cmd/go: fix a typo.

Change-Id: Ic453da17817f66e1073f6ba740b830b8daf2fc38
Reviewed-on: https://go-review.googlesource.com/8829
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

debug/dwarf: add Entry.AttrField method to get *Field by Attr

Currently, Entry has a Val method that looks up an attribute and
returns its value. Now that Field has more fields than the attribute
and its value, it's useful to return the whole Field and let the
caller retrieve the parts it needs.

This change adds an AttrField method to Entry that does the same
lookup at Val, but returns the whole *Field rather than just the
value.

Change-Id: Ic629744c14c0e09d7528fa1026b0e1857789948c
Reviewed-on: https://go-review.googlesource.com/8503
Reviewed-by: Ian Lance Taylor <iant@golang.org>

debug/dwarf: add DWARF attribute value class to Field

To return DWARF attribute values, debug/dwarf maps the DWARF attribute
value classes to Go types. Unfortunately, this mapping is ambiguous in
a way that makes it impossible to correctly interpret some DWARF
attributes as of DWARF 4. For example, AttrStartScope can be either a
constant or a rangelistptr. The attribute is interpreted differently
depending on its class, but debug/dwarf maps both classes to int64, so
the caller can't distinguish them from the Go type.
AttrDataMemberLocation is similar.

To address this, this change adds a field to type Field that indicates
the exact DWARF attribute value class of that field's value. This
makes it possible to distinguish value classes that can't be
distinguished by their Go type alone.

The root of this type ambiguity was DWARF itself. For example, DWARF 2
made no distinction between constants that were just constants and
constants that were section offsets because no attribute could have
both meanings. Hence, the single int64 type was sufficient. To avoid
introducing just another layer of ambiguity, this change takes pains
to canonicalize ambiguous classes in DWARF 2 and 3 files into the
unambiguous classes of DWARF 4.

Of course, there's no guarantee that future DWARF versions won't do
the same thing again and further subdivide the DWARF 4 classes. This
change gets ahead of this somewhat by distinguishing the various *ptr
classes even though the encoding does not. If there's some other form
of split, we can handle this in a backwards-compatible way by
introducing, for example, a Class5 field and type.

Change-Id: I4ef96d1223b0fd7f96ecf44fcc0e704a36af02b4
Reviewed-on: https://go-review.googlesource.com/8502
Reviewed-by: Ian Lance Taylor <iant@golang.org>

cmd/interal/ld: darwin c-archive buildmode support

Uses ar to create an archive when -buildmode=c-archive.

A small example (that I hope to turn into a test in a later CL):

goarchive.go:
package main

import "fmt"

import "C"

func init() {
fmt.Println("ran go init")
}

//export FuncInGo
func FuncInGo() {
fmt.Println("called a go function")
}

func main() {
fmt.Println("in main")
}

This can be compiled with:

go build -ldflags=-buildmode=c-archive -o=libgo.a goarchive.go

main.c:

#include <stdio.h>

extern void FuncInGo();

int main(void) {
printf("c hello\n");
FuncInGo();
printf("c goodbye\n");
return 0;
}

Can be compiled with:

cc main.c libgo.a

Apple provide a warning about the lack of PIE, but still produce a
binary which runs and outputs (on darwin/amd64):

c hello
ran go init
called a go function
c goodbye

Change-Id: I7611925f210a83afa6bd1e66a5601dd636a428c8
Reviewed-on: https://go-review.googlesource.com/8711
Reviewed-by: Ian Lance Taylor <iant@golang.org>

runtime: library entry point for darwin/arm

Tested by using -buildmode=c-archive to generate an archive, add it
to an Xcode project and calling a Go function from an iOS app. (I'm
still investigating proper buildmode tests for all.bash.)

Change-Id: I7890df15246df8e90ad27837b8d64ba2cde409fe
Reviewed-on: https://go-review.googlesource.com/8719
Reviewed-by: Ian Lance Taylor <iant@golang.org>

cmd/addr2line: skip fork test on darwin/arm64

Just like darwin/arm.

Change-Id: Ia84662f58f6b1bb168cce8a9837945b1cbd175e1
Reviewed-on: https://go-review.googlesource.com/8828
Reviewed-by: Minux Ma <minux@golang.org>

cmd/internal/obj/x86: skip test on darwin/arm64

Just like darwin/arm, cannot fork..

Change-Id: If565afbceb79013b9e3103e1e28d93691e9fc0a5
Reviewed-on: https://go-review.googlesource.com/8826
Reviewed-by: Minux Ma <minux@golang.org>

syscall: skip fork test on darwin/arm64

Just like darwin/arm.

Change-Id: Iadc30b7307ae56fd4f8a681d49672bed7ca6966f
Reviewed-on: https://go-review.googlesource.com/8810
Reviewed-by: Minux Ma <minux@golang.org>

cmd/nm: skip fork test on darwin/arm64

Just like darwin/arm.

Change-Id: Iabb6282f18548da43117ee60f7ad6e272502f09d
Reviewed-on: https://go-review.googlesource.com/8825
Reviewed-by: Minux Ma <minux@golang.org>

os: adjust working dir for darwin/arm64 tests

Just like darwin/arm.

Change-Id: Ib64a3e8ff11249a20b0208bd3b900db318c682b7
Reviewed-on: https://go-review.googlesource.com/8817
Reviewed-by: Minux Ma <minux@golang.org>
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

sync/atomic: skip issue 7338 test on darwin/arm64

Similar to darwin/arm. This issue is quite worrying and I hope it
can be addressed for Go 1.5.

Change-Id: Ic095281d6a2e9a38a59973f58d464471db5a2edc
Reviewed-on: https://go-review.googlesource.com/8811
Reviewed-by: Minux Ma <minux@golang.org>

misc/ios: pick clang arch based on GOARCH

Change-Id: Ia49ab729747acb07bf392d90aea9e752471e152e
Reviewed-on: https://go-review.googlesource.com/8789
Reviewed-by: Minux Ma <minux@golang.org>

cmd/6l: support -linkshared

Change-Id: Id469165b1acd383837b1f4e1e6f961e10dfa5d61
Reviewed-on: https://go-review.googlesource.com/8332
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>

cmd/5g, cmd/internal/ld, cmd/internal/obj: destutter composite literals

While here, this changes DWAbbrev's attr field from a [30]DWAttrForm
with zero-termination to a simple []DWAttrForm, and updates its users
accordingly.

Passes "go build -toolexec 'toolstash -cmp' -a std" on linux/amd64.

Change-Id: I52b5f7a749bdb3e7588fc8ebdb8fee2cf8cab602
Reviewed-on: https://go-review.googlesource.com/8762
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Dave Cheney <dave@cheney.net>

cmd/go: fix error message

Change-Id: I440435927fc9f417d81ed4e8656a73787bf0968a
Reviewed-on: https://go-review.googlesource.com/8801
Reviewed-by: Ian Lance Taylor <iant@golang.org>

cmd/internal/ld: make a few more symbols local

The symbols for the actual data in a constant string or bytes literal should
be local.

