]> Cypherpunks repositories - gostls13.git/log
gostls13.git
6 years agoruntime: correct ABI information for all functions
Austin Clements [Thu, 1 Nov 2018 16:25:41 +0000 (12:25 -0400)]
runtime: correct ABI information for all functions

There are three cases where we don't currently have the visibility to
get the ABIs of runtime symbols right, which this CL fixes:

1. For Go functions referenced from non-Go code in other packages.
   This is runtime.morestackc (which is referenced from function
   prologues) and a few syscall symbols. For these we need to generate
   ABI0 wrappers, so this CL adds dummy calls in the assembly code to
   force wrapper generation. There are many other cross-package
   references to runtime and runtime/internal/atomic, but these are
   handled specially by cmd/go.

2. For calls generated by the compiler to runtime Go functions, there
   are a few symbols that aren't declared in builtins.go because we've
   never needed their type information before. Now we at least need
   their ABI information, so these are added to builtins.go.

3. For calls generated by the compiler to runtime assembly functions,
   the compiler is going to assume the internal ABI is available, so
   we add Go stubs to the runtime to trigger wrapper generation. For
   these we're probably going to want to provide internal ABI
   definitions directly in the assembly for performance, but for now
   the ABIs are the same so it doesn't matter.

For #27539.

Change-Id: I9c224e7408d2ef4dd9b0e4c9d7e962ddfe111245
Reviewed-on: https://go-review.googlesource.com/c/146822
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
6 years agoruntime: avoid variable/function alias on runtime._cgo_panic_internal
Austin Clements [Thu, 1 Nov 2018 00:39:59 +0000 (20:39 -0400)]
runtime: avoid variable/function alias on runtime._cgo_panic_internal

The symbol runtime._cgo_panic_internal is defined both as a function
in package runtime and as a (linknamed) variable in package
runtime/cgo. Since we're introducing function ABIs, this is going to
cause problems with resolving the ABI-marked function symbol with the
unmarked data symbol. It's also confusing.

Fix this by declaring runtime._cgo_panic_internal as a function in
runtime/cgo as well and extracting the PC from the function object.

For #27539.

Change-Id: I148a458a600cf9e57791cf4cbe92e79bddbf58d4
Reviewed-on: https://go-review.googlesource.com/c/146821
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agointernal/bytealg, runtime: provide linknames for pushed symbols
Austin Clements [Wed, 31 Oct 2018 19:04:04 +0000 (15:04 -0400)]
internal/bytealg, runtime: provide linknames for pushed symbols

The internal/bytealg package defines several symbols in the runtime,
bytes, and strings packages in assembly, and the runtime package
defines symbols in reflect and sync/atomic. Currently, there's no
corresponding Go prototype for these symbols in the defining package.

We're going to start depending on Go prototypes in the same package as
their assembly definitions in order to provide ABI wrappers. Plus,
these are good documentation and colocate type information with
definitions, which could be useful for vet if it learned a little
about linkname.

This CL adds linknamed Go prototypes for all pushed symbols in
internal/bytealg and runtime.

For #27539.

Change-Id: I9b0c12d935a75bb6af46b6761180d451c00f11b8
Reviewed-on: https://go-review.googlesource.com/c/146820
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
6 years agoruntime, reflect: access runtime.reflectcall directly
Austin Clements [Thu, 8 Nov 2018 15:34:19 +0000 (10:34 -0500)]
runtime, reflect: access runtime.reflectcall directly

Currently, package runtime contains the definition of reflect.call,
even though it's just a jump to runtime.reflectcall. This "push"
symbol is confusing, since it's not clear where the definition of
reflect.call comes from when you're in the reflect package.

Replace this with a "pull" symbol: the runtime now defines only
runtime.reflectcall and package reflect uses a go:linkname to access
this symbol directly. This makes it clear where reflect.call is coming
from without any spooky action at a distance and eliminates all of the
definitions of reflect.call in the runtime.

Change-Id: I3ec73cd394efe9df8d3061a57c73aece2e7048dd
Reviewed-on: https://go-review.googlesource.com/c/148657
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
6 years agogo/build: remove superfluous continues
Ainar Garipov [Sun, 11 Nov 2018 09:19:20 +0000 (12:19 +0300)]
go/build: remove superfluous continues

This cleanup was proposed in CL 148937. The branch is already ended with
a continue, so remove continues from subbranches and use an else-if.

Change-Id: Iaf6eb57afc84e25862f99a342f5824e315bcdcb7
Reviewed-on: https://go-review.googlesource.com/c/148922
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agoruntime: gofmt all improperly formatted code
Michael Anthony Knyszek [Fri, 9 Nov 2018 23:33:40 +0000 (23:33 +0000)]
runtime: gofmt all improperly formatted code

This change fixes incorrect formatting in mheap.go (the result of my
previous heap scavenging changes) and map_test.go.

Change-Id: I2963687504abdc4f0cdf2f0c558174b3bc0ed2df
Reviewed-on: https://go-review.googlesource.com/c/148977
Run-TryBot: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/compile: optimize A->B->C Moves that include VarDefs
Josh Bleecher Snyder [Sun, 21 Oct 2018 02:59:27 +0000 (19:59 -0700)]
cmd/compile: optimize A->B->C Moves that include VarDefs

We have an existing optimization that recognizes
memory moves of the form A -> B -> C and converts
them into A -> C, in the hopes that the store to
B will be end up being dead and thus eliminated.

