]> Cypherpunks repositories - gostls13.git/log
gostls13.git
7 years agoruntime: allow more CPUs on FreeBSD
Austin Clements [Tue, 26 Sep 2017 15:18:37 +0000 (11:18 -0400)]
runtime: allow more CPUs on FreeBSD

Currently the FreeBSD CPU affinity code assumes that the maximum
GOMAXPROCS is 256, but we just removed that limit.

This commit rewrites the FreeBSD CPU affinity code to raise the CPU
count limit to 65,536, like the Linux CPU affinity code, and to
degrade more gracefully if we do somehow go over that.

Change-Id: Ic4ca7f88bd8b9448aae4dbd43ef21a6c1b8fea63
Reviewed-on: https://go-review.googlesource.com/66291
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoruntime: clean up loops over allp
Austin Clements [Tue, 13 Jun 2017 16:01:56 +0000 (12:01 -0400)]
runtime: clean up loops over allp

allp now has length gomaxprocs, which means none of allp[i] are nil or
in state _Pdead. This lets replace several different styles of loops
over allp with normal range loops.

for i := 0; i < gomaxprocs; i++ { ... } loops can simply range over
allp. Likewise, range loops over allp[:gomaxprocs] can just range over
allp.

Loops that check for p == nil || p.state == _Pdead don't need to check
this any more.

Loops that check for p == nil don't have to check this *if* dead Ps
don't affect them. I checked that all such loops are, in fact,
unaffected by dead Ps. One loop was potentially affected, which this
fixes by zeroing p.gcAssistTime in procresize.

Updates #15131.

Change-Id: Ifa1c2a86ed59892eca0610360a75bb613bc6dcee
Reviewed-on: https://go-review.googlesource.com/45575
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agoruntime: eliminate GOMAXPROCS limit
Austin Clements [Tue, 13 Jun 2017 15:33:57 +0000 (11:33 -0400)]
runtime: eliminate GOMAXPROCS limit

Now that allp is dynamically allocated, there's no need for a hard cap
on GOMAXPROCS.

Fixes #15131.

Change-Id: I53eee8e228a711a818f7ebce8d9fd915b3865eed
Reviewed-on: https://go-review.googlesource.com/45574
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agoruntime: dynamically allocate allp
Austin Clements [Tue, 13 Jun 2017 15:32:17 +0000 (11:32 -0400)]
runtime: dynamically allocate allp

This makes it possible to eliminate the hard cap on GOMAXPROCS.

Updates #15131.

Change-Id: I4c422b340791621584c118a6be1b38e8a44f8b70
Reviewed-on: https://go-review.googlesource.com/45573
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agosyscall: correct TCGETS/TCSETS values on ppc64/ppc64le
Tobias Klauser [Wed, 27 Sep 2017 12:26:20 +0000 (14:26 +0200)]
syscall: correct TCGETS/TCSETS values on ppc64/ppc64le

Correcting values is allowed per the syscall package rules, so update
these constants to their correct value on ppc64/ppc64le. The values now
match the corresponding constants in x/sys/unix.

Update #19560
Fixes #22000

Change-Id: I1d358de345766ec96e15dfcc8911fe2f39fb0ddb
Reviewed-on: https://go-review.googlesource.com/66510
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agopath/filepath: add example for Walk
Jean-Francois Cantin [Tue, 26 Sep 2017 22:07:27 +0000 (16:07 -0600)]
path/filepath: add example for Walk

Fixes: #22052
Change-Id: Ia056871b35ecc1a8c5ac891402fc1c5702731623
Reviewed-on: https://go-review.googlesource.com/66330
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agotime: unify three readFile implementations into one
Ian Lance Taylor [Tue, 26 Sep 2017 17:50:57 +0000 (10:50 -0700)]
time: unify three readFile implementations into one

Undoes this part of https://golang.org/cl/5447061 by using the
OS-specific open and close functions, and adding a read function.

Change-Id: If37ef43eb5df8554fc03f3922bbc2f785129bb9c
Reviewed-on: https://go-review.googlesource.com/66271
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Joe Tsai <joetsai@google.com>
7 years agotime: reduce OS dependent timezone related code
Florian Uekermann [Mon, 18 Sep 2017 17:22:29 +0000 (19:22 +0200)]
time: reduce OS dependent timezone related code

Loading and testing timezones is currently implemented using several,
partly redundant, OS specific data structures and functions. This
change merges most of that code into OS independent implementations.

Change-Id: Iae2877c5f48d1e4a9de9ce55d0530d52e24cf96e
Reviewed-on: https://go-review.googlesource.com/64391
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoencoding/json: remove superfluous comment
Marvin Stenger [Thu, 21 Sep 2017 19:03:42 +0000 (21:03 +0200)]
encoding/json: remove superfluous comment

Remove an old comment introduced in golang.org/cl/9073.

Change-Id: I14be27ddfac987f44d839920bc4d02361a576f06
Reviewed-on: https://go-review.googlesource.com/66371
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agonet/mail: parse group in email address
Mihail Minaev [Tue, 26 Sep 2017 11:47:49 +0000 (11:47 +0000)]
net/mail: parse group in email address

This change adds the ability to parse
group into email address. The information about
group name and group members is lost for
backwards compatibility. According to this rule address
`Group: Test <text@example.com>;` would be parsed into
`Test <test@example.com>`.

Fixes #22014

Change-Id: I6e804a62f3ede04f555a1b82500b8ca030eeb431
Reviewed-on: https://go-review.googlesource.com/66250
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agonet: bump TestDialerDualStackFDLeak timeout on iOS
Elias Naur [Wed, 27 Sep 2017 11:36:54 +0000 (13:36 +0200)]
net: bump TestDialerDualStackFDLeak timeout on iOS

On an iPhone 6 running iOS 11, the TestDialerDualStackFDLeak test
started failing with dial durations just above the limit:

FAIL: TestDialerDualStackFDLeak (0.21s)

dial_test.go:90: got 101.154ms; want <= 95ms

Bump the timeout on iOS.

For the iOS builder.

Change-Id: Id42b471e7cf7d0c84f6e83ed04b395fa1a2d449d
Reviewed-on: https://go-review.googlesource.com/66491
Run-TryBot: Elias Naur <elias.naur@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/go/internal/work: skip TestRespectSetgidDir on iOS
Elias Naur [Wed, 27 Sep 2017 11:26:47 +0000 (13:26 +0200)]
cmd/go/internal/work: skip TestRespectSetgidDir on iOS

Attempting to set the SetGID bit on iOS 11 fails with a permission
error. Skip the test.

Change-Id: Idac59750447d668091c44fe7cc5ee178014e0e1d
Reviewed-on: https://go-review.googlesource.com/66490
Run-TryBot: Elias Naur <elias.naur@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/go: delete unused function isInGoToolsRepo
Marvin Stenger [Thu, 21 Sep 2017 15:14:59 +0000 (17:14 +0200)]
cmd/go: delete unused function isInGoToolsRepo

The function isInGoToolsRepo has no use case anymore, so we remove it
with this change.

Change-Id: I71051828eef2e317b83e13f83a14d5f0bc0ec13f
Reviewed-on: https://go-review.googlesource.com/66350
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agodebug/macho: fill Rpath.LoadBytes in NewFile
Hiroshi Ioka [Tue, 26 Sep 2017 23:46:08 +0000 (08:46 +0900)]
debug/macho: fill Rpath.LoadBytes in NewFile

Also, fix some error messages.

Fixes #22065

Change-Id: Iac05c24b7bb128be3f43b8f2aa180b3957d5ee72
Reviewed-on: https://go-review.googlesource.com/66390
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agodebug/macho: parse relocations
Hiroshi Ioka [Thu, 21 Sep 2017 01:53:08 +0000 (10:53 +0900)]
debug/macho: parse relocations

Fixes #21957

Change-Id: I69ef9e257aa2b7b6c4fc4c115e99f8a7f93d8d9c
Reviewed-on: https://go-review.googlesource.com/65150
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agosyscall: make Exit call runtime.exit
Alex Brainman [Tue, 26 Sep 2017 02:35:54 +0000 (12:35 +1000)]
syscall: make Exit call runtime.exit

syscall.Exit and runtime.exit do the same thing.
Why duplicate code?

CL 45115 fixed bug where windows runtime.exit was correct,
but syscall.Exit was broken. So CL 45115 fixed windows
syscall.Exit by calling runtime.exit.

Austin suggested that all OSes should do the same, and
this CL implements his idea.

While making changes, I discovered that nacl syscall.Exit
returned error

func Exit(code int) (err error)

and I changed it into

func Exit(code int)

like all other OSes. I assumed it was a mistake and it
is OK to do because cmd/api does not complain about it.

Also I changed plan9 runtime.exit to accept int32 just
like all other OSes do.

Change-Id: I12f6022ad81406566cf9befcc6edc382eebd413b
Reviewed-on: https://go-review.googlesource.com/66170
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Reviewed-by: David du Colombier <0intro@gmail.com>
7 years agoall: prefer bytes.IndexByte over bytes.Index
Marvin Stenger [Thu, 21 Sep 2017 17:23:51 +0000 (19:23 +0200)]
all: prefer bytes.IndexByte over bytes.Index

bytes.IndexByte can be used wherever the second argument to
strings.Index is exactly one byte long, so we do that with this change.

This avoids generating unnecessary string symbols/converison and saves
a few calls to bytes.Index.

Change-Id: If31c775790e01edfece1169e398ad6a754fb4428
Reviewed-on: https://go-review.googlesource.com/66373
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoall: prefer strings.LastIndexByte over strings.LastIndex
Marvin Stenger [Mon, 25 Sep 2017 13:47:44 +0000 (15:47 +0200)]
all: prefer strings.LastIndexByte over strings.LastIndex

strings.LastIndexByte was introduced in go1.5 and it can be used
effectively wherever the second argument to strings.LastIndex is
exactly one byte long.

This avoids generating unnecessary string symbols and saves
a few calls to strings.LastIndex.

Change-Id: I7b5679d616197b055cffe6882a8675d24a98b574
Reviewed-on: https://go-review.googlesource.com/66372
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agonet/http: error if Transport.Proxy returns https
Daniel Martí [Mon, 25 Sep 2017 16:21:39 +0000 (17:21 +0100)]
net/http: error if Transport.Proxy returns https

Transport.Proxy is documented as only supporting the http and socks5
schemes. If one tries to use it for https URLs, they end up with a
cryptic error like:

http: TLS handshake error from [...]: tls: oversized record received with length 20037

This is because Transport simply skips TLS if Proxy is non-nil, since it
knows it doesn't support Proxy with https.

However, that error is very confusing and it can take a while to figure
out what's going on. Instead, error if Proxy is used and it returns an
unsupported scheme.

Updates #19493.

Change-Id: Ia036357011752f45bb9b8282a4ab5e31bc8d1a69
Reviewed-on: https://go-review.googlesource.com/66010
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Tom Bergan <tombergan@google.com>
7 years agoruntime: make runtime.GC() trigger GC even if GOGC=off
Austin Clements [Mon, 25 Sep 2017 18:58:13 +0000 (14:58 -0400)]
runtime: make runtime.GC() trigger GC even if GOGC=off

Currently, the priority of checks in (gcTrigger).test() puts the
gcpercent<0 test above gcTriggerCycle, which is used for runtime.GC().
This is an unintentional change from 1.8 and before, where
runtime.GC() triggered a GC even if GOGC=off.

Fix this by rearranging the priority so the gcTriggerCycle test
executes even if gcpercent < 0.

Fixes #22023.

Change-Id: I109328d7b643b6824eb9d79061a9e775f0149575
Reviewed-on: https://go-review.googlesource.com/65994
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agointernal/poll: don't return from Close until descriptor is closed
Ian Lance Taylor [Tue, 26 Sep 2017 03:49:37 +0000 (20:49 -0700)]
internal/poll: don't return from Close until descriptor is closed

This permits the program to reliably know that when the Close method
returns, the descriptor has definitely been closed. This matters at
least for listeners.