Change-Id: Idafcfba9a638eaa4e460e5103d96843960559b35
Reviewed-on: https://go-review.googlesource.com/8772
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>

reflect, runtime: check equality, not identity, for method names

When dynamically linking Go code, it is no longer safe to assume that
strings that end up in method names are identical if they are equal.

The performance impact seems to be noise:

benchmark                    old ns/op     new ns/op     delta
BenchmarkAssertI2E2          13.3          13.1          -1.50%
BenchmarkAssertE2I           23.5          23.2          -1.28%
BenchmarkAssertE2E2Blank     0.83          0.82          -1.20%
BenchmarkConvT2ISmall        60.7          60.1          -0.99%
BenchmarkAssertI2T           10.2          10.1          -0.98%
BenchmarkAssertE2T           10.2          10.3          +0.98%
BenchmarkConvT2ESmall        56.7          57.2          +0.88%
BenchmarkConvT2ILarge        59.4          58.9          -0.84%
BenchmarkConvI2E             13.0          12.9          -0.77%
BenchmarkAssertI2E           13.4          13.3          -0.75%
BenchmarkConvT2IUintptr      57.9          58.3          +0.69%
BenchmarkConvT2ELarge        55.9          55.6          -0.54%
BenchmarkAssertI2I           23.8          23.7          -0.42%
BenchmarkConvT2EUintptr      55.4          55.5          +0.18%
BenchmarkAssertE2E           6.12          6.11          -0.16%
BenchmarkAssertE2E2          14.4          14.4          +0.00%
BenchmarkAssertE2T2          10.0          10.0          +0.00%
BenchmarkAssertE2T2Blank     0.83          0.83          +0.00%
BenchmarkAssertE2TLarge      10.7          10.7          +0.00%
BenchmarkAssertI2E2Blank     0.83          0.83          +0.00%
BenchmarkConvI2I             23.4          23.4          +0.00%

Change-Id: I0b3dfc314215a4d4e09eec6b42c1e3ebce33eb56
Reviewed-on: https://go-review.googlesource.com/8239
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/go: start support for -linkshared

This will fruitlessly rebuild stale packages that are in a shared
library.

Change-Id: I66a6e1adf7818558e7d1351ab215a5021b4a8a6b
Reviewed-on: https://go-review.googlesource.com/8333
Reviewed-by: Ian Lance Taylor <iant@golang.org>

api: update next.txt

Change-Id: Ibfb5a96995a954c8c1dc563c33bdddbdb4866425
Reviewed-on: https://go-review.googlesource.com/8765
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

go/types: move go/types/internal/gcimport => go/internal/gcimporter

This will make it possible to access the gcimporter (and gccgoimporter,
eventually) from the forthcoming gc/importer package, without exposing
compiler names in package names.

This change was created by manually adjusting the gcimporter paths in
go/types.bash and then running sh go/types.bash (i.e., by revendoring
gcimporter). The only manual changes are in go/types.bash.

Change-Id: Idc282439742288c09caa58b3a66d77aec0325faf
Reviewed-on: https://go-review.googlesource.com/8764
Reviewed-by: Rob Pike <r@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>

doc/go1.5.txt: log: LUTC flag

Change-Id: Ieadbd2b1ad442ba3eb2ad5f552d96c93ba3e5a2e
Reviewed-on: https://go-review.googlesource.com/8790
Reviewed-by: Rob Pike <r@golang.org>

log: add flag LUTC, to use UTC time zone for time stamp

Issue 9483 suggests several approaches to correlating logs from
machines in different time zones. This approach is the simplest and
really should be sufficient: provide a way to clamp the time stamps
to UTC.

Fixes #9483.

Change-Id: If540b991d758c4d845a719779f8255ece7c452e7
Reviewed-on: https://go-review.googlesource.com/8761
Reviewed-by: Ian Lance Taylor <iant@golang.org>

cmd/6g: call duffcopy, duffzero via got when dynamically linking go

Jumping to an offset past a symbol isn't something that is really
supported by dynamic linkers, so do it by hand.

Change-Id: Ifff8834c6cdfa3d521ebd8479d2e93906df9b258
Reviewed-on: https://go-review.googlesource.com/8238
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>

debug/gosym: skip tests when .gosymtab section not found

Skip the test when there is no .gosymtab section in the executable
rather than crashing.

Change-Id: Ieb3df07e307f50c33cdafab38f9b5d1ac0e55c04
Reviewed-on: https://go-review.googlesource.com/5110
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Michael Hudson-Doyle <michael.hudson@canonical.com>

doc/go1.5.txt: note new options for go, gc, asm, ld

Change-Id: I353ff7eb35b066a1a2693c087c9876adac8e3fd0
Reviewed-on: https://go-review.googlesource.com/8763
Reviewed-by: Ian Lance Taylor <iant@golang.org>

test: add gccgo test case for #10407

Change-Id: I8d17e2b0fbc529ca7958c75222964a5e419aa3db
Reviewed-on: https://go-review.googlesource.com/8717
Reviewed-by: Ian Lance Taylor <iant@golang.org>

runtime: Fix GDB integration with Python 2

A similar fix was applied in 545686857bc4c2e7a5306d97e5ef48f631d277bc
but another instance of 'pc' was missed.

Also adds a test for the goroutine gdb command.

It currently uses goroutine 2 for the test, since goroutine 1 has
its stack pointer set to 0 for some reason.

Change-Id: I53ca22be6952f03a862edbdebd9b5c292e0853ae
Reviewed-on: https://go-review.googlesource.com/8729
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

doc/go1.5.txt: time.AppendFormat

Change-Id: I9e8a0dc152ef9403ff5ece0c161bda3a2f4448a8
Reviewed-on: https://go-review.googlesource.com/8760
Reviewed-by: Rob Pike <r@golang.org>

time: add Time.AppendFormat

This is a version of Time.Format that doesn't require allocation.

This is an updated version of 0af302f50745b93e90a4507993a555d246acef45
submitted by @bradfitz which was later rolled back.

Fixes #5192
Updates #5195

Change-Id: I4e6255bee1cf3914a6cc8d9d2a881cfeb273c08e
Reviewed-on: https://go-review.googlesource.com/1760
Reviewed-by: Rob Pike <r@golang.org>

cmd/gc: fix handling of OGETG in race mode

Now that getg is an intrinsic, more runtime functions
gets inlined (in particular, LockOSThread).
Runtime code gets race instrumented after inlining into
other packages. This can lead to false positives,
as race detector ignores all internal synchronization in runtime.
Inling of LockOSThread lead to false race reports on m contents.
See the issue for an example.

Fixes #10380

Change-Id: Ic9b760b53c28c2350bc54a5d4677fcd1c1f86e5f
Reviewed-on: https://go-review.googlesource.com/8690
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Dmitry Vyukov <dvyukov@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

runtime: start concurrent GC promptly when we reach its trigger

Currently, when allocation reaches the concurrent GC trigger size, we
start the concurrent collector by ready'ing its G. This simply puts it
on the end of the P's run queue, which means we may not actually start
GC for some time as the current G continues to run and then the P
drains other Gs already on its run queue. Since the mutator can
continue to allocate, the heap can potentially be much larger than we
intended by the time GC actually starts. Furthermore, how much larger
is difficult to predict since it depends on the scheduler.