However, when A, B, and C are large types,
the front end sometimes emits VarDef ops for the moves.
This change adds an optimization to match that pattern.

This required changing an old compiler test.
The test assumed that a temporary was required
to deal with a large return value.
With this optimization in place, that temporary
ended up being eliminated.

Triggers 649 times during 'go build -a std cmd'.

Cuts 16k off cmd/go.

name        old object-bytes  new object-bytes  delta
Template          507kB ± 0%        507kB ± 0%  -0.15%  (p=0.008 n=5+5)
Unicode           225kB ± 0%        225kB ± 0%    ~     (all equal)
GoTypes          1.85MB ± 0%       1.85MB ± 0%    ~     (all equal)
Flate             328kB ± 0%        328kB ± 0%    ~     (all equal)
GoParser          402kB ± 0%        402kB ± 0%  -0.00%  (p=0.008 n=5+5)
Reflect          1.41MB ± 0%       1.41MB ± 0%  -0.20%  (p=0.008 n=5+5)
Tar               458kB ± 0%        458kB ± 0%    ~     (all equal)
XML               601kB ± 0%        599kB ± 0%  -0.21%  (p=0.008 n=5+5)

Change-Id: I9b5f25c8663a0b772ad1ee51fa61f74b74d26dd3
Reviewed-on: https://go-review.googlesource.com/c/143479
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Munday <mike.munday@ibm.com>
6 years agogo/build, go/doc: fix tautological conditions
Ainar Garipov [Fri, 9 Nov 2018 20:28:54 +0000 (23:28 +0300)]
go/build, go/doc: fix tautological conditions

These issues were found by the new vet's nilness check. The variables
were already checked against nil, so remove extra checks.

Change-Id: Ie252ccfcc755f3d06f691f354bf13d5a623fe17b
Reviewed-on: https://go-review.googlesource.com/c/148937
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agoruntime: ensure m.p is never stale
Nikhil Benesch [Fri, 9 Nov 2018 05:55:13 +0000 (00:55 -0500)]
runtime: ensure m.p is never stale

When a goroutine enters a syscall, its M unwires from its P to allow
the P to be retaken by another M if the syscall is slow. The M retains a
reference to its old P, however, so that if its old P has not been
retaken when the syscall returns, it can quickly reacquire that P.

The implementation, however, was confusing, as it left the reference to
the potentially-retaken P in m.p, which implied that the P was still
wired.

Make the code clearer by enforcing the invariant that m.p is never
stale. entersyscall now moves m.p to m.oldp and sets m.p to 0;
exitsyscall does the reverse, provided m.oldp has not been retaken.

With this scheme in place, the issue described in #27660 (assertion
failures in the race detector) would have resulted in a clean segfault
instead of silently corrupting memory.

Change-Id: Ib3e03623ebed4f410e852a716919fe4538858f0a
Reviewed-on: https://go-review.googlesource.com/c/148899
Run-TryBot: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/compile: re-run stringer
Josh Bleecher Snyder [Sat, 10 Nov 2018 15:45:19 +0000 (07:45 -0800)]
cmd/compile: re-run stringer

Some of the Ops got re-ordered with OSLICEHEADER. Re-generate.

Change-Id: I8ec3f5056537dd0b18cac2d4267453451ade2b00
Reviewed-on: https://go-review.googlesource.com/c/148824
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Martin Möhrmann <martisch@uos.de>
6 years agocmd/compile: simplify walk OCONVIFACE
Josh Bleecher Snyder [Mon, 5 Nov 2018 23:42:14 +0000 (15:42 -0800)]
cmd/compile: simplify walk OCONVIFACE

n.Type and n.Left.Type are used heavily. Give them useful names.

We generate the type word frequently. Make it a closure.
(We don't want to generate it up front, since there are some code
paths that don't need it, and generating it has side-effects.)

Simplify and document the final call construction.

Follow-up to address feedback on CL 147360.

Change-Id: I251134a55cf80d8b1676280a345d150f2288c09a
Reviewed-on: https://go-review.googlesource.com/c/147538
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Martin Möhrmann <moehrmann@google.com>
6 years agoall: add support for synchronous callbacks to js/wasm
Richard Musiol [Thu, 11 Oct 2018 10:46:14 +0000 (12:46 +0200)]
all: add support for synchronous callbacks to js/wasm

With this change, callbacks returned by syscall/js.NewCallback
get executed synchronously. This is necessary for the APIs of
many JavaScript libraries.

A callback triggered during a call from Go to JavaScript gets executed
on the same goroutine. A callback triggered by JavaScript's event loop
gets executed on an extra goroutine.

Fixes #26045
Fixes #27441

Change-Id: I591b9e85ab851cef0c746c18eba95fb02ea9e85b
Reviewed-on: https://go-review.googlesource.com/c/142004
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/compile: improve typechecking of OSLICEHEADER nodes
Martin Möhrmann [Fri, 2 Nov 2018 13:38:33 +0000 (14:38 +0100)]
cmd/compile: improve typechecking of OSLICEHEADER nodes

Create a new node for OSLICEHEADER nodes to ensure typechecks are applied.
Add nil checks for OSLICEHEADER type and pointer parameters
for better error messages when these are not set.
Improve formatting of OSLICEHEADER nodes in compiler error messages.

Change-Id: Idea8f41bb4beb636f0e1fc381ff8d79b1d44fbae
Reviewed-on: https://go-review.googlesource.com/c/146997
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
6 years agocmd/go: add go mod edit -go flag
Ian Lance Taylor [Sat, 3 Nov 2018 00:08:24 +0000 (17:08 -0700)]
cmd/go: add go mod edit -go flag

It can be used to set the Go language version used by the module.