Fixes #21856
Updates #7970

Change-Id: I1fd0cfd2333649e6e67c6ae956e19fdff3a35a83
Reviewed-on: https://go-review.googlesource.com/66150
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Joe Tsai <joetsai@google.com>
7 years agoruntime: improve comments for nextSample
Giovanni Bajo [Sun, 24 Sep 2017 16:13:26 +0000 (18:13 +0200)]
runtime: improve comments for nextSample

The previous comment of nextSample didn't mention Poisson processes,
which is the reason why it needed to create an exponential
distribution, so it was hard to follow the reasoning for people
not highly familiar with statistics.

Since we're at it, we also make it clear that we are just creating
a random number with exponential distribution by moving the
bulk of the function into a new fastexprand().

No functional changes.

Change-Id: I9c275e87edb3418ee0974257af64c73465028ad7
Reviewed-on: https://go-review.googlesource.com/65657
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: fix sign-extension merging rules
Keith Randall [Thu, 21 Sep 2017 19:52:38 +0000 (12:52 -0700)]
cmd/compile: fix sign-extension merging rules

If we have

  y = <int16> (MOVBQSX x)
  z = <int32> (MOVWQSX y)

We used to use this rewrite rule:

(MOVWQSX x:(MOVBQSX _)) -> x

But that resulted in replacing z with a value whose type
is only int16.  Then if z is spilled and restored, it gets
zero extended instead of sign extended.

Instead use the rule

(MOVWQSX (MOVBQSX x)) -> (MOVBQSX x)

The result is has the correct type, so it can be spilled
and restored correctly.  It might mean that a few more extension
ops might not be eliminated, but that's the price for correctness.

Fixes #21963

Change-Id: I6ec82c3d2dbe43cc1fee6fb2bd6b3a72fca3af00
Reviewed-on: https://go-review.googlesource.com/65290
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agosyscall: allow abstract unix socket to use the full Path len
Nicolas BRULEZ [Tue, 26 Sep 2017 08:35:04 +0000 (10:35 +0200)]
syscall: allow abstract unix socket to use the full Path len

The previous implementation forced all Unix socket to have a name
strictly shorter than len(sa.raw.Path) to allow a terminating NULL
byte to be added. This requirement does not apply to abstract socket
names under Linux, so for this case we allow the full length.

Fixes #21965

Change-Id: I1d1f58b6b6172d589428c7230cfeae984de78b4b
Reviewed-on: https://go-review.googlesource.com/66190
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/link: don't use internal linking mode for cgo on PPC64
Ian Lance Taylor [Tue, 26 Sep 2017 13:55:49 +0000 (06:55 -0700)]
cmd/link: don't use internal linking mode for cgo on PPC64

The internal linker doesn't know how to handle multiple TOC sections
in internal linking mode. This used to work because before CL 64793 we
invoked ld -r on multiple objects, and that merged the TOC sections
for us.

Updates #21961

Change-Id: I48260a7195be660016f2f358ebc8cb79652210ab
Reviewed-on: https://go-review.googlesource.com/66270
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: fix regression in PPC64.rules move zero
Lynn Boger [Tue, 19 Sep 2017 21:36:57 +0000 (17:36 -0400)]
cmd/compile: fix regression in PPC64.rules move zero

When a MOVDstorezero (8 bytes) is used the offset field
in the instruction must be a multiple of 4. This situation
had been corrected in the rules for other types of stores
but not for the zero case.

This also removes some of the special MOVDstorezero cases since
they can be handled by the general LowerZero case.

Updates made to the ssa test for lowering zero moves to
include cases where the target is not aligned to at least 4.

Fixes #21947

Change-Id: I7cceceb1be4898c77cd3b5e78b58dce0a7e28edd
Reviewed-on: https://go-review.googlesource.com/64970
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Carlos Eduardo Seo <cseo@linux.vnet.ibm.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agointernal/poll: be explicit when using runtime netpoller
Alex Brainman [Mon, 25 Sep 2017 08:54:14 +0000 (18:54 +1000)]
internal/poll: be explicit when using runtime netpoller

internal/poll package assumes that only net sockets use runtime
netpoller on windows. We get memory corruption if other file
handles are passed into runtime poller. Make FD.Init receive
and use useNetpoller argument, so FD.Init caller is explicit
about using runtime netpoller.

Fixes #21172

Change-Id: I60e2bfedf9dda9b341eb7a3e5221035db29f5739
Reviewed-on: https://go-review.googlesource.com/65810
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: fix print/println when input is uint
Anfernee Yongkun Gui [Fri, 15 Sep 2017 06:06:20 +0000 (23:06 -0700)]
cmd/compile: fix print/println when input is uint

Fixes #21887

Change-Id: I30e8e03ecfb67a2c4deedc2c8436da4c4782136d
Reviewed-on: https://go-review.googlesource.com/63971
Run-TryBot: Emmanuel Odeke <emm.odeke@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
7 years agoarchive/tar: avoid empty IO operations
Joe Tsai [Mon, 25 Sep 2017 22:03:49 +0000 (15:03 -0700)]
archive/tar: avoid empty IO operations

The interfaces for io.Reader and io.Writer permit calling Read/Write
with an empty buffer. However, this condition is often not well tested
and can lead to bugs in various implementations of io.Reader and io.Writer.
For example, see #22028 for buggy io.Reader in the bzip2 package.

We reduce the likelihood of hitting these bugs by adjusting
regFileReader.Read and regFileWriter.Write to avoid performing
Read and Write calls when the buffer is known to be empty.

Fixes #22029

Change-Id: Ie4a26be53cf87bc4d2abd951fa005db5871cc75c
Reviewed-on: https://go-review.googlesource.com/66111
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
Reviewed-by: Giovanni Bajo <rasky@develer.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocompress/bzip2: fix checksum mismatch on empty reads
Joe Tsai [Mon, 25 Sep 2017 21:23:38 +0000 (14:23 -0700)]
compress/bzip2: fix checksum mismatch on empty reads

Previously, the read method checked whether the current block
was fully consumed or not based on whether the buffer could be filled
with a non-zero number of bytes. This check is problematic because
zero bytes could be read if the provided buffer is empty.

We fix this case by simply checking for whether the input buffer
provided by the user was empty or not. If empty, we assume that
we could not read any bytes because the buffer was too small,
rather than indicating that the current block was fully exhausted.

This check causes bzip2.Reader to be unable to make progress
on the next block unless a non-empty buffer is provided.
However, that is an entirely reasonable expectation since a
non-empty buffer needs to be provided eventually anyways to
read the actual contents of subsequent blocks.

Fixes #22028

Change-Id: I2bb1b2d54e78567baf2bf7b490a272c0853d7bfe
Reviewed-on: https://go-review.googlesource.com/66110
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/compile: add runtime GC funcs to inlining test
Daniel Martí [Sat, 23 Sep 2017 20:44:18 +0000 (21:44 +0100)]
cmd/compile: add runtime GC funcs to inlining test

This is based on a list that Austin Clements provided in mid-2016. It is
mostly untouched, except for the fact that the wbufptr funcs were
removed from the runtime thus removed from the lits here too.

Add a section for these GC funcs, since there are quite a lot of them
and the runtime has tons of funcs that we want to inline. As before,
sort this section too.

Also place some of these funcs out of the GC section, as they are not
directly related to the GC.

Updates #21851.

Change-Id: I35eb777a4c50b5f655618920dc2bc568c7c30ff5
Reviewed-on: https://go-review.googlesource.com/65654
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
7 years agoregexp: make (*bitState).push inlinable
Ilya Tocar [Fri, 22 Sep 2017 20:15:23 +0000 (15:15 -0500)]
regexp: make (*bitState).push inlinable

By refactoring job.arg from int with 0/1 as the only valid values into bool
and simplifying (*bitState).push, we reduce the number of nodes below the inlining threshold.
This improves backtracking regexp performance by 5-10% and go1 geomean  by 1.7%
Full performance data below:

