 2%   261µs ± 2%  -2.02%  (p=0.000 n=25+25)
SortInt64K      23.8ms ± 1%  23.6ms ± 0%  -0.97%  (p=0.000 n=25+23)
StableInt64K    22.8ms ± 0%  22.4ms ± 1%  -1.76%  (p=0.000 n=24+25)
Sort1e2          123µs ± 1%   124µs ± 1%    ~     (p=0.256 n=23+23)
Stable1e2        248µs ± 1%   247µs ± 1%  -0.69%  (p=0.000 n=23+25)
Sort1e4         24.3ms ± 2%  24.6ms ± 5%  +1.36%  (p=0.017 n=22+25)
Stable1e4       77.2ms ± 6%  76.2ms ± 5%  -1.36%  (p=0.020 n=25+25)
Sort1e6          3.95s ± 8%   3.95s ± 8%    ~     (p=0.863 n=25+25)
Stable1e6        15.7s ± 1%   15.5s ± 1%  -1.11%  (p=0.000 n=22+23)

Change-Id: I377b3817af2ed27ddeecf24edef97fad91fc1afc
Reviewed-on: https://go-review.googlesource.com/10500
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Dave Cheney <dave@cheney.net>
9 years agoruntime, cmd/internal/obj/arm: improve arm function prologue
Josh Bleecher Snyder [Thu, 28 May 2015 22:18:35 +0000 (15:18 -0700)]
runtime, cmd/internal/obj/arm: improve arm function prologue

When stack growth is not needed, as it usually is not,
execute only a single conditional branch
rather than three conditional instructions.
This adds 4 bytes to every function,
but might speed up execution in the common case.

Sample disassembly for

func f() {
_ = [128]byte{}
}

Before:

TEXT main.f(SB) x.go
x.go:3 0x2000 e59a1008 MOVW 0x8(R10), R1
x.go:3 0x2004 e59fb028 MOVW 0x28(R15), R11
x.go:3 0x2008 e08d200b ADD R11, R13, R2
x.go:3 0x200c e1520001 CMP R1, R2
x.go:3 0x2010 91a0300e MOVW.LS R14, R3
x.go:3 0x2014 9b0118a9 BL.LS runtime.morestack_noctxt(SB)
x.go:3 0x2018 9afffff8 B.LS main.f(SB)
x.go:3 0x201c e52de084 MOVW.W R14, -0x84(R13)
x.go:4 0x2020 e28d1004 ADD $4, R13, R1
x.go:4 0x2024 e3a00000 MOVW $0, R0
x.go:4 0x2028 eb012255 BL 0x4a984
x.go:5 0x202c e49df084 RET #132
x.go:5 0x2030 eafffffe B 0x2030
x.go:5 0x2034 ffffff7c ?

After:

TEXT main.f(SB) x.go
x.go:3 0x2000 e59a1008 MOVW 0x8(R10), R1
x.go:3 0x2004 e59fb02c MOVW 0x2c(R15), R11
x.go:3 0x2008 e08d200b ADD R11, R13, R2
x.go:3 0x200c e1520001 CMP R1, R2
x.go:3 0x2010 9a000004 B.LS 0x2028
x.go:3 0x2014 e52de084 MOVW.W R14, -0x84(R13)
x.go:4 0x2018 e28d1004 ADD $4, R13, R1
x.go:4 0x201c e3a00000 MOVW $0, R0
x.go:4 0x2020 eb0124dc BL 0x4b398
x.go:5 0x2024 e49df084 RET #132
x.go:5 0x2028 e1a0300e MOVW R14, R3
x.go:5 0x202c eb011b0d BL runtime.morestack_noctxt(SB)
x.go:5 0x2030 eafffff2 B main.f(SB)
x.go:5 0x2034 eafffffe B 0x2034
x.go:5 0x2038 ffffff7c ?

Updates #10587.

package sort benchmarks on an iPhone 6:

name            old time/op  new time/op  delta
SortString1K     569µs ± 0%   565µs ± 1%  -0.75%  (p=0.000 n=23+24)
StableString1K   872µs ± 1%   870µs ± 1%  -0.16%  (p=0.009 n=23+24)
SortInt1K        317µs ± 2%   316µs ± 2%    ~     (p=0.410 n=26+26)
StableInt1K      343µs ± 1%   339µs ± 1%  -1.07%  (p=0.000 n=22+23)
SortInt64K      30.0ms ± 1%  30.0ms ± 1%    ~     (p=0.091 n=25+24)
StableInt64K    30.2ms ± 0%  30.0ms ± 0%  -0.69%  (p=0.000 n=22+22)
Sort1e2          147µs ± 1%   146µs ± 0%  -0.48%  (p=0.000 n=25+24)
Stable1e2        290µs ± 1%   286µs ± 1%  -1.30%  (p=0.000 n=23+24)
Sort1e4         29.5ms ± 2%  29.7ms ± 1%  +0.71%  (p=0.000 n=23+23)
Stable1e4       88.7ms ± 4%  88.6ms ± 8%  -0.07%  (p=0.022 n=26+26)
Sort1e6          4.81s ± 7%   4.83s ± 7%    ~     (p=0.192 n=26+26)
Stable1e6        18.3s ± 1%   18.1s ± 1%  -0.76%  (p=0.000 n=25+23)
SearchWrappers   318ns ± 1%   344ns ± 1%  +8.14%  (p=0.000 n=23+26)

package sort benchmarks on a first generation rpi:

name            old time/op  new time/op  delta
SearchWrappers  4.13µs ± 0%  3.95µs ± 0%   -4.42%  (p=0.000 n=15+13)
SortString1K    5.81ms ± 1%  5.82ms ± 2%     ~     (p=0.400 n=14+15)
StableString1K  9.69ms ± 1%  9.73ms ± 0%     ~     (p=0.121 n=15+11)
SortInt1K       3.30ms ± 2%  3.66ms ±19%  +10.82%  (p=0.000 n=15+14)
StableInt1K     5.97ms ±15%  4.17ms ± 8%  -30.05%  (p=0.000 n=15+15)
SortInt64K       319ms ± 1%   295ms ± 1%   -7.65%  (p=0.000 n=15+15)
StableInt64K     343ms ± 0%   332ms ± 0%   -3.26%  (p=0.000 n=12+13)
Sort1e2         3.36ms ± 2%  3.22ms ± 4%   -4.10%  (p=0.000 n=15+15)
Stable1e2       6.74ms ± 1%  6.43ms ± 2%   -4.67%  (p=0.000 n=15+15)
Sort1e4          247ms ± 1%   247ms ± 1%     ~     (p=0.331 n=15+14)
Stable1e4        864ms ± 0%   820ms ± 0%   -5.15%  (p=0.000 n=14+15)
Sort1e6          41.2s ± 0%   41.2s ± 0%   +0.15%  (p=0.000 n=13+14)
Stable1e6         192s ± 0%    182s ± 0%   -5.07%  (p=0.000 n=14+14)

Change-Id: I8a9db77e1d4ea1956575895893bc9d04bd81204b
Reviewed-on: https://go-review.googlesource.com/10497
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/internal/obj/x86: make function prologue more predictable
Josh Bleecher Snyder [Fri, 29 May 2015 00:31:24 +0000 (17:31 -0700)]
cmd/internal/obj/x86: make function prologue more predictable

Static branch prediction guesses that forward branches aren't taken.
Since stacks are rarely grown, make the forward branch mean grow.