RELNOTES=yes

Updates #28221

Change-Id: Ief0dd185c01195a17be20dff8627c80943c436e7
Reviewed-on: https://go-review.googlesource.com/c/147282
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
6 years agocmd/go: add go statement when initializing go.mod
Ian Lance Taylor [Fri, 2 Nov 2018 23:58:42 +0000 (16:58 -0700)]
cmd/go: add go statement when initializing go.mod

When creating a go.mod file, add a go statement mentioning the current
Go version. We can be reasonably confident that the current version is
able to build the module. This is as described in the language
transition proposal at https://golang.org/issue/28221.

Updates #28221

Change-Id: I70a99b3a53f4b6c0288da07473c5a71bb28cd86f
Reviewed-on: https://go-review.googlesource.com/c/147281
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
6 years agoruntime: don't check _defer against nil twice
Ainar Garipov [Fri, 9 Nov 2018 21:01:14 +0000 (00:01 +0300)]
runtime: don't check _defer against nil twice

This issue was found by the new vet's nilness check. _defer was already
checked against nil, so don't check it again.

Change-Id: I78725eaec7234b262b3c941e06441ca57f82bdd9
Reviewed-on: https://go-review.googlesource.com/c/148917
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agoRevert "os: remove sleep in windows Process.Wait"
Alex Brainman [Fri, 9 Nov 2018 21:32:45 +0000 (08:32 +1100)]
Revert "os: remove sleep in windows Process.Wait"

This reverts CL 145221 (commit 5c359736f8d67338b53c26aaef52139ae8cd0538)

Reason for revert: breaks the build occasionally.

Updates #23171
Updates #25965

Change-Id: Ie1e3c76ab9bcd8d28b6118440b5f80c76f9b1852
Reviewed-on: https://go-review.googlesource.com/c/148957
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agocmd/cgo: make the gccgo init function no_split_stack
Ian Lance Taylor [Fri, 9 Nov 2018 22:53:32 +0000 (14:53 -0800)]
cmd/cgo: make the gccgo init function no_split_stack

This works around what appears to be a bug in current clang (2018-11-09).
Details are in the comment in the code.

Change-Id: Ib4783b6c03d531c69ebc4cb0ac023bea5bee7d40
Reviewed-on: https://go-review.googlesource.com/c/148819
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agoruntime: never call into race detector with retaken P
Nikhil Benesch [Fri, 9 Nov 2018 05:55:13 +0000 (00:55 -0500)]
runtime: never call into race detector with retaken P

cgocall could previously invoke the race detector on an M whose P had
been retaken. The race detector would attempt to use the P-local state
from this stale P, racing with the thread that was actually wired to
that P. The result was memory corruption of ThreadSanitizer's internal
data structures that presented as hard-to-understand assertion failures
and segfaults.

Reorder cgocall so that it always acquires a P before invoking the race
detector, and add a test that stresses the interaction between cgo and
the race detector to protect against future bugs of this kind.

Fixes #27660.

Change-Id: Ide93f96a23490314d6647547140e0a412a97f0d4
Reviewed-on: https://go-review.googlesource.com/c/148717
Run-TryBot: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
6 years agoruntime: stop unnecessary span scavenges on free
Michael Anthony Knyszek [Tue, 6 Nov 2018 00:10:33 +0000 (00:10 +0000)]
runtime: stop unnecessary span scavenges on free

This change fixes a bug wherein freeing a scavenged span that didn't
coalesce with any neighboring spans would result in that span getting
scavenged again. This case may actually be a common occurance because
"freeing" span trimmings and newly-grown spans end up using the same
codepath. On systems where madvise is relatively expensive, this can
have a large performance impact.

This change also cleans up some of this logic in freeSpanLocked since
a number of factors made the coalescing code somewhat difficult to
reason about with respect to scavenging. Notably, the way the
needsScavenge boolean is handled could be better expressed and the
inverted conditions (e.g. !after.released) can make things even more
confusing.

Fixes #28595.

Change-Id: I75228dba70b6596b90853020b7c24fbe7ab937cf
Reviewed-on: https://go-review.googlesource.com/c/147559
Run-TryBot: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
6 years agocmd/compile: eliminate dead code
Josh Bleecher Snyder [Fri, 9 Nov 2018 19:45:04 +0000 (11:45 -0800)]
cmd/compile: eliminate dead code

During walkexpr, we were assessing whether shifts were bounded.
However, that information was dropped on the floor during SSA conversion.
The SSA backend already finds all bounded shifts that walkexpr could have,
and at negligible extra cost (0.02% in alloc, CPU undetectable).

Change-Id: Ieda1af1a2a3ec99bfdc2b0b704c9b80ce8a34486
Reviewed-on: https://go-review.googlesource.com/c/148897
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/compile: use ANDCC, ORCC, XORCC to avoid CMP on ppc64x
Lynn Boger [Wed, 10 Oct 2018 17:47:18 +0000 (13:47 -0400)]
cmd/compile: use ANDCC, ORCC, XORCC to avoid CMP on ppc64x

This change makes use of the cc versions of the AND, OR, XOR
instructions, omitting the need for a CMP instruction.

In many test programs and in the go binary, this reduces the
size of 20-30 functions by at least 1 instruction, many in
runtime.