name                                      old time/op    new time/op     delta
Find-6                                       510ns ± 0%      480ns ± 1%   -5.90%  (p=0.000 n=10+10)
FindString-6                                 504ns ± 1%      479ns ± 1%   -5.10%  (p=0.000 n=10+10)
FindSubmatch-6                               689ns ± 1%      659ns ± 1%   -4.27%  (p=0.000 n=9+10)
FindStringSubmatch-6                         659ns ± 0%      628ns ± 1%   -4.69%  (p=0.000 n=8+10)
Literal-6                                    174ns ± 1%      171ns ± 1%   -1.50%  (p=0.000 n=10+10)
NotLiteral-6                                2.89µs ± 1%     2.72µs ± 0%   -5.84%  (p=0.000 n=10+9)
MatchClass-6                                4.65µs ± 1%     4.28µs ± 1%   -7.96%  (p=0.000 n=10+10)
MatchClass_InRange-6                        4.15µs ± 1%     3.80µs ± 0%   -8.61%  (p=0.000 n=10+8)
ReplaceAll-6                                2.72µs ± 1%     2.60µs ± 1%   -4.68%  (p=0.000 n=10+10)
AnchoredLiteralShortNonMatch-6               158ns ± 1%      153ns ± 1%   -3.03%  (p=0.000 n=10+10)
AnchoredLiteralLongNonMatch-6                176ns ± 1%      176ns ± 0%     ~     (p=1.000 n=10+9)
AnchoredShortMatch-6                         260ns ± 0%      255ns ± 1%   -1.84%  (p=0.000 n=9+10)
AnchoredLongMatch-6                          456ns ± 0%      455ns ± 0%   -0.19%  (p=0.008 n=8+10)
OnePassShortA-6                             1.13µs ± 1%     1.12µs ± 0%   -0.57%  (p=0.046 n=10+8)
NotOnePassShortA-6                          1.14µs ± 1%     1.14µs ± 1%     ~     (p=0.162 n=10+10)
OnePassShortB-6                              908ns ± 0%      893ns ± 0%   -1.60%  (p=0.000 n=8+9)
NotOnePassShortB-6                           857ns ± 0%      803ns ± 1%   -6.34%  (p=0.000 n=8+10)
OnePassLongPrefix-6                          190ns ± 0%      190ns ± 1%     ~     (p=0.059 n=8+10)
OnePassLongNotPrefix-6                       722ns ± 1%      722ns ± 1%     ~     (p=0.451 n=10+10)
MatchParallelShared-6                        810ns ± 2%      807ns ± 2%     ~     (p=0.643 n=10+10)
MatchParallelCopied-6                       72.1ns ± 1%     69.4ns ± 1%   -3.81%  (p=0.000 n=10+10)
QuoteMetaAll-6                               213ns ± 2%      216ns ± 3%     ~     (p=0.284 n=10+10)
QuoteMetaNone-6                             89.7ns ± 1%     89.8ns ± 1%     ~     (p=0.616 n=10+10)
Match/Easy0/32-6                             127ns ± 1%      127ns ± 1%     ~     (p=0.977 n=10+10)
Match/Easy0/1K-6                             566ns ± 0%      566ns ± 0%     ~     (p=1.000 n=8+8)
Match/Easy0/32K-6                           9.30µs ± 1%     9.28µs ± 1%     ~     (p=0.529 n=10+10)
Match/Easy0/1M-6                             460µs ± 1%      460µs ± 1%     ~     (p=0.853 n=10+10)
Match/Easy0/32M-6                           15.0ms ± 0%     15.1ms ± 0%   +0.77%  (p=0.000 n=9+8)
Match/Easy0i/32-6                           2.10µs ± 1%     1.98µs ± 0%   -6.02%  (p=0.000 n=10+8)
Match/Easy0i/1K-6                           61.5µs ± 0%     57.2µs ± 0%   -6.97%  (p=0.000 n=10+9)
Match/Easy0i/32K-6                          2.75ms ± 0%     2.72ms ± 0%   -1.10%  (p=0.000 n=9+9)
Match/Easy0i/1M-6                           88.0ms ± 0%     86.9ms ± 1%   -1.29%  (p=0.000 n=8+10)
Match/Easy0i/32M-6                           2.82s ± 0%      2.77s ± 1%   -1.81%  (p=0.000 n=8+10)
Match/Easy1/32-6                             123ns ± 1%      124ns ± 1%   +0.90%  (p=0.001 n=10+10)
Match/Easy1/1K-6                            1.70µs ± 1%     1.65µs ± 0%   -3.18%  (p=0.000 n=9+10)
Match/Easy1/32K-6                           69.1µs ± 0%     68.4µs ± 1%   -0.95%  (p=0.000 n=8+10)
Match/Easy1/1M-6                            2.46ms ± 1%     2.42ms ± 1%   -1.66%  (p=0.000 n=10+10)
Match/Easy1/32M-6                           78.4ms ± 1%     77.5ms ± 0%   -1.08%  (p=0.000 n=10+9)
Match/Medium/32-6                           2.07µs ± 1%     1.91µs ± 1%   -7.69%  (p=0.000 n=10+10)
Match/Medium/1K-6                           62.8µs ± 0%     58.0µs ± 1%   -7.70%  (p=0.000 n=8+10)
Match/Medium/32K-6                          2.63ms ± 1%     2.58ms ± 1%   -2.14%  (p=0.000 n=10+10)
Match/Medium/1M-6                           84.6ms ± 0%     82.5ms ± 0%   -2.37%  (p=0.000 n=8+9)
Match/Medium/32M-6                           2.71s ± 0%      2.64s ± 0%   -2.46%  (p=0.000 n=10+9)
Match/Hard/32-6                             3.26µs ± 1%     2.98µs ± 1%   -8.49%  (p=0.000 n=10+10)
Match/Hard/1K-6                              100µs ± 0%       90µs ± 1%   -9.55%  (p=0.000 n=9+10)
Match/Hard/32K-6                            3.82ms ± 0%     3.82ms ± 1%     ~     (p=0.515 n=8+10)
Match/Hard/1M-6                              122ms ± 1%      123ms ± 0%   +0.66%  (p=0.000 n=10+8)
Match/Hard/32M-6                             3.89s ± 1%      3.91s ± 1%     ~     (p=0.105 n=10+10)
Match/Hard1/32-6                            18.1µs ± 1%     16.1µs ± 1%  -11.31%  (p=0.000 n=10+10)
Match/Hard1/1K-6                             565µs ± 0%      493µs ± 1%  -12.65%  (p=0.000 n=8+10)
Match/Hard1/32K-6                           18.8ms ± 0%     18.8ms ± 1%     ~     (p=0.905 n=9+10)
Match/Hard1/1M-6                             602ms ± 1%      602ms ± 1%     ~     (p=0.278 n=9+10)
Match/Hard1/32M-6                            19.1s ± 1%      19.2s ± 1%   +0.31%  (p=0.035 n=9+10)
Match_onepass_regex/32-6                    6.32µs ± 1%     6.34µs ± 1%     ~     (p=0.060 n=10+10)
Match_onepass_regex/1K-6                     204µs ± 1%      204µs ± 1%     ~     (p=0.842 n=9+10)
Match_onepass_regex/32K-6                   6.53ms ± 0%     6.55ms ± 1%   +0.36%  (p=0.005 n=10+10)
Match_onepass_regex/1M-6                     209ms ± 0%      208ms ± 1%   -0.65%  (p=0.034 n=8+10)
Match_onepass_regex/32M-6                    6.72s ± 0%      6.68s ± 1%   -0.74%  (p=0.000 n=9+10)
CompileOnepass/^(?:(?:(?:.(?:$))?))...-6    7.02µs ± 1%     7.02µs ± 1%     ~     (p=0.671 n=10+10)
CompileOnepass/^abcd$-6                     5.65µs ± 1%     5.65µs ± 1%     ~     (p=0.411 n=10+9)
CompileOnepass/^(?:(?:a{0,})*?)$-6          7.06µs ± 1%     7.06µs ± 1%     ~     (p=0.912 n=10+10)
CompileOnepass/^(?:(?:a+)*)$-6              6.40µs ± 1%     6.41µs ± 1%     ~     (p=0.699 n=10+10)
CompileOnepass/^(?:(?:a|(?:aa)))$-6         8.18µs ± 2%     8.16µs ± 1%     ~     (p=0.529 n=10+10)
CompileOnepass/^(?:[^\s\S])$-6              5.08µs ± 1%     5.17µs ± 1%   +1.77%  (p=0.000 n=9+10)
CompileOnepass/^(?:(?:(?:a*)+))$-6          6.86µs ± 1%     6.85µs ± 0%     ~     (p=0.190 n=10+9)
CompileOnepass/^[a-c]+$-6                   5.14µs ± 1%     5.11µs ± 0%   -0.53%  (p=0.041 n=10+10)
CompileOnepass/^[a-c]*$-6                   5.62µs ± 1%     5.63µs ± 1%     ~     (p=0.382 n=10+10)
CompileOnepass/^(?:a*)$-6                   5.76µs ± 1%     5.73µs ± 1%   -0.41%  (p=0.008 n=9+10)
CompileOnepass/^(?:(?:aa)|a)$-6             7.89µs ± 1%     7.84µs ± 1%   -0.66%  (p=0.020 n=10+10)
CompileOnepass/^...$-6                      5.38µs ± 1%     5.38µs ± 1%     ~     (p=0.857 n=9+10)
CompileOnepass/^(?:a|(?:aa))$-6             7.80µs ± 2%     7.82µs ± 1%     ~     (p=0.342 n=10+10)
CompileOnepass/^a((b))c$-6                  7.75µs ± 1%     7.78µs ± 1%     ~     (p=0.172 n=10+10)
CompileOnepass/^a.[l-nA-Cg-j]?e$-6          8.39µs ± 1%     8.42µs ± 1%     ~     (p=0.138 n=10+10)
CompileOnepass/^a((b))$-6                   6.92µs ± 1%     6.95µs ± 1%     ~     (p=0.159 n=10+10)
CompileOnepass/^a(?:(b)|(c))c$-6            10.0µs ± 1%     10.0µs ± 1%     ~     (p=0.896 n=10+10)
CompileOnepass/^a(?:b|c)$-6                 5.62µs ± 1%     5.66µs ± 1%   +0.71%  (p=0.023 n=10+10)
CompileOnepass/^a(?:b?|c)$-6                8.49µs ± 1%     8.43µs ± 1%   -0.69%  (p=0.010 n=10+10)
CompileOnepass/^a(?:b?|c+)$-6               9.26µs ± 1%     9.28µs ± 1%     ~     (p=0.448 n=10+10)
CompileOnepass/^a(?:bc)+$-6                 6.52µs ± 1%     6.46µs ± 2%   -1.02%  (p=0.003 n=10+10)
CompileOnepass/^a(?:[bcd])+$-6              6.29µs ± 1%     6.32µs ± 1%     ~     (p=0.256 n=10+10)
CompileOnepass/^a((?:[bcd])+)$-6            7.77µs ± 1%     7.79µs ± 1%     ~     (p=0.105 n=10+10)
CompileOnepass/^a(:?b|c)*d$-6               14.0µs ± 1%     13.9µs ± 1%   -0.69%  (p=0.003 n=10+10)
CompileOnepass/^.bc(d|e)*$-6                8.96µs ± 1%     9.06µs ± 1%   +1.20%  (p=0.000 n=10+9)
CompileOnepass/^loooooooooooooooooo...-6     219µs ± 1%      220µs ± 1%   +0.63%  (p=0.006 n=9+10)
[Geo mean]                                  31.6µs          31.1µs        -1.82%

name                                      old speed      new speed       delta
QuoteMetaAll-6                            65.5MB/s ± 2%   64.8MB/s ± 3%     ~     (p=0.315 n=10+10)
QuoteMetaNone-6                            290MB/s ± 1%    290MB/s ± 1%     ~     (p=0.755 n=10+10)
Match/Easy0/32-6                           250MB/s ± 0%    251MB/s ± 1%     ~     (p=0.277 n=8+9)
Match/Easy0/1K-6                          1.81GB/s ± 0%   1.81GB/s ± 0%     ~     (p=0.408 n=8+10)
Match/Easy0/32K-6                         3.52GB/s ± 1%   3.53GB/s ± 1%     ~     (p=0.529 n=10+10)
Match/Easy0/1M-6                          2.28GB/s ± 1%   2.28GB/s ± 1%     ~     (p=0.853 n=10+10)
Match/Easy0/32M-6                         2.24GB/s ± 0%   2.23GB/s ± 0%   -0.76%  (p=0.000 n=9+8)
Match/Easy0i/32-6                         15.2MB/s ± 1%   16.2MB/s ± 0%   +6.43%  (p=0.000 n=10+9)
Match/Easy0i/1K-6                         16.6MB/s ± 0%   17.9MB/s ± 0%   +7.48%  (p=0.000 n=10+9)
Match/Easy0i/32K-6                        11.9MB/s ± 0%   12.0MB/s ± 0%   +1.11%  (p=0.000 n=9+9)
Match/Easy0i/1M-6                         11.9MB/s ± 0%   12.1MB/s ± 1%   +1.31%  (p=0.000 n=8+10)
Match/Easy0i/32M-6                        11.9MB/s ± 0%   12.1MB/s ± 1%   +1.84%  (p=0.000 n=8+10)
Match/Easy1/32-6                           260MB/s ± 1%    258MB/s ± 1%   -0.91%  (p=0.001 n=10+10)
Match/Easy1/1K-6                           601MB/s ± 1%    621MB/s ± 0%   +3.28%  (p=0.000 n=9+10)
Match/Easy1/32K-6                          474MB/s ± 0%    479MB/s ± 1%   +0.96%  (p=0.000 n=8+10)
Match/Easy1/1M-6                           426MB/s ± 1%    433MB/s ± 1%   +1.68%  (p=0.000 n=10+10)
Match/Easy1/32M-6                          428MB/s ± 1%    433MB/s ± 0%   +1.09%  (p=0.000 n=10+9)
Match/Medium/32-6                         15.4MB/s ± 1%   16.7MB/s ± 1%   +8.23%  (p=0.000 n=10+9)
Match/Medium/1K-6                         16.3MB/s ± 1%   17.7MB/s ± 1%   +8.43%  (p=0.000 n=9+10)
Match/Medium/32K-6                        12.5MB/s ± 1%   12.7MB/s ± 1%   +2.15%  (p=0.000 n=10+10)
Match/Medium/1M-6                         12.4MB/s ± 0%   12.7MB/s ± 0%   +2.44%  (p=0.000 n=8+9)
Match/Medium/32M-6                        12.4MB/s ± 0%   12.7MB/s ± 0%   +2.52%  (p=0.000 n=10+9)
Match/Hard/32-6                           9.82MB/s ± 1%  10.73MB/s ± 1%   +9.29%  (p=0.000 n=10+10)
Match/Hard/1K-6                           10.2MB/s ± 0%   11.3MB/s ± 1%  +10.56%  (p=0.000 n=9+10)
Match/Hard/32K-6                          8.58MB/s ± 0%   8.58MB/s ± 1%     ~     (p=0.554 n=8+10)
Match/Hard/1M-6                           8.59MB/s ± 1%   8.53MB/s ± 0%   -0.70%  (p=0.000 n=10+8)
Match/Hard/32M-6                          8.62MB/s ± 1%   8.59MB/s ± 1%     ~     (p=0.098 n=10+10)
Match/Hard1/32-6                          1.77MB/s ± 1%   1.99MB/s ± 1%  +12.40%  (p=0.000 n=10+8)
Match/Hard1/1K-6                          1.81MB/s ± 1%   2.08MB/s ± 1%  +14.55%  (p=0.000 n=10+10)
Match/Hard1/32K-6                         1.74MB/s ± 0%   1.74MB/s ± 0%     ~     (p=0.108 n=9+10)
Match/Hard1/1M-6                          1.74MB/s ± 0%   1.74MB/s ± 1%     ~     (p=1.000 n=9+10)
Match/Hard1/32M-6                         1.75MB/s ± 0%   1.75MB/s ± 1%     ~     (p=0.157 n=9+10)
Match_onepass_regex/32-6                  5.05MB/s ± 0%   5.05MB/s ± 1%     ~     (p=0.262 n=8+10)
Match_onepass_regex/1K-6                  5.02MB/s ± 1%   5.02MB/s ± 1%     ~     (p=0.677 n=9+10)
Match_onepass_regex/32K-6                 5.02MB/s ± 0%   4.99MB/s ± 0%   -0.47%  (p=0.000 n=10+9)
Match_onepass_regex/1M-6                  5.01MB/s ± 0%   5.04MB/s ± 1%   +0.68%  (p=0.017 n=8+10)
Match_onepass_regex/32M-6                 4.99MB/s ± 0%   5.03MB/s ± 1%   +0.74%  (p=0.000 n=10+10)
[Geo mean]                                29.1MB/s        29.8MB/s        +2.44%