Sample disassembly for

func f() {
_ = [128]byte{}
}

Before:

TEXT main.f(SB) x.go
x.go:3 0x2000 65488b0c25a0080000 GS MOVQ GS:0x8a0, CX
x.go:3 0x2009 483b6110 CMPQ 0x10(CX), SP
x.go:3 0x200d 7707 JA 0x2016
x.go:3 0x200f e88c410400 CALL runtime.morestack_noctxt(SB)
x.go:3 0x2014 ebea JMP main.f(SB)
x.go:3 0x2016 4881ec80000000 SUBQ $0x80, SP
x.go:4 0x201d 488d3c24 LEAQ 0(SP), DI
x.go:4 0x2021 31c0 XORL AX, AX
x.go:4 0x2023 e8cc640400 CALL 0x484f4
x.go:5 0x2028 4881c480000000 ADDQ $0x80, SP
x.go:5 0x202f c3 RET

After:

TEXT main.f(SB) x.go
x.go:3 0x2000 65488b0c25a0080000 GS MOVQ GS:0x8a0, CX
x.go:3 0x2009 483b6110 CMPQ 0x10(CX), SP
x.go:3 0x200d 761a JBE 0x2029
x.go:3 0x200f 4881ec80000000 SUBQ $0x80, SP
x.go:4 0x2016 488d3c24 LEAQ 0(SP), DI
x.go:4 0x201a 31c0 XORL AX, AX
x.go:4 0x201c e813740400 CALL 0x49434
x.go:5 0x2021 4881c480000000 ADDQ $0x80, SP
x.go:5 0x2028 c3 RET
x.go:3 0x2029 e8224f0400 CALL runtime.morestack_noctxt(SB)
x.go:3 0x202e ebd0 JMP main.f(SB)

Updates #10587.

Sample benchmarks on a 2.8 GHz Intel Core i7:

package sort

name            old mean              new mean              delta
SearchWrappers   134ns × (0.99,1.01)   132ns × (0.99,1.01)  -1.73% (p=0.000 n=15+14)
SortString1K     215µs × (0.99,1.01)   213µs × (0.99,1.01)  -0.61% (p=0.020 n=14+15)
StableString1K   311µs × (0.99,1.02)   309µs × (0.99,1.02)    ~    (p=0.077 n=14+15)
SortInt1K        103µs × (0.99,1.02)   100µs × (0.98,1.01)  -3.34% (p=0.000 n=15+15)
StableInt1K      102µs × (0.99,1.01)    98µs × (0.97,1.04)  -3.53% (p=0.000 n=15+15)
SortInt64K      10.1ms × (0.98,1.02)   9.7ms × (0.99,1.01)  -3.86% (p=0.000 n=14+15)
StableInt64K    8.70ms × (0.99,1.01)  8.44ms × (0.99,1.03)  -2.93% (p=0.000 n=14+15)
Sort1e2         51.2µs × (1.00,1.01)  48.9µs × (0.99,1.02)  -4.48% (p=0.000 n=13+15)
Stable1e2        100µs × (0.99,1.02)    99µs × (0.99,1.01)  -1.15% (p=0.000 n=14+13)
Sort1e4         11.1ms × (0.99,1.02)  10.4ms × (0.99,1.01)  -6.02% (p=0.000 n=15+14)
Stable1e4       30.6ms × (0.99,1.01)  30.3ms × (0.99,1.02)  -1.02% (p=0.001 n=15+14)
Sort1e6          1.75s × (0.99,1.02)   1.66s × (0.98,1.03)  -4.95% (p=0.000 n=14+15)
Stable1e6        6.31s × (0.99,1.01)   6.26s × (0.99,1.01)  -0.79% (p=0.002 n=15+15)

package regexp

name                          old mean              new mean              delta
Literal                        131ns × (0.99,1.01)   130ns × (0.99,1.03)  -1.07% (p=0.004 n=14+15)
NotLiteral                    2.13µs × (0.99,1.01)  2.01µs × (0.99,1.03)  -5.71% (p=0.000 n=14+14)
MatchClass                    3.15µs × (0.99,1.01)  3.04µs × (0.99,1.02)  -3.40% (p=0.000 n=15+15)
MatchClass_InRange            2.92µs × (0.99,1.01)  2.77µs × (0.99,1.02)  -5.05% (p=0.000 n=13+15)
ReplaceAll                    2.17µs × (0.99,1.02)  2.06µs × (0.99,1.01)  -5.19% (p=0.000 n=15+13)
AnchoredLiteralShortNonMatch   116ns × (0.99,1.02)   113ns × (0.99,1.01)  -2.75% (p=0.000 n=15+14)
AnchoredLiteralLongNonMatch    125ns × (0.99,1.01)   127ns × (0.98,1.02)  +1.49% (p=0.000 n=15+15)
AnchoredShortMatch             178ns × (0.99,1.02)   175ns × (0.99,1.01)  -1.62% (p=0.000 n=15+13)
AnchoredLongMatch              328ns × (0.99,1.00)   341ns × (0.99,1.01)  +3.73% (p=0.000 n=12+15)
OnePassShortA                  773ns × (0.99,1.02)   752ns × (0.99,1.01)  -2.78% (p=0.000 n=15+13)
NotOnePassShortA               794ns × (0.99,1.03)   780ns × (0.99,1.02)  -1.75% (p=0.001 n=15+15)
OnePassShortB                  608ns × (0.99,1.01)   591ns × (0.99,1.02)  -2.86% (p=0.000 n=15+14)
NotOnePassShortB               576ns × (0.99,1.01)   571ns × (0.99,1.02)  -0.74% (p=0.035 n=15+15)
OnePassLongPrefix              131ns × (0.99,1.02)   130ns × (0.99,1.02)  -1.32% (p=0.003 n=15+15)
OnePassLongNotPrefix           503ns × (0.99,1.02)   481ns × (0.99,1.01)  -4.34% (p=0.000 n=15+13)
MatchEasy0_32                  102ns × (0.98,1.01)   101ns × (0.99,1.02)    ~    (p=0.907 n=15+14)
MatchEasy0_1K                  617ns × (0.99,1.02)   634ns × (0.98,1.02)  +2.77% (p=0.000 n=15+15)
MatchEasy0_32K                10.9µs × (0.99,1.01)  11.1µs × (0.99,1.01)  +1.59% (p=0.000 n=15+15)
MatchEasy0_1M                  406µs × (0.99,1.02)   410µs × (0.99,1.02)  +1.01% (p=0.000 n=14+15)
MatchEasy0_32M                13.4ms × (0.99,1.01)  13.7ms × (0.99,1.02)  +1.64% (p=0.000 n=12+15)
MatchEasy1_32                 83.7ns × (0.98,1.02)  83.0ns × (0.98,1.02)    ~    (p=0.190 n=15+15)
MatchEasy1_1K                 1.46µs × (0.99,1.02)  1.39µs × (0.99,1.02)  -4.83% (p=0.000 n=15+15)
MatchEasy1_32K                49.4µs × (0.99,1.01)  49.4µs × (0.99,1.01)    ~    (p=0.205 n=15+15)
MatchEasy1_1M                 1.72ms × (0.99,1.02)  1.75ms × (0.99,1.01)  +1.34% (p=0.000 n=15+15)
MatchEasy1_32M                55.5ms × (0.99,1.01)  56.1ms × (0.99,1.02)  +1.10% (p=0.002 n=15+15)
MatchMedium_32                1.37µs × (0.99,1.04)  1.33µs × (0.99,1.01)  -2.87% (p=0.000 n=15+15)
MatchMedium_1K                41.1µs × (0.99,1.02)  40.4µs × (0.99,1.02)  -1.59% (p=0.000 n=15+15)
MatchMedium_32K               1.71ms × (0.99,1.01)  1.75ms × (0.99,1.02)  +2.36% (p=0.000 n=14+15)
MatchMedium_1M                54.5ms × (0.99,1.01)  56.1ms × (0.99,1.01)  +2.94% (p=0.000 n=13+15)
MatchMedium_32M                1.75s × (0.99,1.01)   1.80s × (0.99,1.01)  +2.77% (p=0.000 n=15+15)
MatchHard_32                  2.12µs × (0.99,1.02)  2.06µs × (0.99,1.01)  -2.60% (p=0.000 n=15+14)
MatchHard_1K                  64.4µs × (0.98,1.02)  62.2µs × (0.99,1.01)  -3.33% (p=0.000 n=15+15)
MatchHard_32K                 2.74ms × (0.99,1.01)  2.75ms × (0.99,1.01)    ~    (p=0.310 n=15+14)
MatchHard_1M                  87.1ms × (0.99,1.02)  88.2ms × (0.99,1.01)  +1.36% (p=0.000 n=14+15)
MatchHard_32M                  2.79s × (0.99,1.02)   2.83s × (0.99,1.02)  +1.26% (p=0.004 n=15+14)

