runtime/cgo: no-op getwd call as test breakpoint

This is to be used by an lldb script inside go_darwin_arm_exec to pause
the execution of tests on iOS so the working directory can be adjusted
into something resembling a GOROOT.

Change-Id: I69ea2d4d871800ae56634b23ffa48583559ddbc6
Reviewed-on: https://go-review.googlesource.com/6363
Reviewed-by: Minux Ma <minux@golang.org>

image/draw: add a fast path for Gray src images.

Grayscale PNG and JPEG images are not uncommon. We should have a fast path.

Also add a benchmark for the recently added CMYK fast path.

benchmark                    old ns/op     new ns/op     delta
BenchmarkGray                13960348      324152        -97.68%

Change-Id: I72b5838c8c3d1f2d0a4536a848e020e80b10c0f7
Reviewed-on: https://go-review.googlesource.com/6237
Reviewed-by: Rob Pike <r@golang.org>

cmd/internal/gc, cmd/internal/obj: remove pointless fmt.Sprintf calls

This is a follow-up to CL 6265. No behavior changes.
The diff was generated with eg, using template:

package p

import "fmt"

func before(a string) string { return fmt.Sprintf(a) }
func after(a string) string  { return a }

Change-Id: I7b3bebf31be5cd1ae2233da06cb4502a3d73f092
Reviewed-on: https://go-review.googlesource.com/6269
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

net: disable TestTCPReadWriteAllocs on darwin

Having this test fail, as it does reliably for me,
makes working frustrating. Disable it for now,
until we can diagnose the issue.

Update issue #8859.

Change-Id: I9dda30d60793e7a51f48f445c78ccb158068cc25
Reviewed-on: https://go-review.googlesource.com/6381
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/internal/obj: clean up handling of register list operand on ARM

ARM operands for MOVM have lists of registers: [R1,R2,R5-R8].
Handle them cleanly.

It was TYPE_CONST with special handling, which meant operand printing
didn't work right and the special handling was ugly. Add a new TYPE_REGLIST
for this case and it all gets cleaner.

Change-Id: I4a64f70fb9765e63cb636619a7a8553611bfe970
Reviewed-on: https://go-review.googlesource.com/6300
Run-TryBot: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>

syscall: make pwd process-wide on Plan 9

On Plan 9, the pwd is apparently per-thread not per process. That
means different goroutines saw different current directories, even
changing within a goroutine as they were scheduled.

Instead, track the the process-wide pwd protected by a mutex in the
syscall package and set the current goroutine thread's pwd to the
correct once at critical points.

Fixes #9428

Change-Id: I928e90886355be4a95c2be834f5883e2b50fc0cf
Reviewed-on: https://go-review.googlesource.com/6350
Reviewed-by: David du Colombier <0intro@gmail.com>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

Revert "os: set TMPDIR on darwin/arm"

This reverts commit 87a0d395c3d454c6efa560c406b5b3d8ad71c8f6.

Looks like introducing file_darwin_arm.go is confusing something in the API checker (probably go/types) into ignoring file.go, so the O_SYNC symbol is being lost.

No actual bug in this CL AFAIK, but I'll fix the other bug later and then roll this forward.

Change-Id: Ic132fb101e4b5f2716f7a0d15872bf35bdf42139
Reviewed-on: https://go-review.googlesource.com/6331
Reviewed-by: David Crawshaw <crawshaw@golang.org>

image/draw: optimize drawFillSrc.

benchmark                    old ns/op     new ns/op     delta
BenchmarkFillSrc             46781         46000         -1.67%

Change-Id: I0ab25d42d5763f1a0fe5a67ee00b83f0aa55f1f6
Reviewed-on: https://go-review.googlesource.com/6235
Reviewed-by: Rob Pike <r@golang.org>

os: fix tests on darwin/arm

Tests that fork are skipped. Tests that create files do so in a
temporary directory, as the initial PWD is read-only. And
darwin/arm returns a strange mkdir error when trying to write to /.

Change-Id: I2de661a85524300bbac226693d72142992dc188d
Reviewed-on: https://go-review.googlesource.com/6312
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

os: set TMPDIR on darwin/arm

Change-Id: Iee25f4b0a31ece0aae79c68aec809e1e4308f865
Reviewed-on: https://go-review.googlesource.com/6311
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/internal/gc, cmd/internal/obj: clean up string returns

An artifact of the c2go translation was
a handful of instances of code like:

var s string
s += "foo"
return s

This CL converts those to simply 'return "foo"'.

The conversion was done mechanically with the
quick-and-dirty cleanup script at
https://gist.github.com/josharian/1fa4408044c163983e62.

I then manually moved a couple of comments in fmt.go.

toolstash -cmp thinks that there are no functional changes.

Change-Id: Ic0ebdd10f0fb8de0360a1041ce5cd10ae1168be9
Reviewed-on: https://go-review.googlesource.com/6265
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

os: fix LinkError creation on windows.

Not only carrying invalid info but also this caused Error to crash with
null pointer exception.

Change-Id: Ibfe63d20eb9b9178ea618e59c74111e9245a6779
Reviewed-on: https://go-review.googlesource.com/6270
Run-TryBot: Hyang-Ah Hana Kim <hyangah@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/internal/obj: make ppc64's CR subregisters print as CRn rather than Cn

These 8 registers are windows into the CR register. They are officially CR0
through CR7 and that is what the assembler accepts, but for some reason
they have always printed as C0 through C7. Fix the naming and printing.