go1 data for reference

name                     old time/op    new time/op    delta
BinaryTree17-6              4.39s ± 1%     4.37s ± 0%   -0.58%  (p=0.006 n=9+9)
Fannkuch11-6                5.13s ± 0%     5.18s ± 0%   +0.87%  (p=0.000 n=8+8)
FmtFprintfEmpty-6          74.2ns ± 0%    71.7ns ± 3%   -3.41%  (p=0.000 n=10+10)
FmtFprintfString-6          120ns ± 1%     122ns ± 2%     ~     (p=0.333 n=10+10)
FmtFprintfInt-6             127ns ± 1%     127ns ± 1%     ~     (p=0.809 n=10+10)
FmtFprintfIntInt-6          186ns ± 0%     188ns ± 1%   +1.02%  (p=0.002 n=8+10)
FmtFprintfPrefixedInt-6     223ns ± 1%     222ns ± 2%     ~     (p=0.421 n=10+10)
FmtFprintfFloat-6           374ns ± 0%     376ns ± 1%   +0.43%  (p=0.030 n=8+10)
FmtManyArgs-6               795ns ± 0%     788ns ± 1%   -0.79%  (p=0.000 n=8+9)
GobDecode-6                10.9ms ± 1%    10.9ms ± 0%     ~     (p=0.079 n=10+9)
GobEncode-6                8.60ms ± 1%    8.56ms ± 0%   -0.52%  (p=0.004 n=10+10)
Gzip-6                      378ms ± 1%     386ms ± 1%   +2.28%  (p=0.000 n=10+10)
Gunzip-6                   63.7ms ± 0%    62.3ms ± 0%   -2.22%  (p=0.000 n=9+8)
HTTPClientServer-6          120µs ± 3%     114µs ± 3%   -4.99%  (p=0.000 n=10+10)
JSONEncode-6               20.3ms ± 1%    19.9ms ± 0%   -1.90%  (p=0.000 n=9+10)
JSONDecode-6               84.3ms ± 0%    83.7ms ± 0%   -0.76%  (p=0.000 n=8+8)
Mandelbrot200-6            6.91ms ± 0%    6.89ms ± 0%   -0.31%  (p=0.000 n=9+8)
GoParse-6                  5.49ms ± 0%    5.47ms ± 1%     ~     (p=0.101 n=8+10)
RegexpMatchEasy0_32-6       130ns ± 0%     128ns ± 0%   -1.54%  (p=0.002 n=8+10)
RegexpMatchEasy0_1K-6       322ns ± 1%     322ns ± 0%     ~     (p=0.525 n=10+9)
RegexpMatchEasy1_32-6       124ns ± 0%     124ns ± 0%   -0.32%  (p=0.046 n=8+10)
RegexpMatchEasy1_1K-6       570ns ± 0%     548ns ± 1%   -3.76%  (p=0.000 n=10+10)
RegexpMatchMedium_32-6      196ns ± 0%     183ns ± 1%   -6.61%  (p=0.000 n=8+10)
RegexpMatchMedium_1K-6     64.3µs ± 0%    59.0µs ± 1%   -8.31%  (p=0.000 n=8+10)
RegexpMatchHard_32-6       3.08µs ± 0%    2.80µs ± 0%   -8.96%  (p=0.000 n=8+9)
RegexpMatchHard_1K-6       93.0µs ± 0%    84.5µs ± 1%   -9.17%  (p=0.000 n=8+9)
Revcomp-6                   647ms ± 2%     646ms ± 1%     ~     (p=0.720 n=10+9)
Template-6                 92.3ms ± 0%    91.7ms ± 0%   -0.65%  (p=0.000 n=8+8)
TimeParse-6                 490ns ± 0%     488ns ± 0%   -0.43%  (p=0.000 n=10+10)
TimeFormat-6                513ns ± 0%     513ns ± 1%     ~     (p=0.144 n=9+10)
[Geo mean]                 79.1µs         77.7µs        -1.73%

name                     old speed      new speed      delta
GobDecode-6              70.1MB/s ± 1%  70.3MB/s ± 0%     ~     (p=0.078 n=10+9)
GobEncode-6              89.2MB/s ± 1%  89.7MB/s ± 0%   +0.52%  (p=0.004 n=10+10)
Gzip-6                   51.4MB/s ± 1%  50.2MB/s ± 1%   -2.23%  (p=0.000 n=10+10)
Gunzip-6                  304MB/s ± 0%   311MB/s ± 0%   +2.27%  (p=0.000 n=9+8)
JSONEncode-6             95.8MB/s ± 1%  97.7MB/s ± 0%   +1.93%  (p=0.000 n=9+10)
JSONDecode-6             23.0MB/s ± 0%  23.2MB/s ± 0%   +0.76%  (p=0.000 n=8+8)
GoParse-6                10.6MB/s ± 0%  10.6MB/s ± 1%     ~     (p=0.111 n=8+10)
RegexpMatchEasy0_32-6     244MB/s ± 0%   249MB/s ± 0%   +2.06%  (p=0.000 n=9+10)
RegexpMatchEasy0_1K-6    3.18GB/s ± 1%  3.17GB/s ± 0%     ~     (p=0.211 n=10+9)
RegexpMatchEasy1_32-6     257MB/s ± 0%   258MB/s ± 0%   +0.37%  (p=0.000 n=8+8)
RegexpMatchEasy1_1K-6    1.80GB/s ± 0%  1.87GB/s ± 1%   +3.91%  (p=0.000 n=10+10)
RegexpMatchMedium_32-6   5.08MB/s ± 0%  5.43MB/s ± 1%   +7.03%  (p=0.000 n=8+10)
RegexpMatchMedium_1K-6   15.9MB/s ± 0%  17.4MB/s ± 1%   +9.08%  (p=0.000 n=8+10)
RegexpMatchHard_32-6     10.4MB/s ± 0%  11.4MB/s ± 0%   +9.82%  (p=0.000 n=8+9)
RegexpMatchHard_1K-6     11.0MB/s ± 0%  12.1MB/s ± 1%  +10.10%  (p=0.000 n=8+9)
Revcomp-6                 393MB/s ± 2%   394MB/s ± 1%     ~     (p=0.720 n=10+9)
Template-6               21.0MB/s ± 0%  21.2MB/s ± 0%   +0.66%  (p=0.000 n=8+8)
[Geo mean]               74.2MB/s       76.2MB/s        +2.70%

Updates #21851

Change-Id: Ie88455db925f422a828f8528293790726a9c036b
Reviewed-on: https://go-review.googlesource.com/65491
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: improve static map initialization
Keith Randall [Mon, 25 Sep 2017 17:28:20 +0000 (10:28 -0700)]
cmd/compile: improve static map initialization

When static maps are large, we try to initialize them
by iterating over an array of key/value pairs.

Currently this optimization only works if the keys and values
are of primitive type.  This CL improves this optimization
by allowing any static composite literals as well.

Fixes #22010

Change-Id: Ie493e02ab8b8a228a3472b5c6025a33f7b92daf3
Reviewed-on: https://go-review.googlesource.com/66050
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agostrings: improve readability of IndexAny and LastIndexAny functions.
Gabriel Aszalos [Mon, 25 Sep 2017 12:54:37 +0000 (14:54 +0200)]
strings: improve readability of IndexAny and LastIndexAny functions.

This change removes the check of len(chars) > 0 inside the Index and
IndexAny functions which was redundant.

Change-Id: Iffbc0f2b3332c6e31c7514b5f644b6fe7bdcfe0d
Reviewed-on: https://go-review.googlesource.com/65910
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
7 years agopath/filepath: ensure Walk report unreadable directories once
Wei Congrui [Fri, 15 Sep 2017 07:36:16 +0000 (15:36 +0800)]
path/filepath: ensure Walk report unreadable directories once

Before this change, if Walk encounters an unreadable directory,
it will call walkFn with this directory twice. Argument err in
the first call is nil, and the second is the permission error.

This change removes the former call and makes Walk call walkFn
with permission error.

Fixes #21758

Change-Id: I21e57c67f3c5a8370fc80a43db3c8009fbce6439
Reviewed-on: https://go-review.googlesource.com/63994
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agobytes: improve readability of IndexAny and LastIndexAny functions
Gabriel Aszalos [Mon, 25 Sep 2017 11:01:36 +0000 (13:01 +0200)]
bytes: improve readability of IndexAny and LastIndexAny functions

This change removes the check of len(chars) > 0 inside the Index and
IndexAny functions which was redundant.

Change-Id: Ic4bf8b8a37d7f040d3ebd81b4fc45fcb386b639a
Reviewed-on: https://go-review.googlesource.com/65851
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoall: prefer strings.IndexByte over strings.Index
Marvin Stenger [Thu, 21 Sep 2017 17:01:27 +0000 (19:01 +0200)]
all: prefer strings.IndexByte over strings.Index

strings.IndexByte was introduced in go1.2 and it can be used
effectively wherever the second argument to strings.Index is
exactly one byte long.

This avoids generating unnecessary string symbols and saves
a few calls to strings.Index.

Change-Id: I1ab5edb7c4ee9058084cfa57cbcc267c2597e793
Reviewed-on: https://go-review.googlesource.com/65930
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoruntime: fix TestGdbConst on windows
Alessandro Arzilli [Mon, 25 Sep 2017 10:30:41 +0000 (12:30 +0200)]
runtime: fix TestGdbConst on windows

Some (all?) versions of gdb on windows output "\r\n" as line ending
instead of "\n".

Fixes #22012

Change-Id: I798204fd9f616d6d2c9c28eb5227fadfc63c0d45
Reviewed-on: https://go-review.googlesource.com/65850
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agogo/types: don't accept incorrect shift expression arguments
griesemer [Fri, 22 Sep 2017 14:24:25 +0000 (16:24 +0200)]
go/types: don't accept incorrect shift expression arguments

Under certain circumstances involving shifts, go/types didn't verify
that untyped constant values were representable by the relevant type,
leading to the acceptance of incorrect programs (see the issue).

Fixing this code exposed another problem with int-to-string conversions
which suddenly failed because now the type-checker complained that a
(constant) integer argument wasn't representable as a string. Fixed that
as well.

Added many additional tests covering the various scenarious.

Found two cmd/compile bugs in the process (#21979, #21981) and filed
a go/types TODO (#21982).

Fixes #21727.

Change-Id: If443ee0230979cd7d45d2fc669e623648caa70da
Reviewed-on: https://go-review.googlesource.com/65370
Reviewed-by: Alan Donovan <adonovan@google.com>
7 years agocmd/compile: merge bytes inline test with the rest
Daniel Martí [Sun, 24 Sep 2017 16:55:19 +0000 (17:55 +0100)]
cmd/compile: merge bytes inline test with the rest

In golang.org/cl/42813, a test was added in the bytes package to check
if a Buffer method was being inlined, using 'go tool nm'.

Now that we have a compiler test that verifies that certain funcs are
inlineable, merge it there. Knowing whether the funcs are inlineable is
also more reliable than whether or not their symbol appears in the
binary, too. For example, under some circumstances, inlineable funcs
can't be inlined, such as if closures are used.

While at it, add a few more bytes.Buffer methods that are currently
inlined and should clearly stay that way.

Updates #21851.

Change-Id: I62066e32ef5542d37908bd64f90bda51276da4de
Reviewed-on: https://go-review.googlesource.com/65658
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: refactor some more gotos away
Daniel Martí [Sat, 23 Sep 2017 18:39:08 +0000 (19:39 +0100)]
cmd/compile: refactor some more gotos away

The ones in racewalk.go are almost all useless, since they were just
breaks.

typecheck.go wasn't trivial, but still doable with an if/else chain.

Also remove a single silly goto in const.go, while at it.