Fix this by preempting the current G and switching directly to the
concurrent GC G as soon as we reach the trigger heap size.

On the garbage benchmark from the benchmarks subrepo with
GOMAXPROCS=4, this reduces the time from triggering the GC to the
beginning of sweep termination by 10 to 30 milliseconds, which reduces
allocation after the trigger by up to 10MB (a large fraction of the
64MB live heap the benchmark tries to maintain).

One other known source of delay before we "really" start GC is the
sweep finalization performed before sweep termination. This has
similar negative effects on heap size and predictability, but is an
orthogonal problem. This change adds a TODO for this.

Change-Id: I8bae98cb43685c1bf353ff55868e4647e3743c47
Reviewed-on: https://go-review.googlesource.com/8513
Reviewed-by: Rick Hudson <rlh@golang.org>

runtime: remove GoSched/GoStart trace events around GC

These were appropriate for STW GC, since it interrupted the allocating
Goroutine, but don't apply to concurrent GC, which runs on its own
Goroutine. Forced GC is still STW, but it makes sense to attribute the
GC to the goroutine that called runtime.GC().

Change-Id: If12418ca66dc7e53b8b16025af4e03adb5d9577e
Reviewed-on: https://go-review.googlesource.com/8715
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Rick Hudson <rlh@golang.org>

internal/trace: don't assume GC will start and end on same P

Currently, GC disables preemption between the traceGCStart and
traceGCDone, so it never moves Ps. Consequently, the trace verifier
attaches information about GC to its per-P state and will fail if GC
starts on one P and ends on another.

GC will soon be preemptible and may end on a different P than it
began. Hence, this change lifts this per-P verifier state to global
state.

Change-Id: I82256e2baab1ff3c4453fec312079018423b4b51
Reviewed-on: https://go-review.googlesource.com/8714
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Rick Hudson <rlh@golang.org>

runtime: make test for freezetheworld more precise

exitsyscallfast checks for freezetheworld, but does so only by
checking if stopwait is positive. This can also happen during
stoptheworld, which is harmless, but confusing. Shortly, it will be
important that we get to the p.status cas even if stopwait is set.

Hence, make this test more specific so it only triggers with
freezetheworld and not other uses of stopwait.

Change-Id: Ibb722cd8360c3ed5a9654482519e3ceb87a8274d
Reviewed-on: https://go-review.googlesource.com/8205
Reviewed-by: Russ Cox <rsc@golang.org>

debug/dwarf: document DWARF class -> Go type mapping

Currently, the only way to know the Go type of an attribute of some
DWARF attribute class was to read the dwarf package code (or
experiment).  This makes it hard to go from the DWARF specification to
writing code that uses the dwarf package.

Fix this by adding a table to the documentation comment of the Field
type that gives the correspondence between DWARF attribute classes and
Go types.

Change-Id: I57c678a551fa1eb46f8207085d5a53d44985e3e7
Reviewed-on: https://go-review.googlesource.com/7280
Reviewed-by: Rob Pike <r@golang.org>
Reviewed-by: Nigel Tao <nigeltao@golang.org>

strconv: use 64bit uint for decimal conversion if available

The existing code used ints for the (slow) decimal conversion and
assumed that they were 32bit wide.

This change uses uints and the appropriate width (32 or 64bit)
depending on platform.

The performance difference is in the noise for the usual (optimized)
case which does not use the slow path conversion:

benchmark                               old ns/op     new ns/op     delta
BenchmarkFormatFloatDecimal             298           299           +0.34%
BenchmarkFormatFloat                    388           392           +1.03%
BenchmarkFormatFloatExp                 365           364           -0.27%
BenchmarkFormatFloatNegExp              364           362           -0.55%
BenchmarkFormatFloatBig                 482           476           -1.24%
BenchmarkAppendFloatDecimal             100           102           +2.00%
BenchmarkAppendFloat                    199           201           +1.01%
BenchmarkAppendFloatExp                 174           175           +0.57%
BenchmarkAppendFloatNegExp              169           174           +2.96%
BenchmarkAppendFloatBig                 286           286           +0.00%
BenchmarkAppendFloat32Integer           99.9          102           +2.10%
BenchmarkAppendFloat32ExactFraction     161           164           +1.86%
BenchmarkAppendFloat32Point             199           201           +1.01%
BenchmarkAppendFloat32Exp               167           168           +0.60%
BenchmarkAppendFloat32NegExp            163           169           +3.68%
BenchmarkAppendFloat64Fixed1            137           134           -2.19%
BenchmarkAppendFloat64Fixed2            144           146           +1.39%
BenchmarkAppendFloat64Fixed3            138           140           +1.45%
BenchmarkAppendFloat64Fixed4            144           145           +0.69%

The performance difference is significant if the fast path conversion is
explicitly turned off (ftoa.go:101):

benchmark                               old ns/op     new ns/op     delta
BenchmarkFormatFloatDecimal             459           427           -6.97%
BenchmarkFormatFloat                    1560          1180          -24.36%
BenchmarkFormatFloatExp                 5501          3128          -43.14%
BenchmarkFormatFloatNegExp              24085         14360         -40.38%
BenchmarkFormatFloatBig                 1409          1081          -23.28%
BenchmarkAppendFloatDecimal             248           226           -8.87%
BenchmarkAppendFloat                    1315          982           -25.32%
BenchmarkAppendFloatExp                 5274          2869          -45.60%
BenchmarkAppendFloatNegExp              23905         14054         -41.21%
BenchmarkAppendFloatBig                 1194          860           -27.97%
BenchmarkAppendFloat32Integer           167           175           +4.79%
BenchmarkAppendFloat32ExactFraction     182           184           +1.10%
BenchmarkAppendFloat32Point             556           564           +1.44%
BenchmarkAppendFloat32Exp               1134          918           -19.05%
BenchmarkAppendFloat32NegExp            2679          1801          -32.77%
BenchmarkAppendFloat64Fixed1            274           238           -13.14%
BenchmarkAppendFloat64Fixed2            494           368           -25.51%
BenchmarkAppendFloat64Fixed3            1833          1008          -45.01%
BenchmarkAppendFloat64Fixed4            6133          3596          -41.37%

Change-Id: I829b8abcca882b1c10d8ae421d3249597c31f3c9
Reviewed-on: https://go-review.googlesource.com/3811
Reviewed-by: Russ Cox <rsc@golang.org>

runtime: fix tracing of syscall exit

Fix tracing of syscall exit after:
https://go-review.googlesource.com/#/c/7504/

Change-Id: Idcde2aa826d2b9a05d0a90a80242b6bfa78846ab
Reviewed-on: https://go-review.googlesource.com/8728
Reviewed-by: Rick Hudson <rlh@golang.org>
Run-TryBot: Dmitry Vyukov <dvyukov@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

syscall: correct code for cover cmd

Fixes #10378

This is clumsy, but currently cover tool fails as:

$ go test -run=none -cover syscall
syscall_linux_amd64.go:15: can only use //go:noescape with external func implementations
FAIL syscall [build failed]

This happens because cover tool mishandles //go: comments.
r and gri said that fixing cover is infeasible due to go/ast limitations.

So at least fix the offending code so that coverage works.
This come up in context of coverage-guided fuzzing which works best
with program-wide coverage.