Change-Id: I55822c0322c29d3e01a1f2776b3b210ebf9ded21
Reviewed-on: https://go-review.googlesource.com/6290
Reviewed-by: Russ Cox <rsc@golang.org>

cmd/internal/gc: add missing verb to Yyerror

When a function had no body, Yyerror was called with an extra
argument, leading to extraneous printouts.
Add the missing verb to the Yyerror call and display the name of the
bodiless function.

Fixes #10030

Change-Id: I76d76c4547fb9cad1782cb11f7a5c63065a6e0c5
Reviewed-on: https://go-review.googlesource.com/6263
Reviewed-by: Rob Pike <r@golang.org>

runtime/pprof: skip tests that fork on darwin/arm

Change-Id: I9b08b74214e5a41a7e98866a993b038030a4c073
Reviewed-on: https://go-review.googlesource.com/6251
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>

runtime: trivial clean ups to greyobject

Previously, the typeDead check in greyobject was under a separate
!useCheckmark conditional.  Put it with the rest of the !useCheckmark
code.  Also move a comment about atomic update of the marked bit to
where we actually do that update now.

Change-Id: Ief5f16401a25739ad57d959607b8d81ffe0bc211
Reviewed-on: https://go-review.googlesource.com/6271
Reviewed-by: Rick Hudson <rlh@golang.org>

crypto/x509: skip tests not made for darwin/arm

Change-Id: I8b18dc840425b72d7172a35cb0ba004bd156492d
Reviewed-on: https://go-review.googlesource.com/6252
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>

cmd/objdump: skip fork test on darwin/arm

Change-Id: I1d1eb71014381452d1ef368431cb2556245a35ab
Reviewed-on: https://go-review.googlesource.com/6250
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>

build: restore original timeouts in run.rc

The timeouts were increased in CL 2462 and CL 2510
to work around a slowness issue when running Go
programs on a Plan 9 machine on GCE.

Since we figured out this issue, we can restore
the timeouts to their original values.

Updates #10028.

Change-Id: I2e5b91666461715df69df97ea791f3d88d9de4d0
Reviewed-on: https://go-review.googlesource.com/6261
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

cmd/internal/gc: restore mkbuiltin

Change-Id: I6cc4061fef9d3c3530406ab3d4811192b7175e89
Reviewed-on: https://go-review.googlesource.com/6233
Reviewed-by: Minux Ma <minux@golang.org>

cmd/internal/obj: make Rconv a global function

Clean up the obj API by making Rconv (register pretty printer) a top-level
function. This means that Dconv (operand pretty printer) doesn't need
an Rconv argument.

To do this, we make the register numbers, which are arbitrary inside an
operand (obj.Addr), disjoint sets for each architecture. Each architecture
registers (ha) a piece of the space and then the global Rconv knows which
architecture-specific printer to use.

Clean up all the code that uses Dconv.

Now register numbers are large, so a couple of fields in Addr need to go
from int8 to int16 because they sometimes hold register numbers. Clean
up their uses, which meant regenerating the yacc grammars for the
assemblers. There are changes in this CL triggered by earlier changes
to yacc, which had not been run in this directory.

There is still cleanup to do in Addr, but we're getting closer to that being
easy to do.

Change-Id: I9290ebee013b62f7d24e886743ea5a6b232990ab
Reviewed-on: https://go-review.googlesource.com/6220
Reviewed-by: Russ Cox <rsc@golang.org>

runtime: skip tests that need fork on darwin/arm

Change-Id: I1bb0b8b11e8c7686b85657050fd7cf926afe4d29
Reviewed-on: https://go-review.googlesource.com/6200
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>

syscall: Reimplement linux syscalls in terms of their *at replacements.

Updates #9974

This proposal tackles the body of syscalls which have been replaced,
and are now deprecated in linux. This is needed for the arm64 port as
arm64 is the first linux architecture to remove the "legacy" forms of
these syscalls.

The *AT variants were added in kernel 2.6.16, so well before our 2.6.23
cutoff (hey, it'll even work on RHEL5).

Discussion: https://groups.google.com/forum/#!topic/golang-dev/zpeFtN2z5Fc

Change-Id: I473a7c9a295d6f776fcdc75dcce06cbe9e3564ee
Reviewed-on: https://go-review.googlesource.com/5837
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

syscall: exclude Go tool test on darwin/arm

Change-Id: I44c1911beceaedaa35dad71d8be8a814528dce67
Reviewed-on: https://go-review.googlesource.com/6192
Reviewed-by: Minux Ma <minux@golang.org>

cmd/nm: exclude Go tool test on darwin/arm

Change-Id: I44f1240a766f20de5997faca4f13f96af6da3534
Reviewed-on: https://go-review.googlesource.com/6190
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>

cmd/addr2line: exclude Go tool test on darwin/arm

Change-Id: Icee6c88b7eed5fb27f046373ecf53bf64b68c696
Reviewed-on: https://go-review.googlesource.com/6191
Reviewed-by: Minux Ma <minux@golang.org>

net/http/cgi: skip exec test on darwin/arm

Change-Id: I2fc3cf94b465bf9d7ff8d7bf935b45e334b401e3
Reviewed-on: https://go-review.googlesource.com/6180
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>

cmd/internal/obj: implement Dconv for TYPE_INDIR

It was just missing, and apparently always was.

Change-Id: I84c057bb0ec72940201075f3e6078262fe4bce05
Reviewed-on: https://go-review.googlesource.com/6120
Reviewed-by: Russ Cox <rsc@golang.org>

runtime: fix memory allocator on Plan 9

Previously, the memory allocator on Plan 9 did
not free memory properly. It was only able to
free the last allocated block.