Testcase added to test/codegen/comparisons.go

Change-Id: I6cc1ca8b80b065d7390749c625bc9784b0039adb
Reviewed-on: https://go-review.googlesource.com/c/143059
Reviewed-by: Carlos Eduardo Seo <cseo@linux.vnet.ibm.com>
Reviewed-by: Michael Munday <mike.munday@ibm.com>
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agobufio: fix indexes in TestWriter
Ryan Dahl [Fri, 9 Nov 2018 18:13:52 +0000 (18:13 +0000)]
bufio: fix indexes in TestWriter

Change-Id: I393c53d6f7b526d156226502544725a4cb9fb118
GitHub-Last-Rev: 5d53406c70ebbbbedd7650b9f43c873ead63c184
GitHub-Pull-Request: golang/go#28693
Reviewed-on: https://go-review.googlesource.com/c/148818
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agoRevert "cmd/vet: lostcancel: suppress the check in the main.main function"
Brad Fitzpatrick [Fri, 9 Nov 2018 17:27:39 +0000 (17:27 +0000)]
Revert "cmd/vet: lostcancel: suppress the check in the main.main function"

This reverts CL 148758 (commit 5e17ce22ece18571597ef9a214cdfcd461894e9c)

Reason for revert: breaks the build.

Change-Id: I6ed15b7b8f6b74d84edab9402ddf7ae87a0d0387
Reviewed-on: https://go-review.googlesource.com/c/148817
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/vet: lostcancel: suppress the check in the main.main function
Alan Donovan [Fri, 13 Apr 2018 18:27:33 +0000 (14:27 -0400)]
cmd/vet: lostcancel: suppress the check in the main.main function

When main.main returns, the process exits, so there's no need to cancel contexts.

This change was initially reviewed as
https://go-review.googlesource.com/c/go/+/106915/4
but somehow I messed up and committed patchset 5, which was
effectively empty.

Change-Id: Ic4250eb6563af9bc734e429aafc7081ca7d0e012
Reviewed-on: https://go-review.googlesource.com/c/148758
Reviewed-by: Alan Donovan <adonovan@google.com>
6 years agoruntime: reduce linear search through pcvalue cache
Josh Bleecher Snyder [Sun, 1 Apr 2018 23:39:34 +0000 (16:39 -0700)]
runtime: reduce linear search through pcvalue cache

This change introduces two optimizations together,
one for recursive and one for non-recursive stacks.

For recursive stacks, we introduce the new entry
at the beginning of the cache, so it can be found first.
This adds an extra read and write.
While we're here, switch from fastrandn, which does a multiply,
to fastrand % n, which does a shift.

For non-recursive stacks, split the cache from [16]pcvalueCacheEnt
into [2][8]pcvalueCacheEnt, and add a very cheap associative lookup.

name                old time/op  new time/op  delta
StackCopyPtr-8       118ms ± 1%   106ms ± 2%  -9.56%  (p=0.000 n=17+18)
StackCopy-8         95.8ms ± 1%  87.0ms ± 3%  -9.11%  (p=0.000 n=19+20)
StackCopyNoCache-8   135ms ± 2%   139ms ± 1%  +3.06%  (p=0.000 n=19+18)

During make.bash, the association function used has this return distribution:

percent count  return value
 53.23% 678797 1
 46.74% 596094 0

It is definitely not perfect, but it is pretty good,
and that's all we need.

Change-Id: I2cabb1d26b99c5111bc28f427016a2a5e6c620fd
Reviewed-on: https://go-review.googlesource.com/c/110564
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
6 years agocrypto/md5: fix md5block asm to work on big endian ppc64
Lynn Boger [Thu, 25 Oct 2018 15:13:16 +0000 (11:13 -0400)]
crypto/md5: fix md5block asm to work on big endian ppc64

This handles a TODO in the md5block_ppc64le.s file to
make use of byte reverse loads so the function works for
big endian as well as little endian. File name is now
md5block_ppc64x.s.

name                 old time/op    new time/op    delta
Hash8Bytes              537ns ± 0%     299ns ± 0%  -44.32%  (p=0.000 n=8+1)
Hash1K                 4.74µs ± 0%    3.24µs ± 0%  -31.64%  (p=0.250 n=7+1)
Hash8K                 34.4µs ± 0%    23.6µs ± 0%  -31.56%  (p=0.222 n=8+1)
Hash8BytesUnaligned     537ns ± 0%     298ns ± 0%  -44.51%  (p=0.000 n=8+1)
Hash1KUnaligned        4.74µs ± 0%    3.24µs ± 0%  -31.48%  (p=0.222 n=8+1)
Hash8KUnaligned        34.4µs ± 0%    23.6µs ± 0%  -31.39%  (p=0.222 n=8+1)

name                 old speed      new speed      delta
Hash8Bytes           14.9MB/s ± 0%  26.8MB/s ± 0%  +79.76%  (p=0.222 n=8+1)
Hash1K                216MB/s ± 0%   316MB/s ± 0%  +46.29%  (p=0.250 n=7+1)
Hash8K                238MB/s ± 0%   348MB/s ± 0%  +46.11%  (p=0.222 n=8+1)
Hash8BytesUnaligned  14.9MB/s ± 0%  26.8MB/s ± 0%  +79.76%  (p=0.222 n=8+1)
Hash1KUnaligned       216MB/s ± 0%   316MB/s ± 0%  +45.95%  (p=0.222 n=8+1)
Hash8KUnaligned       238MB/s ± 0%   347MB/s ± 0%  +45.75%  (p=0.222 n=8+1)

Change-Id: I2e226bf7e69e0acd49db1af42e4fd8b87b155606
Reviewed-on: https://go-review.googlesource.com/c/144599
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: Michael Munday <mike.munday@ibm.com>
6 years agocmd/vet: fix printf false negative with nested pointers
Daniel Martí [Tue, 6 Nov 2018 21:35:23 +0000 (21:35 +0000)]
cmd/vet: fix printf false negative with nested pointers

Pointers to compound objects (structs, slices, arrays, maps) are only
followed by fmt if the pointer is at the top level of an argument. This
is to minimise the chances of fmt running into loops.