Change-Id: I776a78df6bb3b6bd4f7e5feec546c772baf4e02e
Reviewed-on: https://go-review.googlesource.com/65652
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agonet/mail: skip trailing comment while parsing email
Minaev Mike [Mon, 7 Aug 2017 08:22:21 +0000 (08:22 +0000)]
net/mail: skip trailing comment while parsing email

The existing implementation doesn't handle
comment constructions in email address.
So addresses that are consistent with RFC 5322
don't parse at all.

Fixes #21257

Change-Id: Iae3ba951dfb26b7cf0e1885a680bbceb9123d6d5
Reviewed-on: https://go-review.googlesource.com/53550
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agosort: update main example to use Slice along with Sort
Agniva De Sarker [Sat, 23 Sep 2017 16:59:42 +0000 (22:29 +0530)]
sort: update main example to use Slice along with Sort

This is to let sort.Slice get more prominence since
it's the most common use case.

Fixes #21989

Change-Id: I0b180cc20256f5f09065b722e191c508c872f4f7
Reviewed-on: https://go-review.googlesource.com/65710
Reviewed-by: Rob Pike <r@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>

7 years agocmd/go: move GOOS/GOARCH and tags checks to Init
Daniel Martí [Sun, 24 Sep 2017 10:16:52 +0000 (11:16 +0100)]
cmd/go: move GOOS/GOARCH and tags checks to Init

They were in Do, which is the method that actually starts the build.
However, we can already run these checks in Init, since we already have
all the information necessary to do the checks.

More importantly, some work happens between Init and Do, namely package
loading. That may exit with an error, meaning that in some cases the
user gets a confusing error instead of the correct one.

For example, using a GOOS typo, before showed:

$ GOOS=windwos go build
can't load package: package p: build constraints exclude all Go files in ...

And now:

$ GOOS=windwos go build
cmd/go: unsupported GOOS/GOARCH pair windwos/amd64

Also had to tweak TestGoEnv to modify GOOS as well as GOARCH. Otherwise,
on windows this would result in the invalid GOOS/GOARCH pair
windows/arm, which would error given that we now check that in non-build
commands such as "go env".

Fixes #21999.

Change-Id: Iff9890dea472bff0179a9d703d6f698a0e3187c1
Reviewed-on: https://go-review.googlesource.com/65656
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
7 years agoos: don't refer directly to Sysfd in epipecheck
Ian Lance Taylor [Sun, 24 Sep 2017 05:00:53 +0000 (22:00 -0700)]
os: don't refer directly to Sysfd in epipecheck

Instead record in the File whether it is stdout/stderr. This avoids a
race between a call to epipecheck and closing the file.

Fixes #21994

Change-Id: Ic3d552ffa83402136276bcb5029ec3e6691042c2
Reviewed-on: https://go-review.googlesource.com/65750
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
7 years agodatabase/sql: update minor sql docs
Daniel Theophanes [Sun, 24 Sep 2017 00:04:51 +0000 (17:04 -0700)]
database/sql: update minor sql docs

Replace the work "session" with "connection" in docs. Fix
The ErrConnDone documentation. Clarify what the context is used
for in StmtContext.

Change-Id: I2f07e58d0cd6321b386a73b038cf6070cb8e2572
Reviewed-on: https://go-review.googlesource.com/65732
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/compile: fix invalid switch case value panic
Daniel Martí [Sat, 23 Sep 2017 22:04:31 +0000 (23:04 +0100)]
cmd/compile: fix invalid switch case value panic

This is a regression introduced by myself in golang.org/cl/41852,
confirmed by the program that reproduces the crash that can be seen in
the added test.

Fixes #21988.

Change-Id: I18d5b2b3de63ced84db705b18490b00b16b59e02
Reviewed-on: https://go-review.googlesource.com/65655
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agopath/filepath: simplify TestEvalSymlinks
Alex Brainman [Thu, 17 Aug 2017 10:05:15 +0000 (20:05 +1000)]
path/filepath: simplify TestEvalSymlinks

Change-Id: I4a747fca0db3cbd4972feaddcb625041b648620b
Reviewed-on: https://go-review.googlesource.com/56710
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/go: ignore empty path elements in GOPATH
Alex Brainman [Thu, 21 Sep 2017 02:33:11 +0000 (12:33 +1000)]
cmd/go: ignore empty path elements in GOPATH

go command refuses to use GOPATH with empty path elements
(like %GOPATH%=C:\go;). But environment variable change dialog
on Windows 10 produces strings ending with ; (see issue #21928
for a picture). Just accept GOPATH with empty path elements,
and ignore all empty path elements.

Fixes #21928

Change-Id: I1d3c3a19274ed69204d29ae06c3e8ff8c57c1ca0
Reviewed-on: https://go-review.googlesource.com/65151
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agodatabase/sql: add OpenDB to directly create a *DB without a DSN.
James Lawrence [Sat, 5 Aug 2017 10:04:19 +0000 (06:04 -0400)]
database/sql: add OpenDB to directly create a *DB without a DSN.

The current Open method limits the ability for driver maintainers
to expose options for their drivers by forcing all the configuration
to pass through the DSN in order to create a *DB.

This CL allows driver maintainers to write their own initialization
functions that return a *DB making configuration of the underlying
drivers easier.

Fixes #20268

Change-Id: Ib10b794f36a201bbb92c23999c8351815d38eedb
Reviewed-on: https://go-review.googlesource.com/53430
Reviewed-by: Daniel Theophanes <kardianos@gmail.com>
Run-TryBot: Daniel Theophanes <kardianos@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: clarify adjustctxt inlining comment
Daniel Martí [Sat, 23 Sep 2017 11:22:10 +0000 (12:22 +0100)]
cmd/compile: clarify adjustctxt inlining comment

The reason why adjustctxt wasn't being inlined was reported as:

function too complex: cost 92 exceeds budget 80

However, after tweaking the code to be under the budget limit, we see
the real blocker:

non-leaf function

There is little we can do about this one in particular at the moment.
Create a section with funcs that will need mid-stack inlining to be
inlineable, since this will likely come up again in other cases.

Change-Id: I3a8eb1546b289a060ac896506a007b0496946e84
Reviewed-on: https://go-review.googlesource.com/65650
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agocmd/compile/internal/gc: unexport global constants
Hajime Hoshi [Sun, 28 May 2017 14:38:59 +0000 (23:38 +0900)]
cmd/compile/internal/gc: unexport global constants

Change-Id: Ib292ef3b0a31b2c7bdd77519324362667f30389c
Reviewed-on: https://go-review.googlesource.com/44393
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoarchive/tar: fix sparse files support on Darwin
Giovanni Bajo [Fri, 22 Sep 2017 22:57:51 +0000 (00:57 +0200)]
archive/tar: fix sparse files support on Darwin

Apple defined the SEEK_HOLE/SEEK_DATA constants in unistd.h
with swapped values, compared to all other UNIX systems.

Fixes #21970

Change-Id: I84a33e0741f0f33a2e04898e96b788b87aa9890f
Reviewed-on: https://go-review.googlesource.com/65570
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/vet: remove extraneous "//" in dead.go so its comment does not
Shawn Smith [Sat, 23 Sep 2017 08:43:16 +0000 (17:43 +0900)]
cmd/vet: remove extraneous "//" in dead.go so its comment does not

appear at the top of vet's godoc output

Change-Id: I2593d294b0497aeb9e8c54a4dad052b5c33ecaee
Reviewed-on: https://go-review.googlesource.com/65630
Reviewed-by: Rob Pike <r@golang.org>
7 years agoencoding/json: cleanup detection of unexported embedded fields
Joe Tsai [Fri, 22 Sep 2017 22:44:09 +0000 (15:44 -0700)]
encoding/json: cleanup detection of unexported embedded fields

CL 60410 fixes the compiler such that reflect.StructField.PkgPath
is non-empty if and only if the field is unexported.
Given that property, we can cleanup the logic in the json encoder
to avoid parsing the field name to detect export properties.

Updates #21122

Change-Id: Ic01b9c4ca76386774846b742b0c1b9b948f53e7c
Reviewed-on: https://go-review.googlesource.com/65550
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agoruntime: hide <autogenerated> methods from call stack
Austin Clements [Mon, 12 Jun 2017 15:12:12 +0000 (11:12 -0400)]
runtime: hide <autogenerated> methods from call stack

The compiler generates wrapper methods to forward interface method
calls (which are always pointer-based) to value methods. These
wrappers appear in the call stack even though they are an
implementation detail. This leaves ugly "<autogenerated>" functions in
stack traces and can throw off skip counts for stack traces.

Fix this by considering these runtime frames in printed stack traces
so they will only be printed if runtime frames are being printed, and
by eliding them from the call stack expansion used by CallersFrames
and Caller.

This removes the test for issue 4388 since that was checking that
"<autogenerated>" appeared in the stack trace instead of something
even weirder. We replace it with various runtime package tests.

Fixes #16723.

Change-Id: Ice3f118c66f254bb71478a664d62ab3fc7125819
Reviewed-on: https://go-review.googlesource.com/45412
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoruntime: simplify stack walk in panicwrap
Austin Clements [Mon, 12 Jun 2017 18:24:16 +0000 (14:24 -0400)]
runtime: simplify stack walk in panicwrap

panicwrap currently uses runtime.Callers and runtime.CallersFrames to
find the name of its caller. Simplify this by using getcallerpc.

This will be important for #16723, since to fix that we're going to
make CallersFrames skip the wrapper method, which is exactly what
panicwrap needs to see.

Change-Id: Icb0776d399966e31595f3ee44f980290827e32a6
Reviewed-on: https://go-review.googlesource.com/45411
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoruntime: remove getcallerpc argument
Austin Clements [Fri, 22 Sep 2017 19:16:26 +0000 (15:16 -0400)]
runtime: remove getcallerpc argument

Now that getcallerpc is a compiler intrinsic on x86 and non-x86
platforms don't need the argument, we can drop it.

Sadly, this doesn't let us remove any dummy arguments since all of
those cases also use getcallersp, which still takes the argument
pointer, but this is at least an improvement.

Change-Id: I9c34a41cf2c18cba57f59938390bf9491efb22d2
Reviewed-on: https://go-review.googlesource.com/65474
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
7 years agoos/exec: remove protection against simultaneous Wait/Write
Ian Lance Taylor [Fri, 22 Sep 2017 19:03:52 +0000 (12:03 -0700)]
os/exec: remove protection against simultaneous Wait/Write

CL 31148 added code to protect again simultaneous calls to Close and
Wait when using the standard input pipe, to fix the race condition
described in issue #9307. That issue is a special case of the race
between Close and Write described by issue #7970. Since issue #7970
was not fixed, CL 31148 fixed the problem specific to os/exec.

Since then, issue #7970 has been fixed, so the specific fix in os/exec
is no longer necessary. Remove it, effectively reverting CL 31148 and
followup CL 33298.

Updates #7970
Updates #9307
Updates #17647

Change-Id: Ic0b62569cb0aba44b32153cf5f9632bd1f1b411a
Reviewed-on: https://go-review.googlesource.com/65490
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Miguel Bernabeu <miguelbernadi@gmail.com>
Reviewed-by: Russ Cox <rsc@golang.org>
Reviewed-by: Joe Tsai <joetsai@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoruntime: re-add sanity check for GCCPUFraction
Austin Clements [Mon, 10 Jul 2017 18:34:26 +0000 (14:34 -0400)]
runtime: re-add sanity check for GCCPUFraction

This no longer appears to be reproducible on windows/386. Try putting
it back and we'll see if the builders still don't like it.

Fixes #19319.

Change-Id: Ia47b034e18d0a5a1951125c00542b021aacd5e8d
Reviewed-on: https://go-review.googlesource.com/47936
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/compile: add intrinsic for reading caller's pc
David Chase [Mon, 24 Oct 2016 14:25:05 +0000 (10:25 -0400)]
cmd/compile: add intrinsic for reading caller's pc

First step towards removing the mandatory argument for
getcallerpc, which solves certain problems for the runtime.
This might also slightly improve performance.

Intrinsic enabled on 386, amd64, amd64p32,
runtime asm implementation removed on those architectures.

Now-superfluous argument remains in getcallerpc signature
(for a future CL; non-386/amd64 asm funcs ignore it).

Added getcallerpc to the "not a real function" test
in dcl.go, that story is a little odd with respect to
unexported functions but that is not this CL.