Change-Id: I142e5774c9f326ed38cb202693bd4edae93879ba
Reviewed-on: https://go-review.googlesource.com/8723
Reviewed-by: Rob Pike <r@golang.org>

test: add -update_errors flag to run script

The flag updates error annotations in test files from actual compiler output.
This is useful when doing compiler changes that add/remove/change lots of errors,
or when adding lots of new tests.
Also I noticed at least 2 cases where annotation were sub-optimal:
1. The annotation was "leaking param p" when the actual error is
"leaking param p to result ~r1".
2. The annotation was "leaking param m" when the actual errors
are "leaking param m" and "leaking param mv1".

For now it works only for errorcheck mode.

Also, apply the update to escape and liveness tests.
Some files have gccgo-specific errors of the form "gc error|gccgo error",
so it is risky to run update on all files. Gccgo-specific error
does not necessary contain '|', it can be just truncated.

Change-Id: Iaaae767f859dcb8321a8cb4970b2b70969e8a345
Reviewed-on: https://go-review.googlesource.com/5310
Run-TryBot: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

net: make multi-IP resolution more flexible.

Remove the "netaddr" type, which ambiguously represented either one
address, or a list of addresses. Instead, use "addrList" wherever
multiple addresses are supported.

The "first" method returns the first address matching some condition
(e.g. "is it IPv4?"), primarily to support legacy code that can't handle
multiple addresses.

The "partition" method splits an addrList into two categories, as
defined by some strategy function. This is useful for implementing
Happy Eyeballs, and similar two-channel algorithms.

Finally, internetAddrList (formerly resolveInternetAddr) no longer
mangles the ordering defined by getaddrinfo. In the future, this may
be used by a sequential Dial implementation.

Updates #8453, #8455.

Change-Id: I7375f4c34481580ab40e31d33002a4073a0474f3
Reviewed-on: https://go-review.googlesource.com/8360
Reviewed-by: Mikio Hara <mikioh.mikioh@gmail.com>
Run-TryBot: Mikio Hara <mikioh.mikioh@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/internal/gc: clean up Componentgen

Update #9855

In preparation for introducing direct use of a zero register on
platforms that support it, take the opportunity to clean up
Componentgen a bit.

Change-Id: I120ce1ffcca8c4f7603bfe76bfa1aedd27ebb4d2
Reviewed-on: https://go-review.googlesource.com/8691
Run-TryBot: Dave Cheney <dave@cheney.net>
Reviewed-by: Minux Ma <minux@golang.org>

runtime, cmd/internal/ld: rename themoduledata to firstmoduledata

'themoduledata' doesn't really make sense now we support multiple moduledata
objects.

Change-Id: I8263045d8f62a42cb523502b37289b0fba054f62
Reviewed-on: https://go-review.googlesource.com/8521
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/internal/obj/x86: look for go cmd in GOBIN if present.

If GOBIN is not empty the build moves the go executable
to a new path. When this test runs it fails to find the
go cmd in the GOROOT.

Change-Id: I100def0fbcb9691b13776f795b1d1725e36d8102
Reviewed-on: https://go-review.googlesource.com/8735
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

runtime, reflect: support multiple moduledata objects

This changes all the places that consult themoduledata to consult a
linked list of moduledata objects, as will be necessary for
-linkshared to work.

Obviously, as there is as yet no way of adding moduledata objects to
this list, all this change achieves right now is wasting a few
instructions here and there.

Change-Id: I397af7f60d0849b76aaccedf72238fe664867051
Reviewed-on: https://go-review.googlesource.com/8231
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/go: support -buildmode=default,archive,exe,c-shared

Modelled somewhat on the -race support.

Change-Id: I137037addfc76341f7deb216776fdd18e9af9fe5
Reviewed-on: https://go-review.googlesource.com/8680
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/internal/obj/x86, cmd/internal/ld, cmd/6l: 6g/asm -dynlink accesses global data via a GOT

Change-Id: I49862e177045369d6c94d6a58afbdace4f13cc96
Reviewed-on: https://go-review.googlesource.com/8237
Reviewed-by: Ian Lance Taylor <iant@golang.org>

runtime: fix arm64 build

Broken by CL 8541.

Change-Id: Ie2e89a22b91748e82f7bc4723660a24ed4135687
Reviewed-on: https://go-review.googlesource.com/8734
Reviewed-by: Minux Ma <minux@golang.org>

runtime: report next_gc for initial heap size in gctrace

Currently, the initial heap size reported in the gctrace line is the
heap_live right before sweep termination. However, we triggered GC
when heap_live reached next_gc, and there may have been significant
allocation between that point and the beginning of sweep
termination. Ideally these would be essentially the same, but
currently there's scheduler delay when readying the GC goroutine as
well as delay from background sweep finalization.

We should fix this delay, but in the mean time, to give the user a
better idea of how much the heap grew during the whole of garbage
collection, report the trigger rather than what the heap size happened
to be after the garbage collector finished rolling out of bed. This
will also be more useful for heap growth plots.

Change-Id: I08476b9fbcfb2de90592405e9c9f434dfb9eb1f8
Reviewed-on: https://go-review.googlesource.com/8512
Reviewed-by: Rick Hudson <rlh@golang.org>

math/big: better doc strings for ErrNaN functionality

Change-Id: Ia0944e7b47193465d3ec37fc8dc46dea9b5dcc6b
Reviewed-on: https://go-review.googlesource.com/8710
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

runtime: add isarchive, set by the linker

According to Go execution modes, a Go program compiled with
-buildmode=c-archive has a main function, but it is ignored on run.
This gives the runtime the information it needs not to run the main.

I have this working with pending linker changes on darwin/amd64.

Change-Id: I49bd7d65aa619ec847c464a872afa5deea7d4d30
Reviewed-on: https://go-review.googlesource.com/8701
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

log: logging an empty string should still print a line

Print("") was printing a header but no line.

Fixes #9665.

Change-Id: Iac783187786065e1389ad6e8d7ef02c579ed7bd8
Reviewed-on: https://go-review.googlesource.com/8665
Reviewed-by: Ian Lance Taylor <iant@golang.org>

encoding/gob: more checks for corrupted data

Also unify the tests where possible to make it easy to add more.

Fixes #10273.

Change-Id: Idfa4f4a5dcaa05974066bafe17bed6cdd2ebedb7
Reviewed-on: https://go-review.googlesource.com/8662
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

runtime: add arm64 runtime.cmpstring and bytes.Compare

Add arm64 assembly implementation of runtime.cmpstring and bytes.Compare.

benchmark                                old ns/op     new ns/op     delta
BenchmarkCompareBytesEqual               98.0          27.5          -71.94%
BenchmarkCompareBytesToNil               9.38          10.0          +6.61%
BenchmarkCompareBytesEmpty               13.3          10.0          -24.81%
BenchmarkCompareBytesIdentical           98.0          27.5          -71.94%
BenchmarkCompareBytesSameLength          43.3          16.3          -62.36%
BenchmarkCompareBytesDifferentLength     43.4          16.3          -62.44%
BenchmarkCompareBytesBigUnaligned        6979680       1360979       -80.50%
BenchmarkCompareBytesBig                 6915995       1381979       -80.02%
BenchmarkCompareBytesBigIdentical        6781440       1327304       -80.43%

benchmark                             old MB/s     new MB/s     speedup
BenchmarkCompareBytesBigUnaligned     150.23       770.46       5.13x
BenchmarkCompareBytesBig              151.62       758.76       5.00x
BenchmarkCompareBytesBigIdentical     154.63       790.01       5.11x

* note, the machine we are benchmarking on has some issues. What is clear is
compared to a few days ago the old MB/s value has increased from ~115 to 150.
I'm less certain about the new MB/s number, which used to be close to 1Gb/s.