go1 benchmarks

name                   old time/op    new time/op    delta
BinaryTree17              3.34s ± 3%     3.28s ± 2%  -1.86%  (p=0.000 n=67+66)
Fannkuch11                2.50s ± 1%     2.51s ± 1%  +0.24%  (p=0.016 n=63+66)
FmtFprintfEmpty          50.3ns ± 1%    50.2ns ± 2%  -0.30%  (p=0.001 n=62+67)
FmtFprintfString          178ns ± 1%     166ns ± 1%  -7.10%  (p=0.000 n=62+59)
FmtFprintfInt             168ns ± 1%     161ns ± 2%  -4.41%  (p=0.000 n=66+64)
FmtFprintfIntInt          292ns ± 1%     282ns ± 2%  -3.55%  (p=0.000 n=62+60)
FmtFprintfPrefixedInt     245ns ± 2%     239ns ± 2%  -2.24%  (p=0.000 n=66+65)
FmtFprintfFloat           338ns ± 2%     326ns ± 1%  -3.42%  (p=0.000 n=64+59)
FmtManyArgs              1.14µs ± 1%    1.10µs ± 2%  -3.55%  (p=0.000 n=62+62)
GobDecode                8.88ms ± 2%    8.74ms ± 1%  -1.55%  (p=0.000 n=66+62)
GobEncode                6.84ms ± 2%    6.61ms ± 2%  -3.32%  (p=0.000 n=61+67)
Gzip                      356ms ± 2%     352ms ± 2%  -1.07%  (p=0.000 n=67+66)
Gunzip                   90.6ms ± 2%    89.8ms ± 1%  -0.83%  (p=0.000 n=65+64)
HTTPClientServer         82.6µs ± 2%    82.5µs ± 2%    ~     (p=0.832 n=65+63)
JSONEncode               17.5ms ± 2%    16.8ms ± 2%  -3.77%  (p=0.000 n=63+63)
JSONDecode               63.3ms ± 2%    59.0ms ± 2%  -6.85%  (p=0.000 n=64+63)
Mandelbrot200            3.85ms ± 1%    3.85ms ± 1%    ~     (p=0.127 n=65+62)
GoParse                  3.75ms ± 2%    3.66ms ± 2%  -2.39%  (p=0.000 n=66+64)
RegexpMatchEasy0_32       100ns ± 2%     100ns ± 1%  -0.65%  (p=0.000 n=62+64)
RegexpMatchEasy0_1K       342ns ± 1%     341ns ± 1%  -0.43%  (p=0.000 n=65+64)
RegexpMatchEasy1_32      82.8ns ± 2%    82.8ns ± 2%    ~     (p=0.977 n=63+64)
RegexpMatchEasy1_1K       511ns ± 2%     506ns ± 2%  -1.01%  (p=0.000 n=63+64)
RegexpMatchMedium_32      139ns ± 1%     134ns ± 3%  -3.27%  (p=0.000 n=59+60)
RegexpMatchMedium_1K     41.8µs ± 2%    40.5µs ± 2%  -3.05%  (p=0.000 n=62+64)
RegexpMatchHard_32       2.13µs ± 1%    2.09µs ± 1%  -2.22%  (p=0.000 n=60+65)
RegexpMatchHard_1K       64.4µs ± 3%    62.8µs ± 2%  -2.58%  (p=0.000 n=65+59)
Revcomp                   531ms ± 2%     529ms ± 1%  -0.28%  (p=0.022 n=61+61)
Template                 73.2ms ± 1%    73.1ms ± 1%    ~     (p=0.794 n=66+63)
TimeParse                 369ns ± 1%     352ns ± 1%  -4.68%  (p=0.000 n=65+66)
TimeFormat                374ns ± 2%     348ns ± 2%  -7.01%  (p=0.000 n=66+64)

Change-Id: Ib190b5bb48a3e9087711d9e3383621d3103dd342
Reviewed-on: https://go-review.googlesource.com/10367
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoencoding/xml: Reset the parent stack before printing a chardata or comment field...
Hajime Hoshi [Sat, 9 May 2015 19:22:11 +0000 (04:22 +0900)]
encoding/xml: Reset the parent stack before printing a chardata or comment field in a struct

This CL resets the parent stack when printing a character or comment field struct.
In the case of XML elements, the previous parents stack must be considered. However,
charadata or comment fields can't be printed in other fields so it seems required to reset
the parent stack each time a chardata or comment field is printed.

Fixes #5072

Change-Id: I84f61c9bfce94133cd0c076c11211b9be5b4b1ac
Reviewed-on: https://go-review.googlesource.com/9910
Reviewed-by: Nigel Tao <nigeltao@golang.org>
Reviewed-by: roger peppe <rogpeppe@gmail.com>
9 years agocmd/go: clean up after 'go build' during 'go install'
Russ Cox [Mon, 20 Apr 2015 04:37:37 +0000 (00:37 -0400)]
cmd/go: clean up after 'go build' during 'go install'

If 'go install' (with no arguments, meaning the current directory)
succeeds, remove the executable written by 'go build', if present.
This avoids leaving a stale binary behind during a sequence like:

go build
<test, mostly works, make small change>
go install

Before this CL, the current directory still has the stale binary
from 'go build'. If $PATH contains dot, running the name of
the program will find this stale binary instead of the new,
installed one.

Remove the 'go build' target during 'go install', both to clean
up the directory and to avoid accidentally running the stale binary.

Another way to view this CL is that it makes the go command
behave as if 'go install' is implemented by 'go build' followed by
moving the resulting binary to the install location.

See #9645 for discussion and objections.

Fixes #9645.

Change-Id: Ide109572f96bbb5a35be45dda17738317462a7d4
Reviewed-on: https://go-review.googlesource.com/10682
Reviewed-by: Rob Pike <r@golang.org>
9 years agocmd/go: always rebuild GOPATH code that looks out of date
Russ Cox [Mon, 20 Apr 2015 15:07:29 +0000 (11:07 -0400)]
cmd/go: always rebuild GOPATH code that looks out of date

We used to put a rebuilding barrier between GOPATHs, so that if
you had GOPATH=dir1:dir2 and you had "p" in dir1/src/p
and "q" in dir2/src/q, with "p" importing "q", then when you
ran 'go install p', it would see that it was working in dir1
and (since nothing from dir2 was explicitly mentioned)
would assume that everything in dir2 is up-to-date, provided
it is built at all.