Fixes #17327.

Change-Id: I5df1ad91f27ee9ac1f0dd88fa48f1329d6306c3e
Reviewed-on: https://go-review.googlesource.com/31851
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
7 years agoruntime: don't call lockOSThread for every cgo call
Ian Lance Taylor [Wed, 13 Sep 2017 22:53:47 +0000 (15:53 -0700)]
runtime: don't call lockOSThread for every cgo call

For a trivial benchmark with a do-nothing cgo call:

name    old time/op  new time/op  delta
Call-4  64.5ns ± 7%  63.0ns ± 6%  -2.25%  (p=0.027 n=20+16)

Because Windows uses the cgocall mechanism to make system calls,
and passes arguments in a struct held in the m,
we need to do the lockOSThread/unlockOSThread in that code.

Because deferreturn was getting a nosplit stack overflow error,
change it to avoid calling typedmemmove.

Updates #21827.

Change-Id: I9b1d61434c44faeb29805b46b409c812c9acadc2
Reviewed-on: https://go-review.googlesource.com/64070
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
7 years agocmd/compile,cmd/link: export int global consts to DWARF
Alessandro Arzilli [Sun, 3 Sep 2017 09:59:18 +0000 (11:59 +0200)]
cmd/compile,cmd/link: export int global consts to DWARF

Updates #14517

Change-Id: I23ef88e71c89da12dffcadf5562ea2d7557b62cf
Reviewed-on: https://go-review.googlesource.com/61019
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: add more runtime funcs to inline test
Daniel Martí [Fri, 22 Sep 2017 11:21:36 +0000 (12:21 +0100)]
cmd/compile: add more runtime funcs to inline test

This is based from a list that Keith Randall provided in mid-2016. These
are all funcs that, at the time, were important and small enough that
they should be clearly inlined.

The runtime has changed a bit since then. Ctz16 and Ctz8 were removed,
so don't add them. stringtoslicebytetmp was moved to the backend, so
it's no longer a Go function. And itabhash was moved to itabHashFunc.

The only other outlier is adjustctxt, which is not inlineable at the
moment. I've added a TODO and will address it myself in a separate
commit.

While at it, error if any funcs in the input table are duplicated.
They're never useful and typos could lead to unintentionally thinking a
function is inlineable when it actually isn't.

And, since the lists are getting long, start sorting alphabetically.

Finally, rotl_31 is only defined on 64-bit architectures, and the added
runtime/internal/sys funcs are assembly on 386 and thus non-inlineable
in that case.

Updates #21851.

Change-Id: Ib99ab53d777860270e8fd4aefc41adb448f13662
Reviewed-on: https://go-review.googlesource.com/65351
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/compile: remove unused cases from switch
Daniel Martí [Fri, 22 Sep 2017 16:49:12 +0000 (17:49 +0100)]
cmd/compile: remove unused cases from switch

The first just falls through, and the default case does nothing. They
can be deleted.

Change-Id: I82ab1ce3acde0b8423334cfbf35f9e0c806cd494
Reviewed-on: https://go-review.googlesource.com/65410
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: remove a few unnecessary gotos
Daniel Martí [Tue, 12 Sep 2017 20:51:17 +0000 (22:51 +0200)]
cmd/compile: remove a few unnecessary gotos

Rework the logic to remove them. These were the low hanging fruit,
with labels that were used only once and logic that was fairly
straightforward.

Change-Id: I02a01c59c247b8b2972d8d73ff23f96f271de038
Reviewed-on: https://go-review.googlesource.com/63410
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agocmd/compile/internal/types: unexport Type.Copy
Matthew Dempsky [Thu, 21 Sep 2017 21:26:21 +0000 (14:26 -0700)]
cmd/compile/internal/types: unexport Type.Copy

It's only used/needed by SubstAny.

CL prepared with gorename.

Change-Id: I243138f9dcc4e6af9b81a7746414e6d7b3ba10a2
Reviewed-on: https://go-review.googlesource.com/65311
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>

7 years agoarchive/tar: skip TestSparseFiles on Plan 9
David du Colombier [Fri, 22 Sep 2017 13:50:03 +0000 (15:50 +0200)]
archive/tar: skip TestSparseFiles on Plan 9

CL 60871 added TestSparseFiles. This test is succeeding
on Plan 9 when executed on the ramfs file system, but
is failing when executed on the Fossil file system.

This may be due to an issue in the handling of sparse
files in the Fossil file system on Plan 9 that should
be investigated.

Updates #21977.

Change-Id: I177afff519b862a5c548e094203c219504852006
Reviewed-on: https://go-review.googlesource.com/65352
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoruntime: fix typo in the word "preceding"
Tobias Klauser [Fri, 22 Sep 2017 08:42:12 +0000 (10:42 +0200)]
runtime: fix typo in the word "preceding"

Change-Id: I6d8c8ca0dee972cabfcc95fda23aea25692633a5
Reviewed-on: https://go-review.googlesource.com/65350
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
7 years agoall: use sort.Slice in a few more places
Daniel Martí [Tue, 12 Sep 2017 15:09:46 +0000 (17:09 +0200)]
all: use sort.Slice in a few more places

Do the low-hanging fruit - tiny Less functions that are used exactly
once. This reduces the amount of code and puts the logic in a single
place.

Change-Id: I9d4544cd68de5a95e55019bdad1fca0a1dbfae9c
Reviewed-on: https://go-review.googlesource.com/63171
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agosyscall: update syscall.Select to use newselect on ppc64x
Carlos Eduardo Seo [Wed, 20 Sep 2017 20:33:26 +0000 (17:33 -0300)]
syscall: update syscall.Select to use newselect on ppc64x

Analog to the runtime package, syscall.Select should be using
newselect instead of select. This change addresses this problem and
regenerates zsyscall_linux_* for ppc64 and ppc64le.

Updates #21946

Change-Id: I5dc3bf9e7f0b1172d6cce30ddf3bb1e3c95ec8e9
Reviewed-on: https://go-review.googlesource.com/65090
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
7 years agocrypto/elliptic: gofmt
Michael Munday [Mon, 18 Sep 2017 10:50:29 +0000 (11:50 +0100)]
crypto/elliptic: gofmt

Change-Id: Ia4b49736d3b33cddf58905c6b19febbca45b2ad2
Reviewed-on: https://go-review.googlesource.com/64270
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: collect reasons in inlining test
Daniel Martí [Thu, 14 Sep 2017 14:51:18 +0000 (15:51 +0100)]
cmd/compile: collect reasons in inlining test

If we use -gcflags='-m -m', the compiler should give us a reason why a
func couldn't be inlined. Add the extra -m necessary for that extra info
and use it to give better test failures. For example, for the func in
the TODO:

--- FAIL: TestIntendedInlining (1.53s)
inl_test.go:104: runtime.nextFreeFast was not inlined: function too complex

We might increase the number of -m flags to get more information at some
later point, such as getting details on how close the func was to the
inlining budget.

Also started using regexes, as the output parsing is getting a bit too
complex for manual string handling.

While at it, also refactored the test to not buffer the entire output
into memory. This is fine in practice, but it won't scale well as we add
more packages or we depend more on the compiler's debugging output.

For example, "go build -a -gcflags='-m -m' std" prints nearly 40MB of
plaintext - and we only need to see the output line by line anyway.

Updates #21851.

Change-Id: I00986ff360eb56e4e9737b65a6be749ef8540643
Reviewed-on: https://go-review.googlesource.com/63810
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agomath/big: delete solved TODO
Marvin Stenger [Thu, 21 Sep 2017 15:35:32 +0000 (17:35 +0200)]
math/big: delete solved TODO

The TODO is no longer needed as it was solved by a previous CL.
See https://go-review.googlesource.com/14995.

Change-Id: If62d1b296f35758ad3d18d28c8fbb95e797f4464
Reviewed-on: https://go-review.googlesource.com/65232
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agoarchive/tar: perform test for hole-detection on specific builders
Joe Tsai [Thu, 21 Sep 2017 15:44:05 +0000 (08:44 -0700)]
archive/tar: perform test for hole-detection on specific builders

The test for hole-detection is heavily dependent on whether the
OS and underlying FS provides support for it.
Even on Linux, which has support for SEEK_HOLE and SEEK_DATA,
the underlying filesystem may not have support for it.
In order to avoid an ever-changing game of whack-a-mole,
we whitelist the specific builders that we expect the test to pass on.

Updates #21964

Change-Id: I7334e8532c96cc346ea83aabbb81b719685ad7e5
Reviewed-on: https://go-review.googlesource.com/65270
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/nm: accept macho files which don't have symbol table in the archive
Hiroshi Ioka [Thu, 21 Sep 2017 03:56:30 +0000 (12:56 +0900)]
cmd/nm: accept macho files which don't have symbol table in the archive

After https://golang.org/cl/64793, we started to include Mach-O object
files which don't have symbol table into cgo archive.
However, toolchains didn't handle those files yet.

Fixes #21959

Change-Id: Ibb2f6492f1fa59368f2dfd4cff19783997539875
Reviewed-on: https://go-review.googlesource.com/65170
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agobytes: add documentation to reader methods
Gabriel Aszalos [Thu, 21 Sep 2017 10:25:02 +0000 (12:25 +0200)]
bytes: add documentation to reader methods

Some methods that were used to implement various `io` interfaces in the
Reader were documented, whereas others were not. This change adds
documentation to all the missing methods used to implement these
interfaces.

Change-Id: I2dac6e328542de3cd87e89510651cd6ba74a7b7d
Reviewed-on: https://go-review.googlesource.com/65231
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoarchive/tar: make check for hole detection support more liberal
Joe Tsai [Thu, 21 Sep 2017 06:20:28 +0000 (23:20 -0700)]
archive/tar: make check for hole detection support more liberal

On most Unix OSes, lseek reports EINVAL when lacking SEEK_HOLE support.
However, there are reports that ENOTTY is reported instead.
Rather than tracking down every possible errno that may be used to
represent "not supported", just treat any non-nil error as meaning
that there is no support. This is the same strategy taken by the
GNU and BSD tar tools.

Fixes #21958

Change-Id: Iae68afdc934042f52fa914fca45f0ca89220c383
Reviewed-on: https://go-review.googlesource.com/65191
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/compile: optimized ARM code with BFX/BFXU
Ben Shi [Wed, 20 Sep 2017 08:48:34 +0000 (08:48 +0000)]
cmd/compile: optimized ARM code with BFX/BFXU

BFX&BFXU were introduced in ARMv6T2. A single BFX or BFXU is
more efficiently than a pair of left-shift/right-shift in bit
field extraction.

This patch implements this optimization. And the benchmark tests
show big improvement in special cases and little change in total.