However, vet did not follow this rule. It likely doesn't help that fmt
does not document that restriction well, which is being tracked in
 #28625.

Updates #27672.

Change-Id: Ie9bbd9b974eda5ab9a285986d207ef92fca4453e
Reviewed-on: https://go-review.googlesource.com/c/147997
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
6 years agocmd/vet: fix some pointer false positives in printf
Daniel Martí [Tue, 6 Nov 2018 15:14:56 +0000 (15:14 +0000)]
cmd/vet: fix some pointer false positives in printf

fmt's godoc reads:

For compound objects, the elements are printed using these
rules, recursively, laid out like this:

struct:             {field0 field1 ...}
array, slice:       [elem0 elem1 ...]
maps:               map[key1:value1 key2:value2 ...]
pointer to above:   &{}, &[], &map[]

That is, a pointer to a struct, array, slice, or map, can be correctly
printed by fmt if the type pointed to can be printed without issues.

vet was only following this rule for pointers to structs, omitting
arrays, slices, and maps. Fix that, and add tests for all the
combinations.

Updates #27672.

Change-Id: Ie61ebe1fffc594184f7b24d7dbf72d7d5de78309
Reviewed-on: https://go-review.googlesource.com/c/147758
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
6 years agomime/multipart: check for quoted-printable case insensitively
Brad Fitzpatrick [Thu, 8 Nov 2018 18:46:34 +0000 (18:46 +0000)]
mime/multipart: check for quoted-printable case insensitively

Fixes #28674

Change-Id: Id88e0a4b86b50eb45f0d968d7e4bbe66b7f37f82
Reviewed-on: https://go-review.googlesource.com/c/148579
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agocmd/go/internal/cache: fix wrong/old function name in comment
Brad Fitzpatrick [Thu, 8 Nov 2018 20:06:05 +0000 (20:06 +0000)]
cmd/go/internal/cache: fix wrong/old function name in comment

Change-Id: Ia0caf2fb06097ac184f78779334460900e8c0149
Reviewed-on: https://go-review.googlesource.com/c/148580
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agocmd/compile: only optimize chained Moves on disjoint stack mem
Josh Bleecher Snyder [Sun, 28 Oct 2018 19:01:11 +0000 (12:01 -0700)]
cmd/compile: only optimize chained Moves on disjoint stack mem

This optimization is not sound if A, B, or C
might overlap with each other.

Thanks to Michael Munday for pointing this
out during the review of CL 143479.

This reduces the number of times this optimization
triggers during make.bash from 386 to 74.

This is unfortunate, but I don't see an obvious way around it,
short of souping up the disjointness analysis.

name        old object-bytes  new object-bytes  delta
Template          507kB ± 0%        507kB ± 0%   +0.13%  (p=0.008 n=5+5)
Unicode           225kB ± 0%        225kB ± 0%     ~     (all equal)
GoTypes          1.85MB ± 0%       1.85MB ± 0%   +0.02%  (p=0.008 n=5+5)
Flate             328kB ± 0%        328kB ± 0%     ~     (all equal)
GoParser          402kB ± 0%        402kB ± 0%     ~     (all equal)
Reflect          1.41MB ± 0%       1.41MB ± 0%     ~     (all equal)
Tar               457kB ± 0%        458kB ± 0%   +0.20%  (p=0.008 n=5+5)
XML               600kB ± 0%        601kB ± 0%   +0.03%  (p=0.008 n=5+5)

Change-Id: Ida408cb627145ba9faf473a78606f050c2f3f51c
Reviewed-on: https://go-review.googlesource.com/c/145208
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Munday <mike.munday@ibm.com>
6 years agodatabase/sql: add support for returning cursors to client
Daniel Theophanes [Mon, 29 Oct 2018 23:22:37 +0000 (16:22 -0700)]
database/sql: add support for returning cursors to client

This CL add support for converting a returned cursor (presented
to this package as a driver.Rows) and scanning it into a *Rows.

Fixes #28515

Change-Id: Id8191c568dc135af9e5e8555efcd01987708edcb
Reviewed-on: https://go-review.googlesource.com/c/145738
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agostrings,bytes: use inlineable function trampolines instead of linkname
Keith Randall [Mon, 5 Nov 2018 19:00:01 +0000 (11:00 -0800)]
strings,bytes: use inlineable function trampolines instead of linkname

Cleans things up quite a bit.

There's still a few more, like runtime.cmpstring, which might also
be worth fixing.