This has the confusing behavior that if "q" hasn't been built ever,
then if you update sources in q and run 'go install p', the right
thing happens (q is rebuilt and then p), but after that, if you update
sources in q and run 'go install p', nothing happens: the installed
q is assumed up-to-date.

People using code conventions with multiple GOPATH entries
(for example, with commands in one place and libraries in another,
or vendoring conventions that try to avoid rewriting import paths)
run into this without realizing it and end up with incorrect build
results.

The original motivation here was to avoid rebuild standard packages
since a system-installed GOROOT might be unwritable.
The change introduced to separate GOROOT also separated
individual GOPATH entries. Later changes added a different, more
aggressive earlier shortcut for GOROOT in release settings,
so the code here is now only applying to (and confusing)
multiple GOPATH entries. Remove it.

Fixes #10509.

Change-Id: I687a3baa81eff4073b0d67f9acbc5a3ab192eda5
Reviewed-on: https://go-review.googlesource.com/9155
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/go: detect when package or binary is stale due to removed source file
Russ Cox [Mon, 20 Apr 2015 04:00:45 +0000 (00:00 -0400)]
cmd/go: detect when package or binary is stale due to removed source file

The go command uses file modification times to decide when a
package is out of date: if the .a file is older than a source file,
the .a file needs to be rebuilt. This scheme breaks down when
multiple source files compile into a single .a file: if one source file
is removed but no other changes are made, there is no indication
that the .a file is out of date.

The fix is to store a value called a build ID in the package archive itself.
The build ID is a hash of the names of all source files compiled into the package.
A later go command can read the build ID out of the package archive
and compare to the build ID derived from the list of source files it now
sees in the directory. If the build IDs differ, the file list has changed,
and the package must be rebuilt.

There is a cost here: when scanning a package directory, in addition
to reading the beginning of every source file for build tags and imports,
the go command now also reads the beginning of the associated
package archive, for the build ID. This is at most a doubling in the
number of files read. On my 2012 MacBook Pro, the time for
'go list std' increases from about 0.215 seconds to about 0.23 seconds.

For executable binaries, the approach is the same except that the
build ID information is stored in a trailer at the end of the executable file.
It remains to be seen if anything objects to the trailer.
I don't expect problems except maybe on Plan 9.

Fixes #3895.

Change-Id: I21b4ebf5890c1a39e4a013eabe1ddbb5f3510c04
Reviewed-on: https://go-review.googlesource.com/9154
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/compile: merge Node.Opt and Node.Val behind access methods
Russ Cox [Wed, 27 May 2015 04:47:05 +0000 (00:47 -0400)]
cmd/compile: merge Node.Opt and Node.Val behind access methods

$ sizeof -p cmd/compile/internal/gc Node
Node 144
$

Change-Id: I688e3790964fe42f48c19f697ec38094a92fe1c1
Reviewed-on: https://go-review.googlesource.com/10531
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/compile: cleanups for Node trimming sequence
Russ Cox [Wed, 3 Jun 2015 18:16:01 +0000 (14:16 -0400)]
cmd/compile: cleanups for Node trimming sequence

Suggested during code reviews of last 15 CLs (or so).

Change-Id: If780f6eb47a7a31df133c64d5dcf0eaf04d8447b
Reviewed-on: https://go-review.googlesource.com/10675
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agocmd/link: make stkcheck more flexible
Josh Bleecher Snyder [Fri, 29 May 2015 00:23:58 +0000 (17:23 -0700)]
cmd/link: make stkcheck more flexible

stkcheck is flow-insensitive: It processes calls in PC order.
Since morestack was always the first call in a function,
it was a safe, conservative approximation to simply adjust stack
space as we went, recognizing morestack when it showed up.

Subsequent CLS will rearrange the function prologue;
morestack may no longer be the first call in a function.

Introducing flow-sensitivity to stkcheck would allow this,
and possibly allow a smaller stackguard.
It is also a high risk change and possibly expensive.

Instead, assume that all calls to morestack occur as
part of the function prologue, no matter where they
are located in the program text.

Updates #10587.

Change-Id: I4dcdd4256a980fc4bc433a68a10989ff57f7034f
Reviewed-on: https://go-review.googlesource.com/10496
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/internal/gc: accept map literals with omitted key type
Russ Cox [Tue, 19 May 2015 20:49:47 +0000 (16:49 -0400)]
cmd/internal/gc: accept map literals with omitted key type

Fixes #10209.

Change-Id: I248434f9195c868befd1ed8a6000a9cac72d1df8
Reviewed-on: https://go-review.googlesource.com/10263
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agoruntime: fix TestFixedGOROOT to properly restore the GOROOT env var after test
Brad Fitzpatrick [Wed, 3 Jun 2015 23:16:32 +0000 (16:16 -0700)]
runtime: fix TestFixedGOROOT to properly restore the GOROOT env var after test

Otherwise subsequent tests won't see any modified GOROOT.

With this CL I can move my GOROOT, set GOROOT to the new location, and
the runtime tests pass. Previously the crash_tests would instead look
for the GOROOT baked into the binary, instead of the env var:

--- FAIL: TestGcSys (0.01s)
        crash_test.go:92: building source: exit status 2
                go: cannot find GOROOT directory: /home/bradfitz/go
--- FAIL: TestGCFairness (0.01s)
        crash_test.go:92: building source: exit status 2
                go: cannot find GOROOT directory: /home/bradfitz/go
--- FAIL: TestGdbPython (0.07s)
        runtime-gdb_test.go:64: building source exit status 2
                go: cannot find GOROOT directory: /home/bradfitz/go
--- FAIL: TestLargeStringConcat (0.01s)
        crash_test.go:92: building source: exit status 2
                go: cannot find GOROOT directory: /home/bradfitz/go

Update #10029

Change-Id: If91be0f04d3acdcf39a9e773a4e7905a446bc477
Reviewed-on: https://go-review.googlesource.com/10685
Reviewed-by: Andrew Gerrand <adg@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

9 years agosrc/cmd/compile/internal/gc: re-vendor math/big, manually adjust mparith3.go
Robert Griesemer [Wed, 3 Jun 2015 22:11:05 +0000 (15:11 -0700)]
src/cmd/compile/internal/gc: re-vendor math/big, manually adjust mparith3.go

The only unreviewed change is in mparith3.go.

Change-Id: Iec0885e7688981cbaed04c152dc9b1c7032677e6
Reviewed-on: https://go-review.googlesource.com/10665
Reviewed-by: Alan Donovan <adonovan@google.com>
Run-TryBot: Robert Griesemer <gri@golang.org>

9 years agomath/big: turn off debug mode
Robert Griesemer [Wed, 3 Jun 2015 22:05:07 +0000 (15:05 -0700)]
math/big: turn off debug mode

Change-Id: Ie38742cddc5a256e2f0fc0f720c0ed2f1b2e1bca
Reviewed-on: https://go-review.googlesource.com/10664
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agomath/big: trim trailing zeros before hex printing
Dave Cheney [Tue, 2 Jun 2015 03:08:59 +0000 (13:08 +1000)]
math/big: trim trailing zeros before hex printing

m was being resliced as the result of looking for the first
non zero word of the mantissa, however m was not used later
in printing.