This change implements a variant of the
Kernighan & Ritchie memory allocator with
coalescing and splitting.

The most notable differences are:

- no header is prefixing the allocated blocks, since
  the size is always specified when calling sysFree,
- the free list is nil-terminated instead of circular.

Fixes #9736.
Fixes #9803.
Fixes #9952.

Change-Id: I00d533714e4144a0012f69820d31cbb0253031a3
Reviewed-on: https://go-review.googlesource.com/5524
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>

runtime/debug: fix nacl build

Disable the test properly on nacl systems, tested on nacl/amd64p32.

Change-Id: Iffe210be4f9c426bfc47f2dd3a8f0c6b5a398cc3
Reviewed-on: https://go-review.googlesource.com/6093
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

runtime: disable scavenger on 64k page size kernels

Update #9993

If the physical page size of the machine is larger than the logical
heap size, for example 8k logical, 64k physical, then madvise(2) will
round up the requested amount to a 64k boundary and may discard pages
close to the page being madvised.

This patch disables the scavenger in these situations, which at the moment
is only ppc64 and ppc64le systems. NaCl also uses a 64k page size, but
it's not clear if it is affected by this problem.

Change-Id: Ib897f8d3df5bd915ddc0b510f2fd90a30ef329ca
Reviewed-on: https://go-review.googlesource.com/6091
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>

mime/quotedprintable: create the package

This commit creates the mime/quotedprintable package. It moves and
exports the QP reader of mime/internal/quotedprintable.

The code is almost unchanged to preserve the commit history.

Updates #4943

Change-Id: I4b7b5a2a40a4c84346d42e4cdd2c11a91b28f9e3
Reviewed-on: https://go-review.googlesource.com/5940
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

liblink: delete unused code

Liblink is still needed for the linker (for a bit longer) but mostly not.
Delete the unused parts.

Change-Id: Ie63a7c1520dee52b17425b384943cd16262d36e3
Reviewed-on: https://go-review.googlesource.com/6110
Reviewed-by: Rob Pike <r@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

runtime: disable TestGdbPython on ppc64

(issue #10017)

Change-Id: Ia1267dfdb4474247926a998e32d9c6520015757d
Reviewed-on: https://go-review.googlesource.com/6130
Reviewed-by: Minux Ma <minux@golang.org>
Reviewed-by: Austin Clements <austin@google.com>

cmd/internal/gc: transitive inlining

Inlining refuses to inline bodies containing an actual function call, so that
if that call or a child uses runtime.Caller it cannot observe
the inlining.

However, inlining was also refusing to inline bodies that contained
function calls that were themselves inlined away. For example:

func f() int {
return f1()
}

func f1() int {
return f2()
}

func f2() int {
return 2
}

The f2 call in f1 would be inlined, but the f1 call in f would not,
because f1's call to f2 blocked the inlining, despite itself eventually
being inlined away.

Account properly for this kind of transitive inlining and enable.

Also bump the inlining budget a bit, so that the runtime's
heapBits.next is inlined.

This reduces the time for '6g *.go' in html/template by around 12% (!).
(For what it's worth, closing Chrome reduces the time by about 17%.)

Change-Id: If1aa673bf3e583082dcfb5f223e67355c984bfc1
Reviewed-on: https://go-review.googlesource.com/5952
Reviewed-by: Austin Clements <austin@google.com>

cmd/internal/gc: factor bottom-up visiting out of escape analysis

Change-Id: Id217fb6d8faf045a1a4fbda43b102ba989a02c17
Reviewed-on: https://go-review.googlesource.com/5951
Reviewed-by: Austin Clements <austin@google.com>

cmd/internal/gc: restore detail in plain syntax errors

Change-Id: Ifc4b25fa57d0c9242968246d2193aa29f6b87700
Reviewed-on: https://go-review.googlesource.com/5950
Reviewed-by: Austin Clements <austin@google.com>

math/big: export Float.MinPrec

MinPrec returns the minimum precision required to represent a Float
without loss of precision. Added test.

Change-Id: I466c8e492dcdd59fae854fc4e71ef9b1add7d817
Reviewed-on: https://go-review.googlesource.com/6010
Reviewed-by: Alan Donovan <adonovan@google.com>

os/exec: disable tests on darwin/arm

There is only one process under the iOS sandboxd.

Change-Id: I21b5528366a0248a034801a717f24c60f0733c5f
Reviewed-on: https://go-review.googlesource.com/6101
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

runtime: skip test on darwin/arm

Needs the Go tool, which we do not have on iOS. (No Fork.)

Change-Id: Iedf69f5ca81d66515647746546c9b304c8ec10c4
Reviewed-on: https://go-review.googlesource.com/6102
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>

archive/zip: fix size value in ZIP64 end central directory record

Section 4.3.14.1 of the ZIP file format
spec (https://pkware.cachefly.net/webdocs/casestudies/APPNOTE.TXT) says,

    The value stored into the "size of zip64 end of central directory
    record" should be the size of the remaining record and should not
    include the leading 12 bytes.

We were previously writing the full size, including the 12 bytes.

Fixes #9857

Change-Id: I7cf1fc8457c5f306717cbcf61e02304ab549781f
Reviewed-on: https://go-review.googlesource.com/4760
Reviewed-by: Andrew Gerrand <adg@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

go/build: disable deps test on darwin/arm

Change-Id: Ief78a10c4aaa43f300f34519911ff73b6f510d73
Reviewed-on: https://go-review.googlesource.com/6100
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>

runtime: do not do futile netpolls

There is no sense in trying to netpoll while there is
already a thread blocked in netpoll. And in most cases
there must be a thread blocked in netpoll, because
the first otherwise idle thread does blocking netpoll.