Change-Id: I4f31b2c7a06296e13912aacc958525632cb0450d
Reviewed-on: https://go-review.googlesource.com/8541
Reviewed-by: Aram Hăvărneanu <aram@mgk.ro>
Reviewed-by: David Crawshaw <crawshaw@golang.org>

syscall: reduce the set of architecture-dependent files on Plan 9

Change-Id: I98b172181c2fd85aa385341e28bc661dbc274252
Reviewed-on: https://go-review.googlesource.com/2167
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

net/http: handle close/response race more gracefully

There was a logical race in Transport.RoundTrip where a roundtrip with
a pending response would race with the channel for the connection
closing. This usually happened for responses with connection: close
and no body.

We handled this race by reading the close channel, setting a timer
for 100ms and if no response was returned before then, we would then
return an error.

This put a lower bound on how fast a connection could fail. We couldn't
fail a request faster than 100ms.

Reordering the channel operations gets rid of the logical race. If
the readLoop causes the connection to be closed, it would have put
its response into the return channel already and we can fetch it with
a non-blocking receive.

Change-Id: Idf09e48d7a0453d7de0120d3055d0ce5893a5428
Reviewed-on: https://go-review.googlesource.com/1787
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/go: detect which VCS to use with Bitbucket when the API call fails.

      The API call will fail when Bitbucket repositories are private. In
that case, probe for the repository using vcsCmd.ping.

      Fixes #5375

Change-Id: Ia604ecf9014805579dfda4b5c8e627a52783d56e
Reviewed-on: https://go-review.googlesource.com/1910
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/gc: fix escape analysis of closures

Fixes #10353

See test/escape2.go:issue10353. Previously new(int) did not escape to heap,
and so heap-allcated closure was referencing a stack var. This breaks
the invariant that heap must not contain pointers to stack.

Look at the following program:

package main

func main() {
foo(new(int))
bar(new(int))
}

func foo(x *int) func() {
return func() {
println(*x)
}
}

// Models what foo effectively does.
func bar(x *int) *C {
return &C{x}
}

type C struct {
x *int
}

Without this patch escape analysis works as follows:

$ go build -gcflags="-m -m -m -l" esc.go
escflood:1: dst ~r1 scope:foo[0]
escwalk: level:0 depth:0  func literal( l(9) f(1) esc(no) ld(1)) scope:foo[1]
/tmp/live2.go:9: func literal escapes to heap
escwalk: level:0 depth:1   x( l(8) class(PPARAM) f(1) esc(no) ld(1)) scope:foo[1]
/tmp/live2.go:8: leaking param: x to result ~r1

escflood:2: dst ~r1 scope:bar[0]
escwalk: level:0 depth:0  &C literal( l(15) esc(no) ld(1)) scope:bar[1]
/tmp/live2.go:15: &C literal escapes to heap
escwalk: level:-1 depth:1   &C literal( l(15)) scope:bar[0]
escwalk: level:-1 depth:2   x( l(14) class(PPARAM) f(1) esc(no) ld(1)) scope:bar[1]
/tmp/live2.go:14: leaking param: x

/tmp/live2.go:5: new(int) escapes to heap
/tmp/live2.go:4: main new(int) does not escape

new(int) does not escape while being captured by the closure.
With this patch escape analysis of foo and bar works similarly:

$ go build -gcflags="-m -m -m -l" esc.go
escflood:1: dst ~r1 scope:foo[0]
escwalk: level:0 depth:0  &(func literal)( l(9)) scope:foo[0]
escwalk: level:-1 depth:1   func literal( l(9) f(1) esc(no) ld(1)) scope:foo[1]
/tmp/live2.go:9: func literal escapes to heap
escwalk: level:-1 depth:2   x( l(8) class(PPARAM) f(1) esc(no) ld(1)) scope:foo[1]
/tmp/live2.go:8: leaking param: x

escflood:2: dst ~r1 scope:bar[0]
escwalk: level:0 depth:0  &C literal( l(15) esc(no) ld(1)) scope:bar[1]
/tmp/live2.go:15: &C literal escapes to heap
escwalk: level:-1 depth:1   &C literal( l(15)) scope:bar[0]
escwalk: level:-1 depth:2   x( l(14) class(PPARAM) f(1) esc(no) ld(1)) scope:bar[1]
/tmp/live2.go:14: leaking param: x

/tmp/live2.go:4: new(int) escapes to heap
/tmp/live2.go:5: new(int) escapes to heap

Change-Id: Ifd14b7ae3fc11820e3b5eb31eb07f35a22ed0932
Reviewed-on: https://go-review.googlesource.com/8408
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Dmitry Vyukov <dvyukov@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

runtime: make windows exception handler code arch independent

Mainly it is simple copy. But I had to change amd64
lastcontinuehandler return value from uint32 to int32.
I don't remember how it happened to be uint32, but new
int32 is matching better with Windows documentation (LONG).
I don't think it matters one way or the others.

Change-Id: I6935224a2470ad6301e27590f2baa86c13bbe8d5
Reviewed-on: https://go-review.googlesource.com/8686
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

regexp: skip backtracker for long programs

This update makes maxBacktrackLen return 0 if
len(prog.Inst) > maxBacktrackProg. This prevents an attempt to
backtrack against a nil bitstate.

Fixes #10319

Change-Id: Icdbeb2392782ccf66f9d0a70ea57af22fb93f01b
Reviewed-on: https://go-review.googlesource.com/8473
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

net/http: don't send IPv6 zone identifier in outbound request, per RFC 6874

When making a request to an IPv6 address with a zone identifier, for
exmaple [fe80::1%en0], RFC 6874 says HTTP clients must remove the zone
identifier "%en0" before writing the request for security reason.

This change removes any IPv6 zone identifer attached to URI in the Host
header field in requests.

Fixes #9544.

Change-Id: I7406bd0aa961d260d96f1f887c2e45854e921452
Reviewed-on: https://go-review.googlesource.com/3111
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

os: windows Rename should overwrite destination file.

Rename now uses MoveFileEx which was previously not available to
use because it is not supported on Windows 2000.

Change-Id: I583d029c4467c9be6d1574a790c423559b441e87
Reviewed-on: https://go-review.googlesource.com/6140
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

cmd/internal/obj/arm64, cmd/asm/internal/asm: support CSEL instruction on arm64

Add support for arm64 four operand conditional instructions.

Superceedes CL 8405.