Spotted by Gordon Klaus, https://groups.google.com/forum/#!topic/golang-nuts/MdDLbvOjb4o

Change-Id: Ifbebb51ea5e0d86cb8e0422eb184b8634639a733
Reviewed-on: https://go-review.googlesource.com/10604
Reviewed-by: Robert Griesemer <gri@golang.org>
9 years agocmd/internal/gc: add -buildid flag to write build ID into object file
Russ Cox [Mon, 20 Apr 2015 03:56:30 +0000 (23:56 -0400)]
cmd/internal/gc: add -buildid flag to write build ID into object file

The build ID is an opaque token supplied by the build system.
The compiler writes it out early in the Go export metadata
(the second line), in a way that does not bother existing readers.

The intent is that the go command can use this to store information
about the sources for the generated code, so that it can detect
stale packages even in cases (like removed files) where mtimes fail.

Change-Id: Ib5082515d6cde8a07a8d4b5c69d1e8e4190cb5e1
Reviewed-on: https://go-review.googlesource.com/9153
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/dist: one more file for CL 9152
Russ Cox [Wed, 3 Jun 2015 20:41:00 +0000 (16:41 -0400)]
cmd/dist: one more file for CL 9152

This one didn't get written out.

Change-Id: Iee173861fb4dc7cafa64ba5f601f4664b6e8da4e
Reviewed-on: https://go-review.googlesource.com/10681
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/dist: do not rewrite files that already have the correct content
Russ Cox [Mon, 20 Apr 2015 15:41:31 +0000 (11:41 -0400)]
cmd/dist: do not rewrite files that already have the correct content

In particular, this avoids moving the mtime on runtime/zversion.go
forward unless the file is out of date. In turn, this makes cross compiles
that run dist multiple times coexist nicely.

(It's no longer necessary to run dist multiple times to set up cross compiles,
but people still might, and it's easy to fix regardless.)

Fixes #4749.

Change-Id: Id430525f168f106bc4b821ca74b2ca498a748f14
Reviewed-on: https://go-review.googlesource.com/9152
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/dist: fix build tag parser
Russ Cox [Mon, 20 Apr 2015 03:55:47 +0000 (23:55 -0400)]
cmd/dist: fix build tag parser

It was mishandling conjunctions containing negations.

Change-Id: Ife571b28416870ba2ceadbdac5ecb4670432bba1
Reviewed-on: https://go-review.googlesource.com/9151
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/go: make test.bash pass again
Russ Cox [Mon, 20 Apr 2015 17:34:22 +0000 (13:34 -0400)]
cmd/go: make test.bash pass again

Fixes #8809.

Change-Id: Id443fd406e9c611d5dfabc71a98eb71d1cc0972c
Reviewed-on: https://go-review.googlesource.com/9150
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agodoc/go1.5.txt: add note about internal
Russ Cox [Wed, 3 Jun 2015 20:28:45 +0000 (16:28 -0400)]
doc/go1.5.txt: add note about internal

Change-Id: I6476284a2cf80d50bd0a57fd9a0de9bc74273c7e
Reviewed-on: https://go-review.googlesource.com/10680
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/go: expand 'internal' rule to all repositories
Russ Cox [Mon, 20 Apr 2015 17:45:56 +0000 (13:45 -0400)]
cmd/go: expand 'internal' rule to all repositories

See golang.org/s/go14internal.

Fixes #10479.

Change-Id: I8c657dadeb5f10af060f22dedc15c1af989d4519
Reviewed-on: https://go-review.googlesource.com/9156
Reviewed-by: Rob Pike <r@golang.org>
9 years agocmd/compile: remove Node.Nname, now unused
Russ Cox [Wed, 27 May 2015 14:44:43 +0000 (10:44 -0400)]
cmd/compile: remove Node.Nname, now unused

$ sizeof -p cmd/compile/internal/gc Node
Node 160
$

Change-Id: Ib0bd4230c8547f36972b2a9d81ba3eca81496e39
Reviewed-on: https://go-review.googlesource.com/10537
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agotext/template: refactor code to accomodate bi-state requirement for templates
Aamir Khan [Wed, 3 Jun 2015 03:51:56 +0000 (12:51 +0900)]
text/template: refactor code to accomodate bi-state requirement for templates

This is follow-up to CL10607

- Refactor AddParseTree() to use t.associate()
- Refactor Parse() to use AddParseTree() to put entries into common structure
- Clone() should not put entry in t.tmpl for undefined template
- Clarify documentation for Templates()
- Clarify documentation for AddParseTree() to include the error case

Updates #10910
Uodates #10926

Includes test cases for most of the above changes

Change-Id: I25b2fce6f9651272866f881acf44e4dbca04a4a8
Reviewed-on: https://go-review.googlesource.com/10622
Reviewed-by: Rob Pike <r@golang.org>
Run-TryBot: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/compile: remove use of Node.Nname in initplan
Russ Cox [Wed, 27 May 2015 14:44:23 +0000 (10:44 -0400)]
cmd/compile: remove use of Node.Nname in initplan

$ sizeof -p cmd/compile/internal/gc Node
Node 168
$

Change-Id: Ie7e2cee8cec101560bd5dd013b23969278f89b12
Reviewed-on: https://go-review.googlesource.com/10536
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agocmd/compile: move OCASE/OXCASE Node.Nname into Node.Rlist (type switch variable)
Russ Cox [Wed, 27 May 2015 14:43:53 +0000 (10:43 -0400)]
cmd/compile: move OCASE/OXCASE Node.Nname into Node.Rlist (type switch variable)

$ sizeof -p cmd/compile/internal/gc Node
Node 168
$

Change-Id: I22bcea8099f308298c9db75c937f35e7fca906f1
Reviewed-on: https://go-review.googlesource.com/10535
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agocmd/compile: move ODCLFUNC Node.Nname into Node.Func.Nname
Russ Cox [Wed, 27 May 2015 14:42:55 +0000 (10:42 -0400)]
cmd/compile: move ODCLFUNC Node.Nname into Node.Func.Nname

$ sizeof -p cmd/compile/internal/gc Node
Node 168
$

Change-Id: I7decd950fe068c0f294c6c9bff07ef809c394429
Reviewed-on: https://go-review.googlesource.com/10534
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agocmd/compile, cmd/internal: replace %P with %v in a few final prints
Russ Cox [Wed, 27 May 2015 14:40:47 +0000 (10:40 -0400)]
cmd/compile, cmd/internal: replace %P with %v in a few final prints

$ sizeof -p cmd/compile/internal/gc Node
Node 168
$

Change-Id: I1e2f17bfee0e6ca7213706c4cef8d990f4461915
Reviewed-on: https://go-review.googlesource.com/10533
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agocmd/compile: move Node.Curfn into both Node.Func and Node.Name
Russ Cox [Wed, 27 May 2015 11:31:56 +0000 (07:31 -0400)]
cmd/compile: move Node.Curfn into both Node.Func and Node.Name

$ sizeof -p cmd/compile/internal/gc Node
Node 168
$

Change-Id: If624a2d72ec04ef30a1bc7ce76c0d61a526d8a37
Reviewed-on: https://go-review.googlesource.com/10532
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/compile: move Node.Param, Node.Funcdepth into Node.Name; remove Node.Walkgen
Russ Cox [Wed, 27 May 2015 04:44:05 +0000 (00:44 -0400)]
cmd/compile: move Node.Param, Node.Funcdepth into Node.Name; remove Node.Walkgen

$ sizeof -p cmd/compile/internal/gc Node
Node 176
$

Change-Id: Ibf1ab531a60d4af8a0c242c0e504f4fd50cd5b36
Reviewed-on: https://go-review.googlesource.com/10530
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agocmd/compile: move Node.Vargen, Node.Iota into Node.Name
Russ Cox [Wed, 27 May 2015 03:56:14 +0000 (23:56 -0400)]
cmd/compile: move Node.Vargen, Node.Iota into Node.Name