On some program I see that netpoll called from findrunnable
consumes 3% of time.

Change-Id: I0af1a73d637bffd9770ea50cb9278839716e8816
Reviewed-on: https://go-review.googlesource.com/4553
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Dmitry Vyukov <dvyukov@google.com>

runtime: simplify CPU profiling code

This makes Go's CPU profiling code somewhat more idiomatic; e.g.,
using := instead of forward declaring variables, using "int" for
element counts instead of "uintptr", and slices instead of C-style
pointer+length.  This makes the code easier to read and eliminates a
lot of type conversion clutter.

Additionally, in sigprof we can collect just maxCPUProfStack stack
frames, as cpuprof won't use more than that anyway.

Change-Id: I0235b5ae552191bcbb453b14add6d8c01381bd06
Reviewed-on: https://go-review.googlesource.com/6072
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>

image/jpeg: support 4:1:1 and 4:1:0 chroma subsampling.

The test data was generated by:
convert video-001.png tmp.tga
cjpeg -quality 50 -sample 4x2,1x1,1x1 tmp.tga > video-001.q50.410.jpeg
cjpeg -quality 50 -sample 4x1,1x1,1x1 tmp.tga > video-001.q50.411.jpeg
cjpeg -quality 50 -sample 4x2,1x1,1x1 -progressive tmp.tga > video-001.q50.410.progressive.jpeg
cjpeg -quality 50 -sample 4x1,1x1,1x1 -progressive tmp.tga > video-001.q50.411.progressive.jpeg
rm tmp.tga

Change-Id: I5570389c462360f98c3160f3c6963d9466d511de
Reviewed-on: https://go-review.googlesource.com/6041
Reviewed-by: Rob Pike <r@golang.org>

cmd/gc: reduce lexer allocs when parsing numeric constants

This reduces the number of allocs when
running the rotate.go tests by
about 20%, after applying CL 5700.

Combining

s = "const str"
s += <another string>

generally saves an alloc and might be a candidate for
rsc's grind tool. However, I'm sending this CL now
because this also reuses the result of calling lexbuf.String.

Change-Id: If3a7300b7da9612ab62bb910ee90349dca88dde3
Reviewed-on: https://go-review.googlesource.com/5821
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/gc: don't call memequal twice in generated type.eq routines

The first call is pointless. It appears to simply be a mistake.

benchmark                  old ns/op     new ns/op     delta
BenchmarkComplexAlgMap     90.7          76.1          -16.10%

Change-Id: Id0194c9f09cea8b68f17b2ac751a8e3240e47f19
Reviewed-on: https://go-review.googlesource.com/5284
Reviewed-by: Keith Randall <khr@golang.org>

math/big: fix build for 32bit platforms

Change-Id: I9c217e5140294a17e4feb65da5b121ee8d8cadc2
Reviewed-on: https://go-review.googlesource.com/6050
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

image: add YCbCrSubsampleRatio411 and YCbCrSubsampleRatio410.

Some real world JPEG images are in 4:1:1 and 4:1:0 formats.

See also http://en.wikipedia.org/wiki/Chroma_subsampling

Change-Id: I2d51a41944f581cf11f4ab975046b1737271842f
Reviewed-on: https://go-review.googlesource.com/5838
Reviewed-by: Rob Pike <r@golang.org>

encoding/xml: add more marshalTests tests.

There are no behavior changes in this CL, only specifying the status
quo. A follow-up CL, https://go-review.googlesource.com/#/c/5910/, will
change marshaling behavior.

Change-Id: Ib3f4d62e8c4758da2f11a6d26b285c10d3b0d98a
Reviewed-on: https://go-review.googlesource.com/6040
Reviewed-by: Andrew Gerrand <adg@golang.org>

os: Use GetComputerNameEx to get Hostname on win32

The existing Hostname function uses the GetComputerName system
function in windows to determine the hostname. It has some downsides:

  - The name is limited to 15 characters.
  - The name returned is for NetBIOS, other OS's return a DNS name

This change adds to the internal/syscall/windows package a
GetComputerNameEx function, and related enum constants. They are used
instead of the syscall.ComputerName function to implement os.Hostname
on windows.

Fixes #9982

Change-Id: Idc8782785eb1eea37e64022bd201699ce9c4b39c
Reviewed-on: https://go-review.googlesource.com/5852
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
Reviewed-by: Carlos Castillo <cookieo9@gmail.com>
Reviewed-by: Yasuhiro MATSUMOTO <mattn.jp@gmail.com>

math/big: clean up Float.SetPrec, use shorter internal representation

Change-Id: I9b78085adc12cbd240d0b8b48db6810ddb2aeadd
Reviewed-on: https://go-review.googlesource.com/5991
Reviewed-by: Alan Donovan <adonovan@google.com>

math/big: apply a round of go vet

Change-Id: Ie8310acc783659497e50dfe629f06d655b51d647
Reviewed-on: https://go-review.googlesource.com/5990
Reviewed-by: Alan Donovan <adonovan@google.com>

math/big: improve some doc strings

Change-Id: Ie37673d4af2fa7476d67ffb686641611ab6a8e6b
Reviewed-on: https://go-review.googlesource.com/5930
Reviewed-by: Alan Donovan <adonovan@google.com>

runtime/cgo: fix darwin/arm build

Macro definition ordering.