Change-Id: Ide18dd621efc129cc686db223f47fa0b044b5580
Reviewed-on: https://go-review.googlesource.com/c/148578
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
6 years agotime: add a missing comma in the documentation of Time
Agniva De Sarker [Thu, 8 Nov 2018 07:22:47 +0000 (12:52 +0530)]
time: add a missing comma in the documentation of Time

Updates #28421

Change-Id: I3262c83669bc3cefd2cea6a612e3dc1d4318b2c2
Reviewed-on: https://go-review.googlesource.com/c/148339
Reviewed-by: Rob Pike <r@golang.org>
6 years agocmd/go: vet: revert $GOVETTOOL env var, restore -vettool flag
Alan Donovan [Thu, 8 Nov 2018 15:43:40 +0000 (10:43 -0500)]
cmd/go: vet: revert $GOVETTOOL env var, restore -vettool flag

The environment variable is no longer necessary as we now plan to
transition to the new vet by replacing it in a single step,
and we really don't want to add more environment variables.

Fixes #28636

Change-Id: Ib85e5c0d61213b7b9f6a53d9376fec29525df971
Reviewed-on: https://go-review.googlesource.com/c/148497
Run-TryBot: Alan Donovan <adonovan@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
6 years agotime: handle negative offsets on js/wasm
Agniva De Sarker [Thu, 8 Nov 2018 04:18:06 +0000 (09:48 +0530)]
time: handle negative offsets on js/wasm

Fixes #28649

Change-Id: I9f6807ee3c3007f670dd509780805c7b255a2bda
Reviewed-on: https://go-review.googlesource.com/c/148338
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agocmd/compile: encourage inlining of functions with single-call bodies
Keith Randall [Mon, 5 Nov 2018 03:23:08 +0000 (19:23 -0800)]
cmd/compile: encourage inlining of functions with single-call bodies

This is a simple tweak to allow a bit more mid-stack inlining.
In cases like this:

func f() {
    g()
}

We'd really like to inline f into its callers. It can't hurt.

We implement this optimization by making calls a bit cheaper, enough
to afford a single call in the function body, but not 2.
The remaining budget allows for some argument modification, or perhaps
a wrapping conditional:

func f(x int) {
    g(x, 0)
}
func f(x int) {
    if x > 0 {
        g()
    }
}

Update #19348

Change-Id: Ifb1ea0dd1db216c3fd5c453c31c3355561fe406f
Reviewed-on: https://go-review.googlesource.com/c/147361
Reviewed-by: Austin Clements <austin@google.com>
Reviewed-by: David Chase <drchase@google.com>
6 years agointernal/syscall/unix: use libc calls on Darwin
Keith Randall [Thu, 8 Nov 2018 05:23:47 +0000 (21:23 -0800)]
internal/syscall/unix: use libc calls on Darwin

Add unexported unlinkat, openat, and fstatat calls, so that
the internal/syscall/unix package can use them.

Change-Id: I1df81ecae6427211dd392ec68c9f020fe131a526
Reviewed-on: https://go-review.googlesource.com/c/148457
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agostrings: remove empty line
沈涛 [Thu, 8 Nov 2018 11:25:20 +0000 (11:25 +0000)]
strings: remove empty line

Change-Id: Ibdca4f7002585b00d7f69d710285a8e0f69c598a
GitHub-Last-Rev: eb8f800c986c8ac4a81705158ecc730c35e1c5c2
GitHub-Pull-Request: golang/go#28659
Reviewed-on: https://go-review.googlesource.com/c/148477
Reviewed-by: Russ Cox <rsc@golang.org>
6 years agonet: simplify nested if-blocks
Mikio Hara [Thu, 8 Nov 2018 00:48:55 +0000 (09:48 +0900)]
net: simplify nested if-blocks

Change-Id: I32e1829c955a48d8c4566430c13679e237bb0611
Reviewed-on: https://go-review.googlesource.com/c/148337
Run-TryBot: Mikio Hara <mikioh.public.networking@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agocrypto/tls: remove a forgotten note to future self
Filippo Valsorda [Tue, 6 Nov 2018 02:00:52 +0000 (21:00 -0500)]
crypto/tls: remove a forgotten note to future self

Now, this is embarrassing. While preparing CL 142818, I noticed a
possible vulnerability in the existing code which I was rewriting. I
took a note to go back and assess if it was indeed an issue, and in case
start the security release process. The note unintentionally slipped
into the commit. Fortunately, there was no vulnerability.

What caught my eye was that I had fixed the calculation of the minimum
encrypted payload length from

    roundUp(explicitIVLen+macSize+1, blockSize)

to (using the same variable names)

    explicitIVLen + roundUp(macSize+1, blockSize)

The explicit nonce sits outside of the encrypted payload, so it should
not be part of the value rounded up to the CBC block size.

You can see that for some values of the above, the old result could be
lower than the correct value. An unexpectedly short payload might cause
a panic during decryption (a DoS vulnerability) or even more serious
issues due to the constant time code that follows it (see for example
Yet Another Padding Oracle in OpenSSL CBC Ciphersuites [1]).

In practice, explicitIVLen is either zero or equal to blockSize, so it
does not change the amount of rounding up necessary and the two
formulations happen to be identical. Nothing to see here.

It looked more suspicious than it is in part due to the fact that the
explicitIVLen definition moved farther into hc.explicitNonceLen() and
changed name from IV (which suggests a block length) to nonce (which
doesn't necessarily). But anyway it was never meant to surface or be
noted, except it slipped, so here we are for a boring explanation.