$ sizeof -p cmd/compile/internal/gc Node
Node 192
$

Change-Id: I8f0c1a3cc2bf9c8eff02bbd8d061ff98affc9eb0
Reviewed-on: https://go-review.googlesource.com/10529
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agocmd/compile: remove Node.Escflowsrc, Node.Escretval, Node.Escloopdepth, Node.Esclevel
Russ Cox [Wed, 27 May 2015 03:42:41 +0000 (23:42 -0400)]
cmd/compile: remove Node.Escflowsrc, Node.Escretval, Node.Escloopdepth, Node.Esclevel

$ sizeof -p cmd/compile/internal/gc Node
Node 200
$

Change-Id: Iba4e88eac6bee3e2349e818a5a2326deabcb96f9
Reviewed-on: https://go-review.googlesource.com/10528
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agocmd/compile: move Node.Pkg to Node.Name.Pkg
Russ Cox [Wed, 27 May 2015 03:18:27 +0000 (23:18 -0400)]
cmd/compile: move Node.Pkg to Node.Name.Pkg

$ sizeof -p cmd/compile/internal/gc Node
Node 224
$

Change-Id: Id0969e8df99c43a5f6f8d77a38f20a71a467e7c6
Reviewed-on: https://go-review.googlesource.com/10527
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agocmd/compile: move Node.Pack to Node.Name.Pack
Russ Cox [Wed, 27 May 2015 03:08:39 +0000 (23:08 -0400)]
cmd/compile: move Node.Pack to Node.Name.Pack

$ sizeof -p cmd/compile/internal/gc Node
Node 232
$

Change-Id: I4be025f4ec11f882f24ae7582821d36d3b122b77
Reviewed-on: https://go-review.googlesource.com/10526
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agostrings: mention UTF-8 in the package comment.
Rob Pike [Wed, 3 Jun 2015 19:21:06 +0000 (12:21 -0700)]
strings: mention UTF-8 in the package comment.

Also add a reference to the strings blog post.

Fixes #11045.

Change-Id: Ic0a8908cbd7b51a36d104849fa0e8abfd54de2b9
Reviewed-on: https://go-review.googlesource.com/10662
Reviewed-by: Andrew Gerrand <adg@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agodoc: move pre-go1 release notes to separate file
Andrew Gerrand [Wed, 27 May 2015 20:10:43 +0000 (13:10 -0700)]
doc: move pre-go1 release notes to separate file

Fixes #10924

Change-Id: I0caf5f8f82682ee48e95270d43328550bfd8b6e1
Reviewed-on: https://go-review.googlesource.com/10440
Reviewed-by: Rob Pike <r@golang.org>
9 years agocmd/go: convert one test from test.bash to Go
Ian Lance Taylor [Thu, 28 May 2015 16:57:57 +0000 (09:57 -0700)]
cmd/go: convert one test from test.bash to Go

Sending out the conversion of a single test to get comments on the
overall approach.  Converting more tests will follow.

Change-Id: I4755442d08aeb6f74c46856ae406fec41cf8d5dc
Reviewed-on: https://go-review.googlesource.com/10464
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/compile: remove Node.Alloc
Russ Cox [Wed, 27 May 2015 03:05:35 +0000 (23:05 -0400)]
cmd/compile: remove Node.Alloc

$ sizeof -p cmd/compile/internal/gc Node
Node 240
$

Change-Id: Id12710c480ed4e0a5bf4f5006f6bd56ef91a2af1
Reviewed-on: https://go-review.googlesource.com/10525
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agocmd/compile: remove Val.Ctype
Russ Cox [Wed, 27 May 2015 02:50:45 +0000 (22:50 -0400)]
cmd/compile: remove Val.Ctype

$ sizeof -p cmd/compile/internal/gc Node
Node 248
$

Change-Id: I0fbfeb0d0b36e225eb282fce9e480a96ec1d278f
Reviewed-on: https://go-review.googlesource.com/10524
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agocmd/compile: move Node.Defn to Node.Name.Defn
Russ Cox [Wed, 27 May 2015 02:19:27 +0000 (22:19 -0400)]
cmd/compile: move Node.Defn to Node.Name.Defn

$ sizeof -p cmd/compile/internal/gc Node
Node 256
$

Change-Id: I89ac8bbe077664aa076092bfd096947e84c0624c
Reviewed-on: https://go-review.googlesource.com/10523
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agocmd/compile: move Node.Paramfld to Node.Param.Field
Russ Cox [Wed, 27 May 2015 01:49:31 +0000 (21:49 -0400)]
cmd/compile: move Node.Paramfld to Node.Param.Field

$ sizeof -p cmd/compile/internal/gc Node
Node 264
$

Change-Id: I5c90089dcf5df51c874250f28a1bc3ec32f764b9
Reviewed-on: https://go-review.googlesource.com/10522
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agocmd/compile: remove Node.Ntest, Node.Stkdelta
Russ Cox [Wed, 27 May 2015 01:30:20 +0000 (21:30 -0400)]
cmd/compile: remove Node.Ntest, Node.Stkdelta

$ sizeof -p cmd/compile/internal/gc Node
Node 272
$

Change-Id: I3d9b67eebfc0be0a4b9768d3de3dc76300abd89c
Reviewed-on: https://go-review.googlesource.com/10521
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agocmd/compile: remove Node.Nincr, Node.Nelse, Node.Initplan
Russ Cox [Fri, 22 May 2015 05:16:52 +0000 (01:16 -0400)]
cmd/compile: remove Node.Nincr, Node.Nelse, Node.Initplan

$ sizeof -p cmd/compile/internal/gc Node
Node 288
$

Change-Id: I4e316efa246132b3faa3a892e4fe9c9039250665
Reviewed-on: https://go-review.googlesource.com/10520
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agocmd/dist: enable extlink test on linux/arm64
Shenghou Ma [Wed, 3 Jun 2015 00:52:46 +0000 (20:52 -0400)]
cmd/dist: enable extlink test on linux/arm64

It was an oversight (but as linux/arm64 doesn't support internal
linking and always use external linking with cgo, no harm is done.)

Change-Id: Ie5f2b445cb67a8e63d6b868e63379c68847554f9
Reviewed-on: https://go-review.googlesource.com/10636
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agoruntime: print start of GC cycle in gctrace, rather than end
Austin Clements [Tue, 2 Jun 2015 18:48:12 +0000 (14:48 -0400)]
runtime: print start of GC cycle in gctrace, rather than end

Currently the GODEBUG=gctrace=1 trace line includes "@n.nnns" to
indicate the time that the GC cycle ended relative to the time the
program started. This was meant to be consistent with the utilization
as of the end of the cycle, which is printed next on the trace line,
but it winds up just being confusing and unexpected.

Change the trace line to include the time that the GC cycle started
relative to the time the program started.

Change-Id: I7d64580cd696eb17540716d3e8a74a9d6ae50650
Reviewed-on: https://go-review.googlesource.com/10634
Reviewed-by: Rick Hudson <rlh@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agotext/template: clarify the documentation around template definitions
Rob Pike [Tue, 2 Jun 2015 18:49:24 +0000 (11:49 -0700)]
text/template: clarify the documentation around template definitions

Due to the requirements of parsing template definitions that mention
other templates that are not yet defined, a Template can be in two states:
defined and undefined. Thus, although one calls New, the resulting
template has no definition even though it exists as a data structure.