Change-Id: I12da8f4822938feec400bbcc426eeaf884536135
Reviewed-on: https://go-review.googlesource.com/8638
Reviewed-by: Aram Hăvărneanu <aram@mgk.ro>

runtime: do not calculate asmstdcall address every time we make syscall

Change-Id: If3c8c9035e12d41647ae4982883f6a979313ea9d
Reviewed-on: https://go-review.googlesource.com/8682
Reviewed-by: Minux Ma <minux@golang.org>

image/jpeg: reject multiple Start-Of-Frame markers.

Fixes #10389

Change-Id: Id1c687122751f9317041d9e425d03b267a26c6de
Reviewed-on: https://go-review.googlesource.com/8681
Reviewed-by: Rob Pike <r@golang.org>

runtime: fix darwin/386, darwin/arm builds

In cl/8652 I broke darwin/arm and darwin/386 because I removed the *g
parameter, which they both expect and use. This CL adjusts both ports
to look for g0 in m, just as darwin/amd64 does.

Tested on darwin{386,arm,amd64}.

Change-Id: Ia56f3d97e126b40d8bbd2e8f677b008e4a1badad
Reviewed-on: https://go-review.googlesource.com/8666
Reviewed-by: Ian Lance Taylor <iant@golang.org>

runtime: use (*context) ip, setip, sp and setsp everywhere on windows

Also move dumpregs into defs_windows_*.go.

Change-Id: Ic077d7dbb133c7b812856e758d696d6fed557afd
Reviewed-on: https://go-review.googlesource.com/4650
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/internal/ld: support for -buildmode=shared

Change-Id: Id4997d611ced29397133f14def6abc88aa9e811e
Reviewed-on: https://go-review.googlesource.com/8252
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

cmd/internal/gc, etc: remove dead code

Found with https://github.com/opennota/check.

Change-Id: I50c173382782fb16b15100e02c1c85610bc233a0
Reviewed-on: https://go-review.googlesource.com/7130
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Minux Ma <minux@golang.org>

cmd/internal/ld: clean up hostlink I/O

Change-Id: I6c3a62403941d357ffd9d0025289c2180139b0bd
Reviewed-on: https://go-review.googlesource.com/8664
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Minux Ma <minux@golang.org>

runtime: darwin/amd64 library entry point

This is a practice run for darwin/arm.

Similar to the linux/amd64 shared library entry point. With several
pending linker changes I am successfully using this to implement
-buildmode=c-archive on darwin/amd64 with external linking.

The same entry point can be reused to implement -buildmode=c-shared
on darwin/amd64, however that will require further ld changes to
remove all text relocations.

One extra runtime change will follow this. According to the Go
execution modes document, -buildmode=c-archive should ignore the Go
main function. Right now it is being executed (and the process exits
if it doesn't block). I'm still searching for the right way to do
this.

Change-Id: Id97901ddd4d46970996f222bd79731dabff66a3d
Reviewed-on: https://go-review.googlesource.com/8652
Reviewed-by: Ian Lance Taylor <iant@golang.org>

cmd/internal/ld: add -buildmode=c-shared as an alternative to -shared

The linker currently (on some platforms) takes a -shared flag, which means
approximately what -buildmode=c-shared means in the in the proposed "Go
Execution Modes" document. As part of implementing other modes, the term
"shared" becomes horribly overloaded, so this replaces -shared with a
-buildmode argument instead (which currently only handles -buildmode=c-shared
and the default -buildmode=exe -- no new behaviour here).

As the linker support for -shared was in 1.4 this retains it as an alias.

Change-Id: Id2ebb8e05ee07f46208a554bc2622d0e67b47082
Reviewed-on: https://go-review.googlesource.com/8304
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

cmd/internal/ld: emit macho .init_array section

Change-Id: Ie75a01e899e68f4f9643410f5e161152a81b8ba0
Reviewed-on: https://go-review.googlesource.com/8655
Reviewed-by: Ian Lance Taylor <iant@golang.org>

encoding/gob: clean up decoderMap after errBadType

When decoding an invalid typeId the associated *decEngine was not
removed from decoderMap. If the decoder was run again on the same input
a nil *decEngine was found in the map and assumed to be initialized,
resulting in a panic.

Fixes #9649

Change-Id: I5bb51808362a21c09228c2705a658f073e5b59b3
Reviewed-on: https://go-review.googlesource.com/3509
Reviewed-by: Rob Pike <r@golang.org>

math/big: make ErrNaN actually implement the error interface (oversight)

There was no way to get to the error message before.

Change-Id: I4aa9d3d9f468c33f9996295bafcbed097de0389f
Reviewed-on: https://go-review.googlesource.com/8660
Reviewed-by: Alan Donovan <adonovan@google.com>

net/rpc: document that the type must be exported, not just the methods

Fixes #10379.

Change-Id: Ia4cdda36ed57a06371f9ace7365ce9e215228487
Reviewed-on: https://go-review.googlesource.com/8654
Reviewed-by: Rob Pike <r@golang.org>

cmd/asm: remove object file if assembly fails.

Just an oversight. Plus the code had an unnecessary call to os.Exit
that now has a purpose.

Fixes #10372.

Change-Id: I456018f3a01ca05b4501c7f8a4961d48ab8c5e16
Reviewed-on: https://go-review.googlesource.com/8651
Reviewed-by: Minux Ma <minux@golang.org>

go/types/internal/gcimporter: update for 7g and 9g

Change-Id: Ied1582d8aabee2eb346e1c23bfd7781e4a091264
Reviewed-on: https://go-review.googlesource.com/8621
Reviewed-by: Robert Griesemer <gri@golang.org>

go/types: more selective disabling of tests

Disable importer-dependent tests on platforms for which the
respective builders don't have access to importable packages.

Fixes #10368.

Change-Id: I8072c59d2bbbc24a43d771fd04fd0b1a678d765a
Reviewed-on: https://go-review.googlesource.com/8611
Reviewed-by: Minux Ma <minux@golang.org>

cmd/internal/ld: change elf64 from int to bool

Change-Id: Iaf2dba7d699a8d52f91ce10222ab0d1a0f1f21fc
Reviewed-on: https://go-review.googlesource.com/8625
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Minux Ma <minux@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

runtime, cmd/internal/ld: initialize themoduledata slices directly

This CL is quite conservative in some ways.  It continues to define
symbols that have no real purpose (e.g. epclntab).  These could be
deleted if there is no concern that external tools might look for them.

It would also now be possible to make some changes to the pcln data but
I get the impression that would definitely require some thought and
discussion.

Change-Id: Ib33cde07e4ec38ecc1d6c319a10138c9347933a3
Reviewed-on: https://go-review.googlesource.com/7616
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

database/sql: Retry with a fresh connection after maxBadConnRetries

Previously if the connection pool was larger than maxBadConnRetries
and there were a lot of bad connections in the pool (for example if
the database server was restarted), a query might have failed with an
ErrBadConn unnecessarily.  Instead of trying to guess how many times
to retry, try maxBadConnRetries times and then force a fresh
connection to be used for the last attempt.  At the same time, lower
maxBadConnRetries to a smaller value now that it's not that important
to retry so many times from the free connection list.

Fixes #8834

Change-Id: I6542f151a766a658980fb396fa4880ecf5874e3d
Reviewed-on: https://go-review.googlesource.com/2034
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

time: unify formatting of decimals for timestamps

Change function appendUint to appendInt with variable-width 0-padding.