Change-Id: I0def4702d19a21a68ffa52ea5b7c22578830c578
Reviewed-on: https://go-review.googlesource.com/6030
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>

runtime/cgo: set the initial working directory

Gives tests a way to find the bundle that contains their testdata, and
is generally useful for finding resources.

Change-Id: Idfa03e8543af927c17bc8ec8aadc5014ec82df28
Reviewed-on: https://go-review.googlesource.com/6000
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>

runtime: skip failing gdb test on linux/arm

Updates #10002

The gdb test added in 1c82e236f5ee is failing on most arm systems.

Temporarily disable this test so that we can return to a working arm build.

Change-Id: Iff96ea8d5a99e1ceacf4979e864ff196e5503535
Reviewed-on: https://go-review.googlesource.com/5902
Reviewed-by: Ian Lance Taylor <iant@golang.org>

runtime: fix build, divide by constant 0 is a compile-time error

Change-Id: Iee319c9f5375c172fb599da77234c10ccb0fd314
Reviewed-on: https://go-review.googlesource.com/6020
Reviewed-by: Keith Randall <khr@golang.org>

doc/install-source.html: fix cd command, already in go dir.

Change-Id: I42929abe5cefd84a1daf4f36ee1806c490ce16ec
Reviewed-on: https://go-review.googlesource.com/5960
Reviewed-by: Minux Ma <minux@golang.org>

runtime: mark pages we return to kernel as NOHUGEPAGE

We return memory to the kernel with madvise(..., DONTNEED).
Also mark returned memory with NOHUGEPAGE to keep the kernel from
merging this memory into a huge page, effectively reallocating it.

Only known to be a problem on linux/{386,amd64,amd64p32} at the moment.
It may come up on other os/arch combinations in the future.

Fixes #8832

Change-Id: Ifffc6627a0296926e3f189a8a9b6e4bdb54c79eb
Reviewed-on: https://go-review.googlesource.com/5660
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>

runtime: handle holes in the heap

We need to distinguish pointers to free spans, which indicate bugs in
our pointer analysis, from pointers to never-in-the-heap spans, which
can legitimately arise from sysAlloc/mmap/etc.  This normally isn't a
problem because the heap is contiguous, but in some situations (32
bit, particularly) the heap must grow around an already allocated
region.

The bad pointer test is disabled so this fix doesn't actually do
anything, but it removes one barrier from reenabling it.

Fixes #9872.

Change-Id: I0a92db4d43b642c58d2b40af69c906a8d9777f88
Reviewed-on: https://go-review.googlesource.com/5780
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>

cmd/internal/obj: make Dconv a portable top-level function

Each architecture had its own Dconv (operand printer) but the syntax is
close to uniform and the code overlap was considerable. Consolidate these
into a single top-level function. A similar but smaller unification is done
for Mconv ("Name" formatter) as well.

The signature is changed. The flag was unused so drop it. Add a
function argument, Rconv, that must be supplied by the caller.
TODO: A future change will unify Rconv as well and this argument
will go away.

Some formats changed, because of the automatic consistency
created by unification. For instance, 0(R1) always prints as (R1)
now, and foo+0(SB) is just foo(SB). Before, some made these
simplifications and some didn't; now they all do.

Update the asm tests that depend on the format.

Change-Id: I6e3310bc19814c0c784ff0b960a154521acd9532
Reviewed-on: https://go-review.googlesource.com/5920
Reviewed-by: Russ Cox <rsc@golang.org>

cmd/internal/gc: omit non-explicit capacity in errors with map/chan make

Fixes #9083.

Change-Id: Ifbdebafb39a73a1dacf7e67171e8e88028d1f10b
Reviewed-on: https://go-review.googlesource.com/1219
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Chris Manghane <cmang@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

runtime: fallback to 128M address space on 32bit

Available darwin/arm devices sporadically have trouble mapping 256M.

I would really appreciate it if anyone could check my working on
this, and make sure sure there aren't obviously bad consequences I
haven't considered.

Change-Id: Id1a8edae104d974fcf5f9333274f958625467f79
Reviewed-on: https://go-review.googlesource.com/5752
Reviewed-by: Keith Randall <khr@golang.org>

cmd/internal/gc: do not show computed value in type error

Fixes #9076.

Change-Id: Ib41a452fa9aa9fecf19f65c36d13715923548041
Reviewed-on: https://go-review.googlesource.com/1250
Reviewed-by: Minux Ma <minux@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Chris Manghane <cmang@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/internal/gc: method selector should not auto-deref named pointer type

Fixes #9017.

Change-Id: I26cb1e7d6e137ff145773169cfe2d8bd4e1b339c
Reviewed-on: https://go-review.googlesource.com/1252
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Chris Manghane <cmang@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

cmd/internal/obj: fix index panic on invalid instruction

Also introduce actual data structure for table.

Change-Id: I6bbe9aff8a872ae254f3739ae4ca17f7b5c4507a
Reviewed-on: https://go-review.googlesource.com/5701
Reviewed-by: Rob Pike <r@golang.org>

cmd/internal/gc: expand DBG macro in lex.go

The dummy implementation was causing lots of argument lists
to be prepared and thrown away.

Change-Id: Id0040dec6b0937f3daa8a8d8911fa3280123e863
Reviewed-on: https://go-review.googlesource.com/5700
Reviewed-by: Rob Pike <r@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>

math/big: permit passing of an *Int to Float.Int to avoid allocation

Change-Id: I50e83248357928e56c94b88a8764de828f4f5c76
Reviewed-on: https://go-review.googlesource.com/5890
Reviewed-by: Alan Donovan <adonovan@google.com>

cmd/go: make asm the first assembler

verifyAsm is still on, but this CL changes the order to asm then 6a.
Before, it was 6a then asm, but that meant that any bugs in asm
for bad input would be prevented from happening because 6a would
catch them. Now asm gets first crack, as it must.