1. There is big improvement in a special test case.
name                     old time/op    new time/op    delta
BFX-4                       665µs ± 1%     595µs ± 0%  -10.61%  (p=0.000 n=20+20)
(The test case: https://github.com/benshi001/ugo1/blob/master/bfx_test.go)

2. The compilecmp benchmark shows no regression.
name        old time/op       new time/op       delta
Template          2.33s ± 2%        2.34s ± 2%    ~     (p=0.356 n=9+10)
Unicode           1.32s ± 2%        1.30s ± 2%    ~     (p=0.139 n=9+8)
GoTypes           7.77s ± 1%        7.76s ± 1%    ~     (p=0.780 n=10+9)
Compiler          37.3s ± 1%        37.1s ± 1%    ~     (p=0.211 n=10+9)
SSA               84.3s ± 2%        84.3s ± 2%    ~     (p=0.842 n=10+9)
Flate             1.45s ± 1%        1.45s ± 3%    ~     (p=0.853 n=10+10)
GoParser          1.83s ± 2%        1.83s ± 2%    ~     (p=0.739 n=10+10)
Reflect           5.08s ± 2%        5.09s ± 2%    ~     (p=0.720 n=9+10)
Tar               2.44s ± 1%        2.44s ± 2%    ~     (p=0.684 n=10+10)
XML               2.62s ± 2%        2.62s ± 2%    ~     (p=0.529 n=10+10)
[Geo mean]        4.80s             4.79s       -0.06%

name        old user-time/op  new user-time/op  delta
Template          2.76s ± 2%        2.75s ± 3%    ~     (p=0.893 n=10+10)
Unicode           1.63s ± 1%        1.60s ± 1%  -2.07%  (p=0.000 n=8+9)
GoTypes           9.54s ± 1%        9.52s ± 1%    ~     (p=0.215 n=10+10)
Compiler          46.0s ± 1%        46.0s ± 1%    ~     (p=0.853 n=10+10)
SSA                110s ± 1%         110s ± 1%    ~     (p=0.838 n=10+10)
Flate             1.69s ± 3%        1.69s ± 5%    ~     (p=0.957 n=10+10)
GoParser          2.15s ± 2%        2.15s ± 2%    ~     (p=0.749 n=10+10)
Reflect           6.03s ± 1%        5.99s ± 2%    ~     (p=0.060 n=9+10)
Tar               3.02s ± 2%        2.99s ± 2%    ~     (p=0.214 n=10+10)
XML               3.10s ± 2%        3.08s ± 2%    ~     (p=0.732 n=9+10)
[Geo mean]        5.82s             5.79s       -0.41%

name        old text-bytes    new text-bytes    delta
HelloSize         589kB ± 0%        589kB ± 0%    ~     (all equal)

name        old data-bytes    new data-bytes    delta
HelloSize        5.46kB ± 0%       5.46kB ± 0%    ~     (all equal)

name        old bss-bytes     new bss-bytes     delta
HelloSize        76.9kB ± 0%       76.9kB ± 0%    ~     (all equal)

name        old exe-bytes     new exe-bytes     delta
HelloSize        1.03MB ± 0%       1.03MB ± 0%    ~     (all equal)

3. The go1 benchmark shows little change in total. (excluding noise)
name                     old time/op    new time/op    delta
BinaryTree17-4              41.5s ± 1%     41.6s ± 1%    ~     (p=0.373 n=30+26)
Fannkuch11-4                23.6s ± 1%     23.6s ± 1%  +0.28%  (p=0.003 n=29+30)
FmtFprintfEmpty-4           826ns ± 1%     827ns ± 1%    ~     (p=0.155 n=30+30)
FmtFprintfString-4         1.35µs ± 1%    1.35µs ± 1%    ~     (p=0.499 n=30+30)
FmtFprintfInt-4            1.43µs ± 1%    1.41µs ± 1%  -1.19%  (p=0.000 n=30+30)
FmtFprintfIntInt-4         2.15µs ± 1%    2.11µs ± 1%  -1.78%  (p=0.000 n=30+30)
FmtFprintfPrefixedInt-4    2.21µs ± 1%    2.21µs ± 1%    ~     (p=0.881 n=30+30)
FmtFprintfFloat-4          4.41µs ± 1%    4.44µs ± 0%  +0.64%  (p=0.000 n=30+30)
FmtManyArgs-4              8.06µs ± 1%    8.06µs ± 0%    ~     (p=0.871 n=30+30)
GobDecode-4                 103ms ± 1%     104ms ± 2%  +0.54%  (p=0.013 n=28+29)
GobEncode-4                92.4ms ± 1%    92.6ms ± 1%    ~     (p=0.447 n=30+29)
Gzip-4                      4.17s ± 1%     4.06s ± 1%  -2.56%  (p=0.000 n=29+30)
Gunzip-4                    603ms ± 1%     602ms ± 1%    ~     (p=0.423 n=30+30)
HTTPClientServer-4          688µs ± 2%     674µs ± 3%  -2.09%  (p=0.000 n=29+30)
JSONEncode-4                237ms ± 1%     237ms ± 1%    ~     (p=0.061 n=29+30)
JSONDecode-4                907ms ± 1%     910ms ± 1%    ~     (p=0.061 n=30+30)
Mandelbrot200-4            41.7ms ± 0%    41.7ms ± 0%  +0.19%  (p=0.000 n=24+20)
GoParse-4                  45.7ms ± 2%    45.5ms ± 2%  -0.29%  (p=0.005 n=30+30)
RegexpMatchEasy0_32-4      1.27µs ± 0%    1.27µs ± 0%  +0.12%  (p=0.031 n=30+30)
RegexpMatchEasy0_1K-4      7.77µs ± 4%    7.73µs ± 3%    ~     (p=0.169 n=30+30)
RegexpMatchEasy1_32-4      1.29µs ± 1%    1.29µs ± 1%    ~     (p=0.126 n=30+30)
RegexpMatchEasy1_1K-4      10.4µs ± 3%    10.3µs ± 2%  -1.32%  (p=0.004 n=30+29)
RegexpMatchMedium_32-4     2.06µs ± 0%    2.06µs ± 0%    ~     (p=0.071 n=30+30)
RegexpMatchMedium_1K-4      531µs ± 1%     530µs ± 0%    ~     (p=0.121 n=30+23)
RegexpMatchHard_32-4       28.7µs ± 1%    28.6µs ± 1%  -0.21%  (p=0.001 n=30+27)
RegexpMatchHard_1K-4        860µs ± 1%     857µs ± 1%    ~     (p=0.105 n=30+27)
Revcomp-4                  67.3ms ± 2%    67.3ms ± 2%    ~     (p=0.805 n=29+29)
Template-4                  1.08s ± 1%     1.08s ± 1%    ~     (p=0.260 n=30+30)
TimeParse-4                7.04µs ± 0%    7.04µs ± 0%    ~     (p=0.315 n=30+30)
TimeFormat-4               13.2µs ± 1%    13.2µs ± 1%    ~     (p=0.077 n=30+30)
[Geo mean]                  715µs          713µs       -0.30%

name                     old speed      new speed      delta
GobDecode-4              7.42MB/s ± 1%  7.38MB/s ± 2%  -0.54%  (p=0.011 n=28+29)
GobEncode-4              8.30MB/s ± 1%  8.29MB/s ± 1%    ~     (p=0.484 n=30+29)
Gzip-4                   4.65MB/s ± 2%  4.78MB/s ± 1%  +2.73%  (p=0.000 n=30+30)
Gunzip-4                 32.2MB/s ± 1%  32.2MB/s ± 1%    ~     (p=0.357 n=30+30)
JSONEncode-4             8.18MB/s ± 1%  8.19MB/s ± 1%    ~     (p=0.052 n=29+30)
JSONDecode-4             2.14MB/s ± 1%  2.13MB/s ± 1%    ~     (p=0.074 n=30+29)
GoParse-4                1.27MB/s ± 1%  1.27MB/s ± 2%    ~     (p=0.618 n=24+30)
RegexpMatchEasy0_32-4    25.2MB/s ± 0%  25.2MB/s ± 0%  -0.12%  (p=0.031 n=30+30)
RegexpMatchEasy0_1K-4     132MB/s ± 5%   132MB/s ± 2%    ~     (p=0.171 n=30+30)
RegexpMatchEasy1_32-4    24.8MB/s ± 1%  24.9MB/s ± 1%    ~     (p=0.106 n=30+30)
RegexpMatchEasy1_1K-4    98.4MB/s ± 3%  99.6MB/s ± 4%  +1.19%  (p=0.011 n=30+30)
RegexpMatchMedium_32-4    483kB/s ± 1%   484kB/s ± 1%    ~     (p=0.426 n=30+30)
RegexpMatchMedium_1K-4   1.93MB/s ± 1%  1.93MB/s ± 0%    ~     (p=0.157 n=30+17)
RegexpMatchHard_32-4     1.12MB/s ± 1%  1.12MB/s ± 0%  +0.33%  (p=0.001 n=30+24)
RegexpMatchHard_1K-4     1.19MB/s ± 1%  1.19MB/s ± 1%    ~     (p=0.290 n=30+30)
Revcomp-4                37.8MB/s ± 2%  37.8MB/s ± 1%    ~     (p=0.815 n=29+29)
Template-4               1.80MB/s ± 1%  1.80MB/s ± 1%    ~     (p=0.586 n=30+30)
[Geo mean]               6.80MB/s       6.81MB/s       +0.25%

fixes #20966

Change-Id: Idb5567bbe988c875315b8c98c128957cd474ccc5
Reviewed-on: https://go-review.googlesource.com/64950
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>

7 years agocontext: fix references to "d" in WithDeadline docs
Michael Darakananda [Thu, 21 Sep 2017 00:25:35 +0000 (10:25 +1000)]
context: fix references to "d" in WithDeadline docs

Docs of WithDeadline refers to variable "d" which does not exist
in the docs.

This commit renames the time argument to "d" to make the doc work.

Change-Id: Ifd2c1be7d2e3f7dfb21cd9bb8ff7fc5039c8d3bd
Reviewed-on: https://go-review.googlesource.com/65130
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/nm: handle cgo archive
Hiroshi Ioka [Sat, 16 Sep 2017 06:28:14 +0000 (15:28 +0900)]
cmd/nm: handle cgo archive

This CL also make cmd/nm accept PE object file.

Fixes #21706

Change-Id: I4a528b7d53da1082e61523ebeba02c4c514a43a7
Reviewed-on: https://go-review.googlesource.com/64890
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/link: Grouping declaration of variables on ld/pe.go
Avelino [Fri, 1 Sep 2017 15:43:04 +0000 (12:43 -0300)]
cmd/link: Grouping declaration of variables on ld/pe.go

Change-Id: I33284d3154db43b2b89418c5076df79407e7cf41
Reviewed-on: https://go-review.googlesource.com/60931
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/compile: use a counter to track whether writebarrier rewriting is done
Cherry Zhang [Wed, 30 Aug 2017 15:12:52 +0000 (11:12 -0400)]
cmd/compile: use a counter to track whether writebarrier rewriting is done

Use a counter, instead of a loop, to see whether there are more
writebarrier ops in the current block that need to be rewritten.

No visible change in normal compiler speed benchmarks.

Passes toolstash -cmp on std cmd.

Fixes #20416.

Change-Id: Ifbbde23611cd668c35b8a4a3e9a92726bfe19956
Reviewed-on: https://go-review.googlesource.com/60310
Reviewed-by: David Chase <drchase@google.com>
7 years agocmd/compile/internal/gc: update comment in plive.go
Matthew Dempsky [Wed, 20 Sep 2017 22:22:21 +0000 (15:22 -0700)]
cmd/compile/internal/gc: update comment in plive.go

onebitwalktype1 no longer appears to be a bottleneck for the mentioned
test case. In fact, we appear to compile it significantly faster now
than Go 1.4 did (~1.8s vs ~3s).

Fixes #21951.

Change-Id: I315313e906092a7d6ff4ff60a918d80a4cff7a7f
Reviewed-on: https://go-review.googlesource.com/65110
Reviewed-by: Keith Randall <khr@golang.org>
7 years agoarchive/tar: add Header.DetectSparseHoles and Header.PunchSparseHoles
Joe Tsai [Fri, 1 Sep 2017 09:57:46 +0000 (02:57 -0700)]
archive/tar: add Header.DetectSparseHoles and Header.PunchSparseHoles

To support the detection and creation of sparse files,
add two new methods:
func Header.DetectSparseHoles(*os.File) error
func Header.PunchSparseHoles(*os.File) error

DetectSparseHoles is intended to be used after FileInfoHeader
prior to serializing the Header with WriteHeader.
For each OS, it uses specialized logic to detect
the location of sparse holes. On most Unix systems, it uses
SEEK_HOLE and SEEK_DATA to query for the holes.
On Windows, it uses a specialized the FSCTL_QUERY_ALLOCATED_RANGES
syscall to query for all the holes.

PunchSparseHoles is intended to be used after Reader.Next
prior to populating the file with Reader.WriteTo.
On Windows, this uses the FSCTL_SET_ZERO_DATA syscall.
On other operating systems it simply truncates the file
to the end-offset of SparseHoles.

DetectSparseHoles and PunchSparseHoles are added as methods on
Header because they are heavily tied to the operating system,
for which there is already an existing precedence for
(since FileInfoHeader makes uses of OS-specific details).

Fixes #13548

Change-Id: I98a321dd1ce0165f3d143d4edadfda5e7db67746
Reviewed-on: https://go-review.googlesource.com/60871
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agomath: implement fast path for Exp
Agniva De Sarker [Sat, 9 Sep 2017 14:17:37 +0000 (19:47 +0530)]
math: implement fast path for Exp

- using FMA and AVX instructions if available to speed-up
Exp calculation on amd64

- using a data table instead of #define'ed constants because
these instructions do not support loading floating point immediates.
One has to use a memory operand / register.