[1] https://blog.cloudflare.com/yet-another-padding-oracle-in-openssl-cbc-ciphersuites/

Change-Id: I365560dfe006513200fa877551ce7afec9115fdf
Reviewed-on: https://go-review.googlesource.com/c/147637
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agotime: document that a marshaled time does not include location name
Agniva De Sarker [Wed, 7 Nov 2018 14:26:30 +0000 (19:56 +0530)]
time: document that a marshaled time does not include location name

Fixes #28421

Change-Id: I00878ec246d5249d910f2b57749f74cfc38dbec6
Reviewed-on: https://go-review.googlesource.com/c/148117
Reviewed-by: Rob Pike <r@golang.org>
6 years agosyscall: move uses of Syscall to libSystem on darwin
Keith Randall [Wed, 7 Nov 2018 23:27:16 +0000 (15:27 -0800)]
syscall: move uses of Syscall to libSystem on darwin

Miscellaneous additional conversions from raw syscalls
to using their libc equivalent.

Update #17490

Change-Id: If9ab22cc1d676c1f20fb161ebf02b0c28f71585d
Reviewed-on: https://go-review.googlesource.com/c/148257
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agocmd/go: add goversion environment variable to testing script language
Ian Lance Taylor [Fri, 2 Nov 2018 23:38:11 +0000 (16:38 -0700)]
cmd/go: add goversion environment variable to testing script language

Updates #28221

Change-Id: I8a1e352cd9122bce200d45c6b19955cb50308d71
Reviewed-on: https://go-review.googlesource.com/c/147280
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
6 years agocmd/go: document the go statement in a go.mod file
Ian Lance Taylor [Fri, 2 Nov 2018 23:29:04 +0000 (16:29 -0700)]
cmd/go: document the go statement in a go.mod file

Change-Id: I8db276ec371de56871ce3250f27de1d1dee4b473
Reviewed-on: https://go-review.googlesource.com/c/147279
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
6 years agocmd/go: pass go language version to cmd/compile
Ian Lance Taylor [Fri, 2 Nov 2018 23:00:33 +0000 (16:00 -0700)]
cmd/go: pass go language version to cmd/compile

Pass the Go language version specified in the go.mod file to
cmd/compile.

Also, change the behavior when the go.mod file requests a Go version
that is later than the current one. Previously cmd/go would give a
fatal error in this situation. With this change it attempts the
compilation, and if (and only if) the compilation fails it adds a note
saying that the requested Go version is newer than the known version.
This is as described in https://golang.org/issue/28221.

Updates #28221.

Change-Id: I46803813e7872d4a418a3fd5299880be3b73a971
Reviewed-on: https://go-review.googlesource.com/c/147278
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
6 years agocmd/go: add cmpenv command to testing script language
Ian Lance Taylor [Fri, 2 Nov 2018 22:42:15 +0000 (15:42 -0700)]
cmd/go: add cmpenv command to testing script language

This will be used by later commits in this sequence.

Updates #28221

Change-Id: I2b22b9f88a0183636cde9509606f03f079eb33f1
Reviewed-on: https://go-review.googlesource.com/c/147277
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
6 years agocmd/go: add /v2 to go.mod require example
diplozoon [Fri, 26 Oct 2018 05:03:15 +0000 (05:03 +0000)]
cmd/go: add /v2 to go.mod require example

I added /v2 to go.mod require example

Fixes #28374

Change-Id: I74cca374838d106eb79acb9189a02fe9443962c0
Reviewed-on: https://go-review.googlesource.com/c/144917
Reviewed-by: Bryan C. Mills <bcmills@google.com>
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agosyscall: implement syscalls on Darwin using libSystem
Keith Randall [Mon, 24 Sep 2018 14:13:36 +0000 (07:13 -0700)]
syscall: implement syscalls on Darwin using libSystem

There are still some references to the bare Syscall functions
in the stdlib. I will root those out in a following CL.
(This CL is big enough as it is.)
Most are in vendor directories:

cmd/vendor/golang.org/x/sys/unix/
vendor/golang_org/x/net/route/syscall.go
syscall/bpf_bsd.go
syscall/exec_unix.go
syscall/flock.go

Update #17490

Change-Id: I69ab707811530c26b652b291cadee92f5bf5c1a4
Reviewed-on: https://go-review.googlesource.com/c/141639
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Elias Naur <elias.naur@gmail.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agogo/types: avoid certain problems with recursive alias type declarations
Robert Griesemer [Mon, 5 Nov 2018 23:51:11 +0000 (15:51 -0800)]
go/types: avoid certain problems with recursive alias type declarations

It is possible to create certain recursive type declarations involving
alias types which cause the type-checker to produce an (invalid) type
for the alias because it is not yet available. By type-checking alias
declarations in a 2nd phase, the problem is mitigated a bit since it
requires more convoluted alias declarations for the problem to appear.

Also re-enable testing of fixedbugs/issue27232.go again (which was the
original cause for this change).

Updates #28576.

Change-Id: If6f9656a95262e6575b01c4a003094d41551564b
Reviewed-on: https://go-review.googlesource.com/c/147597
Reviewed-by: Alan Donovan <adonovan@google.com>
6 years agocmd/internal/obj/arm64: encode large constants into MOVZ/MOVN and MOVK instructions
fanzha02 [Thu, 10 May 2018 05:58:16 +0000 (05:58 +0000)]
cmd/internal/obj/arm64: encode large constants into MOVZ/MOVN and MOVK instructions

Current assembler gets large constants from constant pool, this CL
gets rid of the pool by using MOVZ/MOVN and MOVK to load large
constants.

This CL changes the assembler behavior as follows.