Also implement the -trimpath flag in asm. It's necessary and trivial.

Change-Id: Ifb2ab870de1aa1b53dec76a78ac697a0d36fa80a
Reviewed-on: https://go-review.googlesource.com/5850
Reviewed-by: Russ Cox <rsc@golang.org>

math/big: implemented Float.Rat

Change-Id: If516e12d4b5dfb6f9288437d270569f7e4e2a1cd
Reviewed-on: https://go-review.googlesource.com/5871
Reviewed-by: Alan Donovan <adonovan@google.com>

cmd/asm: add end to end test for 386

Change-Id: I6514f69b979d064b6a3c4b5d0828cc94f485cac2
Reviewed-on: https://go-review.googlesource.com/5694
Reviewed-by: Russ Cox <rsc@golang.org>

cmd/asm: add a couple of operand parses discovered by end-to-end test

Missing cases for JMP $4 and foo+4(SB):AX. Both are odd but 8a accepts them
and they seem valid.

Change-Id: Ic739f626fcc79ace1eaf646c5dfdd96da59df165
Reviewed-on: https://go-review.googlesource.com/5693
Reviewed-by: Russ Cox <rsc@golang.org>

runtime: simplify gcResetGState

Since allglock is held in this function, there's no point to
tip-toeing around allgs.  Just use a for-range loop.

Change-Id: I1ee61c7e8cac8b8ebc8107c0c22f739db5db9840
Reviewed-on: https://go-review.googlesource.com/5882
Reviewed-by: Russ Cox <rsc@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>

runtime: consolidate gcworkdone/gcscanvalid clearing loops

Previously, we had three loops in the garbage collector that all
cleared the per-G GC flags.  Consolidate these into one function.
This one function is designed to work in a concurrent setting.  As a
result, it's slightly more expensive than the loops it replaces during
STW phases, but these happen at most twice per GC.

Change-Id: Id1ec0074fd58865eb0112b8a0547b267802d0df1
Reviewed-on: https://go-review.googlesource.com/5881
Reviewed-by: Russ Cox <rsc@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>

runtime: remove unnecessary gcworkdone resetting loop

The loop in gcMark is redundant with the gcworkdone resetting
performed by markroot, which called a few lines later in gcMark.

Change-Id: Ie0a826a614ecfa79e6e6b866e8d1de40ba515856
Reviewed-on: https://go-review.googlesource.com/5880
Reviewed-by: Russ Cox <rsc@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>

runtime: remove gogetcallerpc and gogetcallersp functions

Package runtime's Go code was converted to directly call getcallerpc
and getcallersp in https://golang.org/cl/138740043, but the assembly
implementations were not removed.

Change-Id: Ib2eaee674d594cbbe799925aae648af782a01c83
Reviewed-on: https://go-review.googlesource.com/5901
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>

runtime: simplify NetBSD semaphores

NetBSD's semaphore implementation is derived from OpenBSD's, but has
subsequently diverged due to cleanups that were only applied to the
latter (https://golang.org/cl/137960043, https://golang.org/cl/5563).
This CL applies analogous cleanups for NetBSD.

Notably, we can also remove the scary NetBSD deadlock warning.
NetBSD's manual pages document that lwp_unpark on a not-yet-parked LWP
will cause that LWP's next lwp_park system call to return immediately,
so there's no race hazard.

Change-Id: Ib06844c420d2496ac289748eba13eb4700bbbbb2
Reviewed-on: https://go-review.googlesource.com/5564
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Joel Sing <jsing@google.com>

syscall: generate AT_* constants and regenerate ztypes_linux_*.go

Updates #9974

The *at family of syscalls requires some constants to be defined in the
syscall package for linux. Add the necessary constants and regenerate
the ztypes_linux_*.go files.

Change-Id: I6df343fef7bcacad30d36c7900dbfb621465a4fe
Reviewed-on: https://go-review.googlesource.com/5836
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

gdb: fix map prettyprinter

(gdb) p x
Python Exception <class 'gdb.error'> There is no member named b.:
$2 = map[string]string
->
(gdb) p x
$1 = map[string]string = {["shane"] = "hansen"}

Change-Id: I874d02a029f2ac9afc5ab666afb65760ec2c3177
Reviewed-on: https://go-review.googlesource.com/5522
Reviewed-by: Ian Lance Taylor <iant@golang.org>

runtime: simplify OpenBSD semaphores

OpenBSD's thrsleep system call includes an "abort" parameter, which
specifies a memory address to be tested after being registered on the
sleep channel (i.e., capable of being woken up by thrwakeup).  By
passing a pointer to waitsemacount for this parameter, we avoid race
conditions without needing a lock.  Instead we just need to use
atomicload, cas, and xadd to mutate the semaphore count.

Change-Id: If9f2ab7cfd682da217f9912783cadea7e72283a8
Reviewed-on: https://go-review.googlesource.com/5563
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Joel Sing <jsing@google.com>

syscall: split implementation of Dup2 per linux GOOS

Updates #9974

This proposal moves the definition of Dup2 from the generic syscall_linux.go
to the GOOS specific variants. This is in preparation for the arm64 port.

For all existing platforms Dup2 is not affected. When arm64 is added we'll use
either a forwarding method to Dup3 or

//sysnb        Dup2(oldfd int, newfd int) (err error) = SYS_DUP3

Because mksycall.pl does not sort symbols before generating the output file
the diff includes some unavoidable code moves as Dup2 is processed latter in
the run.