- Benchmark results on Intel(R) Core(TM) i5-5200U CPU @ 2.20GHz:

Original vs New (non-FMA path)
name  old time/op    new time/op    delta
Exp     16.0ns ± 1%    16.1ns ± 3%   ~     (p=0.308 n=9+10)

Original vs New (FMA path)
name  old time/op    new time/op    delta
Exp     16.0ns ± 1%    13.7ns ± 2%  -14.80%  (p=0.000 n=9+10)

Change-Id: I3d8986925d82b39b95ee979ae06f59d7e591d02e
Reviewed-on: https://go-review.googlesource.com/62590
Reviewed-by: Ilya Tocar <ilya.tocar@intel.com>
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile/internal/gc: better inliner diagnostics
Ilya Tocar [Tue, 12 Sep 2017 18:53:55 +0000 (13:53 -0500)]
cmd/compile/internal/gc: better inliner diagnostics

When debugging inliner with -m -m print cost of complex functions,
instead of simple "function too complex". This helps to understand,
how close to inlining is this particular function.

Change-Id: I6871f69b5b914d23fd0b43a24d7c6fc928f4b716
Reviewed-on: https://go-review.googlesource.com/63330
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/go: stop linking cgo objects together with ld -r
Ian Lance Taylor [Tue, 19 Sep 2017 20:53:20 +0000 (13:53 -0700)]
cmd/go: stop linking cgo objects together with ld -r

https://golang.org/cl/5822049 introduced the idea of linking together
all the cgo objects with -r, while also linking against -lgcc. This
was to fix http://golang.org/issue/3261: cgo code that requires libgcc
would break when using internal linking.

This approach introduced https://golang.org/issue/9510: multiple
different cgo packages could include the same libgcc object, leading
to a multiple definition error during the final link. That problem was
fixed by https://golang.org/cl/16741, as modified by
https://golang.org/cl/16993, which did the link against libgcc only
during the final link.

After https://golang.org/cl/16741, and, on Windows, the later
https://golang.org/cl/26670, ld -r no longer does anything useful.

So, remove it.

Doing this revealed that running ld -r on Darwin simplifies some
relocs by making them specific to a symbol rather than a section.
Correct the handling of unsigned relocations in internal linking mode
by offsetting by the symbol value. This only really comes up when
using the internal linker with C code that initializes a variable to
the address of a local constant, such as a C string (as in const char
*s = "str";). This change does not affect the normal case of external
linking, where the Add field is ignored. The test case is
misc/cgo/test/issue6612.go in internal linking mode.

The cmd/internal/goobj test can now see an external object with no
symbol table; fix it to not crash in that case.

Change-Id: I15e5b7b5a8f48136bc14bf4e1c4c473d5eb58062
Reviewed-on: https://go-review.googlesource.com/64793
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agoruntime: make nextFreeFast inlinable
Ilya Tocar [Wed, 20 Sep 2017 18:09:08 +0000 (13:09 -0500)]
runtime: make nextFreeFast inlinable

https://golang.org/cl/22598 made nextFreeFast inlinable.
But during https://golang.org/cl/63611 it was discovered, that it is no longer inlinable.
Reduce number of statements below inlining threshold to make it inlinable again.
Also update tests, to prevent regressions.
Doesn't reduce readability.

Change-Id: Ia672784dd48ed3b1ab46e390132f1094fe453de5
Reviewed-on: https://go-review.googlesource.com/65030
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
7 years agocmd/compile: fix large global variables in -linkshared mode on s390x
Michael Munday [Wed, 20 Sep 2017 16:12:33 +0000 (12:12 -0400)]
cmd/compile: fix large global variables in -linkshared mode on s390x

When rewriting loads and stores accessing global variables to use the
GOT we were making use of REGTMP (R10). Unfortunately loads and stores
with large offsets (larger than 20-bits) were also using REGTMP,
causing it to be clobbered and subsequently a segmentation fault.

This can be fixed by using REGTMP2 (R11) for the rewrite. This is fine
because REGTMP2 only has a couple of uses in the assembler (division,
high multiplication and storage-to-storage instructions). We didn't
use REGTMP2 originally because it used to be used more frequently,
in particular for stores of constants to memory. However we have now
eliminated those uses.

This was found while writing a test case for CL 63030. That test case
is included in this CL.

Change-Id: I13956f1f3ca258a7c8a7ff0a7570d2848adf7f68
Reviewed-on: https://go-review.googlesource.com/65011
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: change liveness-related functions into methods
Matthew Dempsky [Wed, 20 Sep 2017 19:58:16 +0000 (12:58 -0700)]
cmd/compile: change liveness-related functions into methods

No functional change; just making the code slightly more idiomatic.

Passes toolstash-check.

Change-Id: I66d14a8410bbecf260d0ea5683564aa413ce5747
Reviewed-on: https://go-review.googlesource.com/65070
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: refactor onebitwalktype1
Matthew Dempsky [Tue, 19 Sep 2017 06:29:46 +0000 (23:29 -0700)]
cmd/compile: refactor onebitwalktype1

The existing logic tried to advance the offset for each variable's
width, but then tried to undo this logic with the array and struct
handling code. It can all be much simpler by only worrying about
computing offsets within the array and struct code.

While here, include a short-circuit for zero-width arrays to fix a
pedantic compiler failure case.

Passes toolstash-check.

Fixes #20739.

Change-Id: I98af9bb512a33e3efe82b8bf1803199edb480640
Reviewed-on: https://go-review.googlesource.com/64471
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agocmd/compile/internal/syntax: fix source buffer refilling
Matthew Dempsky [Tue, 19 Sep 2017 21:28:03 +0000 (14:28 -0700)]
cmd/compile/internal/syntax: fix source buffer refilling

The previous code seems to have an off-by-1 in it somewhere, the
consequence being that we didn't properly preserve all of the old
buffer contents that we intended to.

After spending a while looking at the existing window-shifting logic,
I wasn't able to understand exactly how it was supposed to work or
where the issue was, so I rewrote it to be (at least IMO) more
obviously correct.

Fixes #21938.

Change-Id: I1ed7bbc1e1751a52ab5f7cf0411ae289586dc345
Reviewed-on: https://go-review.googlesource.com/64830
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agocmd/compile/internal/types: simplify dclstack
Matthew Dempsky [Tue, 19 Sep 2017 19:45:17 +0000 (12:45 -0700)]
cmd/compile/internal/types: simplify dclstack

We used to backup symbol declarations using complete Syms, but this
was unnecessary: very few of Sym's fields were actually needed. Also,
to restore a symbol, we had to re-Lookup the Sym in its Pkg.

By introducing a new dedicated dsym type for this purpose, we can
address both of these deficiencies.

Passes toolstash-check.

Change-Id: I39f3d672b301f84a3a62b9b34b4b2770cb25df79
Reviewed-on: https://go-review.googlesource.com/64811
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agonet/http: net/http: doc that prefer "must" over "should"
Kunpei Sakai [Sun, 3 Sep 2017 09:18:51 +0000 (18:18 +0900)]
net/http: net/http: doc that prefer "must" over "should"

See https://go-review.googlesource.com/c/go/+/59850

Change-Id: I9f0b6bc009eae86cbbdb56562ee4eb8d5eef653e
Reviewed-on: https://go-review.googlesource.com/61230
Reviewed-by: Tom Bergan <tombergan@google.com>
7 years agocmd/compile: stop rematerializable ops from clobbering flags
Michael Munday [Mon, 11 Sep 2017 20:23:06 +0000 (21:23 +0100)]
cmd/compile: stop rematerializable ops from clobbering flags

Rematerializable ops can be inserted after the flagalloc phase,
they must therefore not clobber flags. This CL adds a check to
ensure this doesn't happen and fixes the instances where it
does currently.

amd64: ADDQconst and ADDLconst were recently changed to be
rematerializable in CL 54393 (only in tip, not 1.9). That change
has been reverted.

s390x: MOVDaddr could clobber flags when using dynamic linking due
to a ADD with immediate instruction. Change the code generation to
use LA/LAY instead.

Fixes #21080.

Change-Id: Ia85c882afa2a820a309e93775354b3169ec6d034
Reviewed-on: https://go-review.googlesource.com/63030
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Ilya Tocar <ilya.tocar@intel.com>
7 years agocmd/compile: remove {Mark,Pop}dcl calls in bimport
Matthew Dempsky [Tue, 19 Sep 2017 18:45:58 +0000 (11:45 -0700)]
cmd/compile: remove {Mark,Pop}dcl calls in bimport

These were previously only relevant for recording scoping level so
that invalid 'fallthrough' statements could be rejected. However,
that's handled differently since CL 61130 (in particular, there's no
use of types.Block anymore), so these calls can be safely removed.

Passes toolstash-check.

Change-Id: I8631b156594df85b8d39f57acad3ebcf099d52f9
Reviewed-on: https://go-review.googlesource.com/64810
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agoio: Add benchmarks for CopyN
Albert Nigmatzianov [Wed, 20 Sep 2017 14:23:11 +0000 (19:23 +0500)]
io: Add benchmarks for CopyN

Copied from CL 60630

Current results:
name          time/op
CopyNSmall-4  2.20µs ±90%
CopyNLarge-4   136µs ±56%

name          alloc/op
CopyNSmall-4  1.84kB ±21%
CopyNLarge-4   128kB ±10%

name          allocs/op
CopyNSmall-4    1.00 ± 0%
CopyNLarge-4    1.00 ± 0%

Change-Id: If08c0132a773e936c9f61bff96e0aabf58006d31
Reviewed-on: https://go-review.googlesource.com/64932
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agoio: Improve performance of CopyN
Albert Nigmatzianov [Thu, 31 Aug 2017 11:00:37 +0000 (16:00 +0500)]
io: Improve performance of CopyN

Benchmarks:
name          old time/op    new time/op    delta
CopyNSmall-4    5.09µs ± 1%    2.25µs ±86%  -55.91%  (p=0.000 n=11+14)
CopyNLarge-4     114µs ±73%     121µs ±72%     ~     (p=0.701 n=14+14)

name          old alloc/op   new alloc/op   delta
CopyNSmall-4    34.6kB ± 0%     1.9kB ±19%  -94.60%  (p=0.000 n=12+14)
CopyNLarge-4     129kB ± 8%     127kB ±18%   -2.00%  (p=0.007 n=14+14)

name          old allocs/op  new allocs/op  delta
CopyNSmall-4      2.00 ± 0%      1.00 ± 0%  -50.00%  (p=0.000 n=14+14)
CopyNLarge-4      2.00 ± 0%      1.00 ± 0%  -50.00%  (p=0.000 n=14+14)

Benchmark code:
type Buffer struct {
bytes.Buffer
io.ReaderFrom
}

func BenchmarkCopyNSmall(b *testing.B) {
bs := bytes.Repeat([]byte{0}, 1024)
rd := bytes.NewReader(bs)
buf := new(Buffer)
b.ResetTimer()

for i := 0; i < b.N; i++ {
io.CopyN(buf, rd, 512)
rd.Reset(bs)
}
}

func BenchmarkCopyNLarge(b *testing.B) {
bs := bytes.Repeat([]byte{0}, 64*1024)
rd := bytes.NewReader(bs)
buf := new(Buffer)
b.ResetTimer()

for i := 0; i < b.N; i++ {
io.CopyN(buf, rd, (32*1024)+1)
rd.Reset(bs)
}
}

Change-Id: Id8d29e55758452c870cf372db640f07baec05849
Reviewed-on: https://go-review.googlesource.com/60630
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agobytes: improve test readability
Gabriel Aszalos [Wed, 20 Sep 2017 09:10:26 +0000 (11:10 +0200)]
bytes: improve test readability

This CL improves the readability of the tests in the bytes package by
naming the `data` test variable `testString`, using the same convention
as its counterpart, `testBytes`.

It additionally removes some type casting which was unnecessary.

Change-Id: If38b5606ce8bda0306bae24498f21cb8dbbb6377
Reviewed-on: https://go-review.googlesource.com/64931
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>