This allows the decimal for the year to be generated without extra code
to handle the wider padding and directly handles negative numbers.

Removes the special casing for numbers with one and two digits.
The special case for 0 was unreachable.

The new version is slightly slower.

benchmark              old ns/op     new ns/op     delta
BenchmarkFormat        444           454           +2.25%
BenchmarkFormatNow     398           415           +4.27%

Change-Id: I4ddef96bf07ad35dca76053321d510441ec6d4f5
Reviewed-on: https://go-review.googlesource.com/2751
Reviewed-by: Robert Griesemer <gri@golang.org>

cmd/7g: reactivate componentgen

Update #10203

When the portable componentgen was introduced in b960263 it
produced broken code on arm64 and was deactivated. In the month since
it looks like the underlying issues have been fixed so componentgen
produces working binaries that are slightly smaller, ~3kb reduction in
size for godoc binary.

Benchmarks are underwhelming, but where visible, trending towards
an improvement (this is with Minux's peep optimiser CL).

benchmark                          old ns/op       new ns/op       delta
BenchmarkBinaryTree17              15336842000     15002766000     -2.18%
BenchmarkFannkuch11                10848984000     10896931000     +0.44%
BenchmarkFmtFprintfEmpty           203             188             -7.39%
BenchmarkFmtFprintfString          753             720             -4.38%
BenchmarkFmtFprintfInt             667             670             +0.45%
BenchmarkFmtFprintfIntInt          1103            1102            -0.09%
BenchmarkFmtFprintfPrefixedInt     981             969             -1.22%
BenchmarkFmtFprintfFloat           1396            1348            -3.44%
BenchmarkFmtManyArgs               4151            4102            -1.18%
BenchmarkGobDecode                 34202360        32933020        -3.71%
BenchmarkGobEncode                 27579180        27438820        -0.51%
BenchmarkGzip                      1296119000      1285096000      -0.85%
BenchmarkGunzip                    291099800       289727200       -0.47%
BenchmarkHTTPClientServer          169476          169803          +0.19%
BenchmarkJSONEncode                70313600        69973400        -0.48%
BenchmarkJSONDecode                227811800       232875200       +2.22%
BenchmarkMandelbrot200             12985600        12996430        +0.08%
BenchmarkGoParse                   14708930        14507320        -1.37%
BenchmarkRegexpMatchEasy0_32       464             460             -0.86%
BenchmarkRegexpMatchEasy0_1K       4516            4517            +0.02%
BenchmarkRegexpMatchEasy1_32       452             454             +0.44%
BenchmarkRegexpMatchEasy1_1K       4664            4669            +0.11%
BenchmarkRegexpMatchMedium_32      602             602             +0.00%
BenchmarkRegexpMatchMedium_1K      172939          172494          -0.26%
BenchmarkRegexpMatchHard_32        9733            9577            -1.60%
BenchmarkRegexpMatchHard_1K        301356          298143          -1.07%
BenchmarkRevcomp                   2754334000      2753874000      -0.02%
BenchmarkTemplate                  315664000       311810800       -1.22%
BenchmarkTimeParse                 1034            989             -4.35%
BenchmarkTimeFormat                1118            1137            +1.70%

benchmark                         old MB/s     new MB/s     speedup
BenchmarkGobDecode                22.44        23.31        1.04x
BenchmarkGobEncode                27.83        27.97        1.01x
BenchmarkGzip                     14.97        15.10        1.01x
BenchmarkGunzip                   66.66        66.98        1.00x
BenchmarkJSONEncode               27.60        27.73        1.00x
BenchmarkJSONDecode               8.52         8.33         0.98x
BenchmarkGoParse                  3.94         3.99         1.01x
BenchmarkRegexpMatchEasy0_32      68.92        69.47        1.01x
BenchmarkRegexpMatchEasy0_1K      226.71       226.65       1.00x
BenchmarkRegexpMatchEasy1_32      70.75        70.42        1.00x
BenchmarkRegexpMatchEasy1_1K      219.55       219.28       1.00x
BenchmarkRegexpMatchMedium_32     1.66         1.66         1.00x
BenchmarkRegexpMatchMedium_1K     5.92         5.94         1.00x
BenchmarkRegexpMatchHard_32       3.29         3.34         1.02x
BenchmarkRegexpMatchHard_1K       3.40         3.43         1.01x
BenchmarkRevcomp                  92.28        92.29        1.00x
BenchmarkTemplate                 6.15         6.22         1.01x

Change-Id: I0b7d95388d6920fcbd7fe305df0c7c630a407726
Reviewed-on: https://go-review.googlesource.com/8636
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/asm/internal/asm: add arm64 end to end tests

Add end to end tests for arm64 to support CL 8405.

There are several instruction forms commented out at the moment
they will be addressed in CL 8405 or later followups.

Change-Id: I6eeeb810c1e03cd49bb3c881bc46a29cdb817822
Reviewed-on: https://go-review.googlesource.com/8631
Reviewed-by: Aram Hăvărneanu <aram@mgk.ro>
Reviewed-by: Rob Pike <r@golang.org>

androidtest.bash: copy pkg for gcimporter tests

The tests for go/types depend on reading gc export data from the
$GOROOT/pkg directory. This is the first use of these files as
testdata, so previously they were not copied to the android device.
Now they are used, copy them.

Fixes android/arm build.

Change-Id: If13bbe603ce0aff697a73a97ae9a7d6b3ea800f9
Reviewed-on: https://go-review.googlesource.com/8624
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

runtime: fix return type for bsdthread_register in comments

The return type for bsdthread_register is int32. See
runtime/os_darwin.go.

This change also rewrites declaration comments for go functions to
use go syntax and fixes vet errors in sys_darwin_amd64.s.

Change-Id: I7482105f7562929e0ede30099efac9e76babd8a3
Reviewed-on: https://go-review.googlesource.com/3260
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

io: clarify Copy docs regarding error handling

"returns ... the first error" was misleading or at least confusing:
in case a Read results in an error with non-zero bytes read, and the
subsequent Write also results in an error, the error from Write is
returned, which is the second one (in the temporal dimension).

Fixes #9744

Change-Id: If8925a701e4fae820cd9df7446503403fc0785d4
Reviewed-on: https://go-review.googlesource.com/3686
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

doc/go1.5: mention cgo is supported on linux/arm64 (ext. link only)

Change-Id: I1dcca264d9cc900aad2d7737073cc01fe574bf55
Reviewed-on: https://go-review.googlesource.com/8623
Reviewed-by: Minux Ma <minux@golang.org>

go/build: cgo is supported on linux/arm64 (external linking only)

Fixes #10107.

Change-Id: I309e3df7608b9eef9339196fdc50dedf5f9439f6
Reviewed-on: https://go-review.googlesource.com/8453
Reviewed-by: Aram Hăvărneanu <aram@mgk.ro>

cmd/internal/ld: force external linking on linux/arm64 with cgo

Update #10373.