Discussion: https://groups.google.com/forum/#!topic/golang-dev/zpeFtN2z5Fc

Change-Id: Icdedf55bb29e749c4230e1ee371bf9d0bd0cfb38
Reviewed-on: https://go-review.googlesource.com/5835
Reviewed-by: Minux Ma <minux@golang.org>
Run-TryBot: Dave Cheney <dave@cheney.net>

doc/go1.5: mention the change in certificate wildcard handling.

https://go-review.googlesource.com/#/c/5691/ shouldn't affect anyone
because it reflects the RFC and behaviour of modern browsers, but it's
probably worth mentioning.

Change-Id: I8a8eedaac5226b1d8018dc56de6b2d41bd82a642
Reviewed-on: https://go-review.googlesource.com/5862
Reviewed-by: Rob Pike <r@golang.org>

math/big: permit passing of (possibly nil) *Float to MantExp to avoid allocation

Change-Id: Ia92eea833283f8b16fa09d4ca1c9cb3bc0eb18a2
Reviewed-on: https://go-review.googlesource.com/5870
Reviewed-by: Rob Pike <r@golang.org>

syscall: split implementation of Pipe/Pipe2 per GOOS

Updates #9974

This proposal moves the definition of Pipe an Pipe2 from the generic
syscall_linux.go to the GOOS specific variants. This is in preparation
for the arm64 port.

For platforms where pipe2(2) is not supported in the minimum 2.6.23 kernel,
amd64 and 386, we retain pipe(2). For all other platforms pipe(2) is removed
and Pipe forwards to pipe2(2).

Because mksycall.pl does not sort symbols before generating the output file
the diff includes some unavoidable code moves as Pipe and Pipe2 are processed
latter in the run.

Discussion: https://groups.google.com/forum/#!topic/golang-dev/zpeFtN2z5Fc

Change-Id: Ie26d6761eeb9760dbaff974ee8bc0d57a9ceaee4
Reviewed-on: https://go-review.googlesource.com/5833
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

cmd/dist: respect runtime.NumCPU when bootstrapping arm hosts

This is a reproposal of CL 2957. This reproposal restricts the
scope of this change to just arm systems.

With respect to rsc's comments on 2957, on all my arm hosts they perform
the build significantly faster with this change in place.

Change-Id: Ie09be1a73d5bb777ec5bca3ba93ba73d5612d141
Reviewed-on: https://go-review.googlesource.com/5834
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

runtime: reset gcscanvalid and gcworkdone when GODEBUG=gctrace=2

When GODEBUG=gctrace=2 two gcs are preformed. During the first gc
the stack scan sets the g's gcscanvalid and gcworkdone flags to true
indicating that the stacks have to be scanned and do not need to
be rescanned. These need to be reset to false for the second GC so the
stacks are rescanned, otherwise if the only pointer to an object is
on the stack it will not be discovered and the object will be freed.
Typically this will include the object that was just allocated in
the mallocgc call that initiated the GC.

Change-Id: Ic25163f4689905fd810c90abfca777324005c02f
Reviewed-on: https://go-review.googlesource.com/5861
Reviewed-by: Russ Cox <rsc@golang.org>

syscall: rebuild zsyscall_linux_*.go

Rebuild the zsyscall_linux_*.go files in preperation for #9974

The only change is the ppc64/ppc64le files which were not rebuilt when
syscall.use was added.

Change-Id: I804c63731e4900c782025de04ea3585d99688958
Reviewed-on: https://go-review.googlesource.com/5831
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

math/big: fix test for 32bit platforms (fix build)

Change-Id: I73509cd0e0866cfe2b2ae46a26fb4d043dd050c4
Reviewed-on: https://go-review.googlesource.com/5840
Reviewed-by: Alan Donovan <adonovan@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

go/ast, go/parser: correct End() position for *ast.EmptyStmt

- added a new field ast.EmptyStmt.Implicit to indicate explicit
  or implicit semicolon
- fix ast.EmptyStmt.End() accordingly
- adjusted parser and added test case

Fixes #9979.

Change-Id: I72b0983b3a0cabea085598e1bf6c8df629776b57
Reviewed-on: https://go-review.googlesource.com/5720
Reviewed-by: Russ Cox <rsc@golang.org>

math/big: change Float.SetMantExp to always multiply mant by 2**exp

Change-Id: If840e647376a2141f8c17729f7ef251bfff13f5f
Reviewed-on: https://go-review.googlesource.com/5810
Reviewed-by: Rob Pike <r@golang.org>

cmd/asm: fix build for new aliases

Missing leading A on names.

Change-Id: I6f3a66bdd3a21220f45a898f0822930b6a7bfa38
Reviewed-on: https://go-review.googlesource.com/5801
Reviewed-by: Rob Pike <r@golang.org>

cmd/asm: add alias for MOVOA=MOVO on amd64

The alias should exist for both 386 and amd64.
There were a few others missing as well. Add them.

Change-Id: Ia0c3e71abc79f67a7a66941c0d932a8d5d6e9989
Reviewed-on: https://go-review.googlesource.com/5800
Reviewed-by: Russ Cox <rsc@golang.org>

crypto/x509: make behaviour of absolute DNS names match Chromium.

Previously, we didn't handle absolute DNS names in certificates the same
way as Chromium, and we probably shouldn't diverge from major browsers.

Change-Id: I56a3962ad1002f68b5dbd65ae90991b82c2f5629
Reviewed-on: https://go-review.googlesource.com/5692
Reviewed-by: Andrew Gerrand <adg@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

crypto/x509: allow wildcards only as the first label.