Thus, for example, will return nil for a template that is named but not
yet defined.

Fixes #10910
Fixes #10926

Clarify the documentation a little to explain this,
Also tidy up the code a little and remove a spurious call to init.

Change-Id: I22cc083291500bca424e83dc12807e0de7b00b7a
Reviewed-on: https://go-review.googlesource.com/10641
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agoruntime: implement GC stack barriers
Austin Clements [Wed, 20 May 2015 20:30:49 +0000 (16:30 -0400)]
runtime: implement GC stack barriers

This commit implements stack barriers to minimize the amount of
stack re-scanning that must be done during mark termination.

Currently the GC scans stacks of active goroutines twice during every
GC cycle: once at the beginning during root discovery and once at the
end during mark termination. The second scan happens while the world
is stopped and guarantees that we've seen all of the roots (since
there are no write barriers on writes to local stack
variables). However, this means pause time is proportional to stack
size. In particularly recursive programs, this can drive pause time up
past our 10ms goal (e.g., it takes about 150ms to scan a 50MB heap).

Re-scanning the entire stack is rarely necessary, especially for large
stacks, because usually most of the frames on the stack were not
active between the first and second scans and hence any changes to
these frames (via non-escaping pointers passed down the stack) were
tracked by write barriers.

To efficiently track how far a stack has been unwound since the first
scan (and, hence, how much needs to be re-scanned), this commit
introduces stack barriers. During the first scan, at exponentially
spaced points in each stack, the scan overwrites return PCs with the
PC of the stack barrier function. When "returned" to, the stack
barrier function records how far the stack has unwound and jumps to
the original return PC for that point in the stack. Then the second
scan only needs to proceed as far as the lowest barrier that hasn't
been hit.

For deeply recursive programs, this substantially reduces mark
termination time (and hence pause time). For the goscheme example
linked in issue #10898, prior to this change, mark termination times
were typically between 100 and 500ms; with this change, mark
termination times are typically between 10 and 20ms. As a result of
the reduced stack scanning work, this reduces overall execution time
of the goscheme example by 20%.

Fixes #10898.

The effect of this on programs that are not deeply recursive is
minimal:

name                   old time/op    new time/op    delta
BinaryTree17              3.16s ± 2%     3.26s ± 1%  +3.31%  (p=0.000 n=19+19)
Fannkuch11                2.42s ± 1%     2.48s ± 1%  +2.24%  (p=0.000 n=17+19)
FmtFprintfEmpty          50.0ns ± 3%    49.8ns ± 1%    ~     (p=0.534 n=20+19)
FmtFprintfString          173ns ± 0%     175ns ± 0%  +1.49%  (p=0.000 n=16+19)
FmtFprintfInt             170ns ± 1%     175ns ± 1%  +2.97%  (p=0.000 n=20+19)
FmtFprintfIntInt          288ns ± 0%     295ns ± 0%  +2.73%  (p=0.000 n=16+19)
FmtFprintfPrefixedInt     242ns ± 1%     252ns ± 1%  +4.13%  (p=0.000 n=18+18)
FmtFprintfFloat           324ns ± 0%     323ns ± 0%  -0.36%  (p=0.000 n=20+19)
FmtManyArgs              1.14µs ± 0%    1.12µs ± 1%  -1.01%  (p=0.000 n=18+19)
GobDecode                8.88ms ± 1%    8.87ms ± 0%    ~     (p=0.480 n=19+18)
GobEncode                6.80ms ± 1%    6.85ms ± 0%  +0.82%  (p=0.000 n=20+18)
Gzip                      363ms ± 1%     363ms ± 1%    ~     (p=0.077 n=18+20)
Gunzip                   90.6ms ± 0%    90.0ms ± 1%  -0.71%  (p=0.000 n=17+18)
HTTPClientServer         51.5µs ± 1%    50.8µs ± 1%  -1.32%  (p=0.000 n=18+18)
JSONEncode               17.0ms ± 0%    17.1ms ± 0%  +0.40%  (p=0.000 n=18+17)
JSONDecode               61.8ms ± 0%    63.8ms ± 1%  +3.11%  (p=0.000 n=18+17)
Mandelbrot200            3.84ms ± 0%    3.84ms ± 1%    ~     (p=0.583 n=19+19)
GoParse                  3.71ms ± 1%    3.72ms ± 1%    ~     (p=0.159 n=18+19)
RegexpMatchEasy0_32       100ns ± 0%     100ns ± 1%  -0.19%  (p=0.033 n=17+19)
RegexpMatchEasy0_1K       342ns ± 1%     331ns ± 0%  -3.41%  (p=0.000 n=19+19)
RegexpMatchEasy1_32      82.5ns ± 0%    81.7ns ± 0%  -0.98%  (p=0.000 n=18+18)
RegexpMatchEasy1_1K       505ns ± 0%     494ns ± 1%  -2.16%  (p=0.000 n=18+18)
RegexpMatchMedium_32      137ns ± 1%     137ns ± 1%  -0.24%  (p=0.048 n=20+18)
RegexpMatchMedium_1K     41.6µs ± 0%    41.3µs ± 1%  -0.57%  (p=0.004 n=18+20)
RegexpMatchHard_32       2.11µs ± 0%    2.11µs ± 1%  +0.20%  (p=0.037 n=17+19)
RegexpMatchHard_1K       63.9µs ± 2%    63.3µs ± 0%  -0.99%  (p=0.000 n=20+17)
Revcomp                   560ms ± 1%     522ms ± 0%  -6.87%  (p=0.000 n=18+16)
Template                 75.0ms ± 0%    75.1ms ± 1%  +0.18%  (p=0.013 n=18+19)
TimeParse                 358ns ± 1%     364ns ± 0%  +1.74%  (p=0.000 n=20+15)
TimeFormat                360ns ± 0%     372ns ± 0%  +3.55%  (p=0.000 n=20+18)

Change-Id: If8a9bfae6c128d15a4f405e02bcfa50129df82a2
Reviewed-on: https://go-review.googlesource.com/10314
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoruntime: avoid double-scanning of stacks
Austin Clements [Thu, 28 May 2015 16:37:12 +0000 (12:37 -0400)]
runtime: avoid double-scanning of stacks

Currently there's a race between stopg scanning another G's stack and
the G reaching a preemption point and scanning its own stack. When
this race occurs, the G's stack is scanned twice. Currently this is
okay, so this race is benign.

However, we will shortly be adding stack barriers during the first
stack scan, so scanning will no longer be idempotent. To prepare for
this, this change ensures that each stack is scanned only once during
each GC phase by checking the flag that indicates that the stack has
been scanned in this phase before scanning the stack.

Change-Id: Id9f4d5e2e5b839bc3f200ec1723a4a12dd677ab4
Reviewed-on: https://go-review.googlesource.com/10458
Reviewed-by: Rick Hudson <rlh@golang.org>
9 years agoruntime: steal space for stack barrier tracking from stack
Austin Clements [Wed, 20 May 2015 20:16:04 +0000 (16:16 -0400)]
runtime: steal space for stack barrier tracking from stack

The stack barrier code will need a bookkeeping structure to keep track
of the overwritten return PCs. This commit introduces and allocates
this structure, but does not yet use the structure.

We don't want to allocate space for this structure during garbage
collection, so this commit allocates it along with the allocation of
the corresponding stack. However, we can't do a regular allocation in
newstack because mallocgc may itself grow the stack (which would lead
to a recursive allocation). Hence, this commit makes the bookkeeping
structure part of the stack allocation itself by stealing the
necessary space from the top of the stack allocation. Since the size
of this bookkeeping structure is logarithmic in the size of the stack,
this has minimal impact on stack behavior.