Change-Id: I309e3df7608b9eef9339196fdc50dedf5f9439f5
Reviewed-on: https://go-review.googlesource.com/8452
Reviewed-by: Aram Hăvărneanu <aram@mgk.ro>
Reviewed-by: David Crawshaw <crawshaw@golang.org>

misc/cgo/test/issue9400: add arm64 implementation

Change-Id: I309e3df7608b9eef9339196fdc50dedf5f9439f4
Reviewed-on: https://go-review.googlesource.com/8451
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Reviewed-by: Aram Hăvărneanu <aram@mgk.ro>

runtime: linux/arm64 cgo support

Change-Id: I309e3df7608b9eef9339196fdc50dedf5f9439f3
Reviewed-on: https://go-review.googlesource.com/8450
Reviewed-by: Aram Hăvărneanu <aram@mgk.ro>

runtime/cgo: linux/arm64 cgo support

Change-Id: I309e3df7608b9eef9339196fdc50dedf5f9439f2
Reviewed-on: https://go-review.googlesource.com/8439
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Reviewed-by: Aram Hăvărneanu <aram@mgk.ro>

cmd/internal/obj/arm64, cmd/internal/ld, cmd/7l: remove absolute addressing in .text

This CL introduces R_ADDRARM64, which is similar to R_ADDRPOWER.

Fixes #10112.

Change-Id: I309e3df7608b9eef9339196fdc50dedf5f9439f1
Reviewed-on: https://go-review.googlesource.com/8438
Reviewed-by: Aram Hăvărneanu <aram@mgk.ro>
Reviewed-by: David Crawshaw <crawshaw@golang.org>

cmd/internal/ld, cmd/7l: external linking support for linux/arm64

Based on Michael Hudson-Doyle's patch:
https://github.com/4ad/go/commit/b735215ee41b4237ec393d4669f6f55d4b27babc

Change-Id: I309e3df7608b9eef9339196fdc50dedf5f9439f0
Reviewed-on: https://go-review.googlesource.com/8437
Reviewed-by: Aram Hăvărneanu <aram@mgk.ro>
Reviewed-by: David Crawshaw <crawshaw@golang.org>

cmd/api: make the test more robust

Previously, the TestCompareAPI test would fail if runtime.Version()
is "dev", or, more importantly, "go1.5"; because compareAPI depends
on runtime.Version and -allow_new flag. Move that logic out make
its test more robust.

Change-Id: I8f40daa1838b8acd26adac8848762d95315053b0
Reviewed-on: https://go-review.googlesource.com/8622
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/7g: enable peephole optimizer

Based on cmd/9g/peep.go.

Go 1 benchmark comparison:
benchmark                          old ns/op       new ns/op       delta
BenchmarkBinaryTree17              24328574000     18351639000     -24.57%
BenchmarkFannkuch11                17029365000     10817758000     -36.48%
BenchmarkFmtFprintfEmpty           291             223             -23.37%
BenchmarkFmtFprintfString          1073            799             -25.54%
BenchmarkFmtFprintfInt             1024            778             -24.02%
BenchmarkFmtFprintfIntInt          1654            1277            -22.79%
BenchmarkFmtFprintfPrefixedInt     1360            1083            -20.37%
BenchmarkFmtFprintfFloat           2272            1415            -37.72%
BenchmarkFmtManyArgs               5933            4742            -20.07%
BenchmarkGobDecode                 53166003        38584736        -27.43%
BenchmarkGobEncode                 37930156        30074874        -20.71%
BenchmarkGzip                      1880638900      1286832100      -31.57%
BenchmarkGunzip                    386343633       292194480       -24.37%
BenchmarkHTTPClientServer          237077          179776          -24.17%
BenchmarkJSONEncode                101731690       73116925        -28.13%
BenchmarkJSONDecode                344655360       241277600       -29.99%
BenchmarkMandelbrot200             28329778        12950809        -54.29%
BenchmarkGoParse                   21670755        16554244        -23.61%
BenchmarkRegexpMatchEasy0_32       557             484             -13.11%
BenchmarkRegexpMatchEasy0_1K       4687            4832            +3.09%
BenchmarkRegexpMatchEasy1_32       539             483             -10.39%
BenchmarkRegexpMatchEasy1_1K       5100            5080            -0.39%
BenchmarkRegexpMatchMedium_32      796             651             -18.22%
BenchmarkRegexpMatchMedium_1K      233099          182047          -21.90%
BenchmarkRegexpMatchHard_32        13202           9897            -25.03%
BenchmarkRegexpMatchHard_1K        401027          303602          -24.29%
BenchmarkRevcomp                   3837679666      2816546600      -26.61%
BenchmarkTemplate                  440608300       324831040       -26.28%
BenchmarkTimeParse                 1460            1019            -30.21%
BenchmarkTimeFormat                1609            1174            -27.04%

benchmark                         old MB/s     new MB/s     speedup
BenchmarkGobDecode                14.44        19.89        1.38x
BenchmarkGobEncode                20.24        25.52        1.26x
BenchmarkGzip                     10.32        15.08        1.46x
BenchmarkGunzip                   50.23        66.41        1.32x
BenchmarkJSONEncode               19.07        26.54        1.39x
BenchmarkJSONDecode               5.63         8.04         1.43x
BenchmarkGoParse                  2.67         3.50         1.31x
BenchmarkRegexpMatchEasy0_32      57.38        66.05        1.15x
BenchmarkRegexpMatchEasy0_1K      218.47       211.91       0.97x
BenchmarkRegexpMatchEasy1_32      59.29        66.21        1.12x
BenchmarkRegexpMatchEasy1_1K      200.76       201.54       1.00x
BenchmarkRegexpMatchMedium_32     1.26         1.53         1.21x
BenchmarkRegexpMatchMedium_1K     4.39         5.62         1.28x
BenchmarkRegexpMatchHard_32       2.42         3.23         1.33x
BenchmarkRegexpMatchHard_1K       2.55         3.37         1.32x
BenchmarkRevcomp                  66.23        90.24        1.36x
BenchmarkTemplate                 4.40         5.97         1.36x

Fixes #10105.

Change-Id: I353cc9fdf922e431821508c9dbbe4d9a85d64bd4
Signed-off-by: Shenghou Ma <minux@golang.org>
Reviewed-on: https://go-review.googlesource.com/8471
Reviewed-by: Dave Cheney <dave@cheney.net>

cmd/api: re-enable TestGolden on nacl

Fixes #10369.

Change-Id: If0a6d2b33c6862c9f7f862bdc997f2204072c6dc
Reviewed-on: https://go-review.googlesource.com/8620
Run-TryBot: Minux Ma <minux@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/asm/internal/asm: parse arm64 register pairs for LDP/STP

Add test, and while we're at here, also add a test for ARM.

Fixes #10343.

Change-Id: Ic914df8233d4f1f495e2cc0743fbd37b7671bc91
Signed-off-by: Shenghou Ma <minux@golang.org>
Reviewed-on: https://go-review.googlesource.com/8472
Reviewed-by: Aram Hăvărneanu <aram@mgk.ro>
Reviewed-by: Rob Pike <r@golang.org>

go/types: enable disabled test

Change-Id: I58de76c49de6e43befb30b2bf677934e7952b5ab
Reviewed-on: https://go-review.googlesource.com/8610
Reviewed-by: Rob Pike <r@golang.org>