RFC 6125 now specifies that wildcards are only allowed for the leftmost
label in a pattern: https://tools.ietf.org/html/rfc6125#section-6.4.3.

This change updates Go to match the behaviour of major browsers in this
respect.

Fixes #9834.

Change-Id: I37c10a35177133624568f2e0cf2767533926b04a
Reviewed-on: https://go-review.googlesource.com/5691
Reviewed-by: Andrew Gerrand <adg@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

crypto/tls: allow larger initial records.

Some servers which misunderstood the point of the CertificateRequest
message send huge reply records. These records are large enough that
they were considered “insane” by the TLS code and rejected.

This change removes the sanity test for record lengths. Although the
maxCiphertext test still remains, just above, which (roughly) enforces
the 16KB protocol limit on record sizes:
https://tools.ietf.org/html/rfc5246#section-6.2.1

Fixes #8928.

Change-Id: Idf89a2561b1947325b7ddc2613dc2da638d7d1c9
Reviewed-on: https://go-review.googlesource.com/5690
Reviewed-by: Andrew Gerrand <adg@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

crypto/x509: don't reject certs with critical policy extensions.

There was a missing continue that caused certificates with critical
certificate-policy extensions to be rejected. Additionally, that code
structure in general was prone to exactly that bug so I changed it
around to hopefully be more robust in the future.

Fixes #9964.

Change-Id: I58fc6ef3a84c1bd292a35b8b700f44ef312ec1c1
Reviewed-on: https://go-review.googlesource.com/5670
Reviewed-by: Andrew Gerrand <adg@golang.org>

math/big: fix several issues with string->Float conversion

Change-Id: I7bf7154e2d8d779fdf7f1d2bb561a06ad174f3b0
Reviewed-on: https://go-review.googlesource.com/4883
Reviewed-by: Alan Donovan <adonovan@google.com>

cmd/go: limit default darwin/arm parallelism

iOS devices can only run tests serially.

Change-Id: I3f4e7abddf812a186895d9d5138999c8bded698f
Reviewed-on: https://go-review.googlesource.com/5751
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>

sync: add active spinning to Mutex

Currently sync.Mutex is fully cooperative. That is, once contention is discovered,
the goroutine calls into scheduler. This is suboptimal as the resource can become
free soon after (especially if critical sections are short). Server software
usually runs at ~~50% CPU utilization, that is, switching to other goroutines
is not necessary profitable.

This change adds limited active spinning to sync.Mutex if:
1. running on a multicore machine and
2. GOMAXPROCS>1 and
3. there is at least one other running P and
4. local runq is empty.
As opposed to runtime mutex we don't do passive spinning,
because there can be work on global runq on on other Ps.

benchmark                   old ns/op     new ns/op     delta
BenchmarkMutexNoSpin        1271          1272          +0.08%
BenchmarkMutexNoSpin-2      702           683           -2.71%
BenchmarkMutexNoSpin-4      377           372           -1.33%
BenchmarkMutexNoSpin-8      197           190           -3.55%
BenchmarkMutexNoSpin-16     131           122           -6.87%
BenchmarkMutexNoSpin-32     170           164           -3.53%
BenchmarkMutexSpin          4724          4728          +0.08%
BenchmarkMutexSpin-2        2501          2491          -0.40%
BenchmarkMutexSpin-4        1330          1325          -0.38%
BenchmarkMutexSpin-8        684           684           +0.00%
BenchmarkMutexSpin-16       414           372           -10.14%
BenchmarkMutexSpin-32       559           469           -16.10%

BenchmarkMutex                 19.1          19.1          +0.00%
BenchmarkMutex-2               81.6          54.3          -33.46%
BenchmarkMutex-4               143           100           -30.07%
BenchmarkMutex-8               154           156           +1.30%
BenchmarkMutex-16              140           159           +13.57%
BenchmarkMutex-32              141           163           +15.60%
BenchmarkMutexSlack            33.3          31.2          -6.31%
BenchmarkMutexSlack-2          122           97.7          -19.92%
BenchmarkMutexSlack-4          168           158           -5.95%
BenchmarkMutexSlack-8          152           158           +3.95%
BenchmarkMutexSlack-16         140           159           +13.57%
BenchmarkMutexSlack-32         146           162           +10.96%
BenchmarkMutexWork             154           154           +0.00%
BenchmarkMutexWork-2           89.2          89.9          +0.78%
BenchmarkMutexWork-4           139           86.1          -38.06%
BenchmarkMutexWork-8           177           162           -8.47%
BenchmarkMutexWork-16          170           173           +1.76%
BenchmarkMutexWork-32          176           176           +0.00%
BenchmarkMutexWorkSlack        160           160           +0.00%
BenchmarkMutexWorkSlack-2      103           99.1          -3.79%
BenchmarkMutexWorkSlack-4      155           148           -4.52%
BenchmarkMutexWorkSlack-8      176           170           -3.41%
BenchmarkMutexWorkSlack-16     170           173           +1.76%
BenchmarkMutexWorkSlack-32     175           176           +0.57%

"No work" benchmarks are not very interesting (BenchmarkMutex and
BenchmarkMutexSlack), as they are absolutely not realistic.

Fixes #8889

Change-Id: I6f14f42af1fa48f73a776fdd11f0af6dd2bb428b
Reviewed-on: https://go-review.googlesource.com/5430
Reviewed-by: Rick Hudson <rlh@golang.org>
Run-TryBot: Dmitry Vyukov <dvyukov@google.com>