Change-Id: Ia14408be06aafa9ca4867f4e70bddb3fe0e96665
Reviewed-on: https://go-review.googlesource.com/10313
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoruntime: decouple stack bounds and stack allocation size
Austin Clements [Wed, 20 May 2015 19:29:53 +0000 (15:29 -0400)]
runtime: decouple stack bounds and stack allocation size

Currently the runtime assumes that the allocation for the stack is
exactly [stack.lo, stack.hi). We're about to steal a small part of
this allocation for per-stack GC metadata. To prepare for this, this
commit adds a field to the G for the allocated size of the stack.
With this change, stack.lo and stack.hi continue to act as the true
bounds on the stack, but are no longer also used as the bounds on the
stack allocation.

(I also tried this the other way around, where stack.lo and stack.hi
remained the allocation bounds and I introduced a new top of stack.
However, there are far more places that assume stack.hi is the true
top of the stack than there are places that assume it's the top of the
allocation.)

Change-Id: Ifa9d956753be53d286d09cbc73d47fb34a18c0c6
Reviewed-on: https://go-review.googlesource.com/10312
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoruntime: clean up signalstack API
Austin Clements [Thu, 21 May 2015 18:12:29 +0000 (14:12 -0400)]
runtime: clean up signalstack API

Currently signalstack takes a lower limit and a length and all calls
hard-code the passed length. Change the API to take a *stack and
compute the lower limit and length from the passed stack.

This will make it easier for the runtime to steal some space from the
top of the stack since it eliminates the hard-coded stack sizes.

Change-Id: I7d2a9f45894b221f4e521628c2165530bbc57d53
Reviewed-on: https://go-review.googlesource.com/10311
Reviewed-by: Rick Hudson <rlh@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoruntime: increase precision of gctrace times
Austin Clements [Sun, 31 May 2015 01:47:00 +0000 (21:47 -0400)]
runtime: increase precision of gctrace times

Currently we truncate gctrace clock and CPU times to millisecond
precision. As a result, many phases are typically printed as 0, which
is fine for user consumption, but makes gathering statistics and
reports over GC traces difficult.

In 1.4, the gctrace line printed times in microseconds. This was
better for statistics, but not as easy for users to read or interpret,
and it generally made the trace lines longer.

This change strikes a balance between these extremes by printing
milliseconds, but including the decimal part to two significant
figures down to microsecond precision. This remains easy to read and
interpret, but includes more precision when it's useful.

For example, where the code currently prints,

gc #29 @1.629s 0%: 0+2+0+12+0 ms clock, 0+2+0+0/12/0+0 ms cpu, 4->4->2 MB, 4 MB goal, 1 P

this prints,

gc #29 @1.629s 0%: 0.005+2.1+0+12+0.29 ms clock, 0.005+2.1+0+0/12/0+0.29 ms cpu, 4->4->2 MB, 4 MB goal, 1 P

Fixes #10970.

Change-Id: I249624779433927cd8b0947b986df9060c289075
Reviewed-on: https://go-review.googlesource.com/10554
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agonet/http: set nosniff header when serving Error
Andrew Gerrand [Tue, 2 Jun 2015 18:01:56 +0000 (11:01 -0700)]
net/http: set nosniff header when serving Error

The Error function is a potential XSS vector if a user can control the
error message.

For example, an http.FileServer when given a request for this path
/<script>alert("xss!")</script>
may return a response with a body like this
open <script>alert("xss!")</script>: no such file or directory
Browsers that sniff the content may interpret this as HTML and execute
the script. The nosniff header added by this CL should help, but we
should also try santizing the output entirely.

Change-Id: I447f701531329a2fc8ffee2df2f8fa69d546f893
Reviewed-on: https://go-review.googlesource.com/10640
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agofmt: fix buffer underflow for negative integers
Håvard Haugen [Mon, 25 May 2015 21:14:35 +0000 (23:14 +0200)]
fmt: fix buffer underflow for negative integers

Allow room for the initial minus sign of negative integers when
computing widths.

Fixes #10945.

Change-Id: I04d80203aaff64611992725d613ec13ed2ae721f
Reviewed-on: https://go-review.googlesource.com/10393
Reviewed-by: Rob Pike <r@golang.org>
9 years agonet: fix parsing literal IP addresses in local database
Mikio Hara [Wed, 20 May 2015 14:20:43 +0000 (23:20 +0900)]
net: fix parsing literal IP addresses in local database

This change fixes incorrect parsing of literal IP addresses in local
database when the addresses contain IPv6 zone identifiers, are in
dotted-decimal notation or in colon-hexadecimal notation with leading
zeros.

https://golang.org/cl/5851 already fixed the code path using getaddrinfo
via cgo. This change fixes the remaining non-cgo code path.

Fixes #8243.
Fixes #8996.

Change-Id: I48443611cbabed0d69667cc73911ba3de396fd44
Reviewed-on: https://go-review.googlesource.com/10306
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agoruntime: fix data race in BenchmarkChanPopular
Mikio Hara [Mon, 1 Jun 2015 04:25:03 +0000 (13:25 +0900)]
runtime: fix data race in BenchmarkChanPopular

Fixes #11014.

Change-Id: I9a18dacd10564d3eaa1fea4d77f1a48e08e79f53
Reviewed-on: https://go-review.googlesource.com/10563
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agomath/big: implement fmt.Formatter-compatible (*Float).Format
Robert Griesemer [Fri, 29 May 2015 01:17:38 +0000 (18:17 -0700)]
math/big: implement fmt.Formatter-compatible (*Float).Format

Change-Id: I22fdba8ecaecf4e9201b845e65d982cac09f254a
Reviewed-on: https://go-review.googlesource.com/10499
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agonet: fix undetected set and not used error
Dave Cheney [Tue, 2 Jun 2015 01:04:45 +0000 (11:04 +1000)]
net: fix undetected set and not used error

Fixes an error where the compiler did not spot that the shadowed err
value was set again after being read. That second assignment was lost
as the value was redeclared in an inner scope.

Spotted by Gordon Klass, https://groups.google.com/forum/#!topic/golang-nuts/MdDLbvOjb4o

Change-Id: I28f2da6f98c52afcbb45e17d2b4f36c586598f98
Reviewed-on: https://go-review.googlesource.com/10600
Reviewed-by: Mikio Hara <mikioh.mikioh@gmail.com>
9 years agocmd/pprof/internal/profile: fix missing error check
Dave Cheney [Tue, 2 Jun 2015 03:28:33 +0000 (13:28 +1000)]
cmd/pprof/internal/profile: fix missing error check

Fix missing error check.

Spotted by Gordon Klass, https://groups.google.com/forum/#!topic/golang-nuts/MdDLbvOjb4o

Change-Id: I453a0cf032e0077d2622d5b85030310d159b9c4b
Reviewed-on: https://go-review.googlesource.com/10606
Reviewed-by: Nigel Tao <nigeltao@golang.org>
9 years agocmd/cover: fix missing error check in test
Dave Cheney [Tue, 2 Jun 2015 03:21:37 +0000 (13:21 +1000)]
cmd/cover: fix missing error check in test

Fix missing error check in test.

Spotted by Gordon Klass, https://groups.google.com/forum/#!topic/golang-nuts/MdDLbvOjb4o

Change-Id: I22f1a438cbb60a2fe1740fc2d43fbf8aa008b6de
Reviewed-on: https://go-review.googlesource.com/10605
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>