1. go assembly  1, MOVD $0x1111222233334444, R1
                2, MOVD $0x1111ffff1111ffff, R1
   previous version: MOVD 0x9a4, R1 (loads constant from pool).
   optimized version: 1, MOVD $0x4444, R1; MOVK $(0x3333<<16), R1; MOVK $(0x2222<<32), R1;
   MOVK $(0x1111<<48), R1. 2, MOVN $(0xeeee<<16), R1; MOVK $(0x1111<<48), R1.

Add test cases, and below are binary size comparison and bechmark results.

1. Binary size before/after
binary                 size change
pkg/linux_arm64        +25.4KB
pkg/tool/linux_arm64   -2.9KB
go                     -2KB
gofmt                  no change

2. compiler benchmark.
name       old time/op       new time/op       delta
Template         574ms ±21%        577ms ±14%     ~     (p=0.853 n=10+10)
Unicode          327ms ±29%        353ms ±23%     ~     (p=0.360 n=10+8)
GoTypes          1.97s ± 8%        2.04s ±11%     ~     (p=0.143 n=10+10)
Compiler         9.13s ± 9%        9.25s ± 8%     ~     (p=0.684 n=10+10)
SSA              29.2s ± 5%        27.0s ± 4%   -7.40%  (p=0.000 n=10+10)
Flate            402ms ±40%        308ms ± 6%  -23.29%  (p=0.004 n=10+10)
GoParser         470ms ±26%        382ms ±10%  -18.82%  (p=0.000 n=9+10)
Reflect          1.36s ±16%        1.17s ± 7%  -13.92%  (p=0.001 n=9+10)
Tar              561ms ±19%        466ms ±15%  -17.08%  (p=0.000 n=9+10)
XML              745ms ±20%        679ms ±20%     ~     (p=0.123 n=10+10)
StdCmd           35.5s ± 6%        37.2s ± 3%   +4.81%  (p=0.001 n=9+8)

name       old user-time/op  new user-time/op  delta
Template         625ms ±14%        660ms ±18%     ~     (p=0.343 n=10+10)
Unicode          355ms ±10%        373ms ±20%     ~     (p=0.346 n=9+10)
GoTypes          2.39s ± 8%        2.37s ± 5%     ~     (p=0.897 n=10+10)
Compiler         11.1s ± 4%        11.4s ± 2%   +2.63%  (p=0.010 n=10+9)
SSA              35.4s ± 3%        34.9s ± 2%     ~     (p=0.113 n=10+9)
Flate            402ms ±13%        371ms ±30%     ~     (p=0.089 n=10+9)
GoParser         513ms ± 8%        489ms ±24%   -4.76%  (p=0.039 n=9+9)
Reflect          1.52s ±12%        1.41s ± 5%   -7.32%  (p=0.001 n=9+10)
Tar              607ms ±10%        558ms ± 8%   -7.96%  (p=0.009 n=9+10)
XML              828ms ±10%        789ms ±12%     ~     (p=0.059 n=10+10)

name       old text-bytes    new text-bytes    delta
HelloSize        714kB ± 0%        712kB ± 0%   -0.23%  (p=0.000 n=10+10)
CmdGoSize       8.26MB ± 0%       8.25MB ± 0%   -0.14%  (p=0.000 n=10+10)

name       old data-bytes    new data-bytes    delta
HelloSize       10.5kB ± 0%       10.5kB ± 0%     ~     (all equal)
CmdGoSize        258kB ± 0%        258kB ± 0%     ~     (all equal)

name       old bss-bytes     new bss-bytes     delta
HelloSize        125kB ± 0%        125kB ± 0%     ~     (all equal)
CmdGoSize        146kB ± 0%        146kB ± 0%     ~     (all equal)

name       old exe-bytes     new exe-bytes     delta
HelloSize       1.18MB ± 0%       1.18MB ± 0%     ~     (all equal)
CmdGoSize       11.2MB ± 0%       11.2MB ± 0%   -0.13%  (p=0.000 n=10+10)

3. go1 benckmark.
name                   old time/op    new time/op    delta
BinaryTree17              6.60s ±18%     7.36s ±22%    ~     (p=0.222 n=5+5)
Fannkuch11                4.04s ± 0%     4.05s ± 0%    ~     (p=0.421 n=5+5)
FmtFprintfEmpty          91.8ns ±14%    91.2ns ± 9%    ~     (p=0.667 n=5+5)
FmtFprintfString          145ns ± 0%     151ns ± 6%    ~     (p=0.397 n=4+5)
FmtFprintfInt             169ns ± 0%     176ns ± 5%  +4.14%  (p=0.016 n=4+5)
FmtFprintfIntInt          229ns ± 2%     243ns ± 6%    ~     (p=0.143 n=5+5)
FmtFprintfPrefixedInt     343ns ± 0%     350ns ± 3%  +1.92%  (p=0.048 n=5+5)
FmtFprintfFloat           400ns ± 3%     394ns ± 3%    ~     (p=0.063 n=5+5)
FmtManyArgs              1.04µs ± 0%    1.05µs ± 0%  +1.62%  (p=0.029 n=4+4)
GobDecode                13.9ms ± 4%    13.9ms ± 5%    ~     (p=1.000 n=5+5)
GobEncode                10.6ms ± 4%    10.6ms ± 5%    ~     (p=0.421 n=5+5)
Gzip                      567ms ± 1%     563ms ± 4%    ~     (p=0.548 n=5+5)
Gunzip                   60.2ms ± 1%    60.4ms ± 0%    ~     (p=0.056 n=5+5)
HTTPClientServer          114µs ± 4%     108µs ± 7%    ~     (p=0.095 n=5+5)
JSONEncode               18.4ms ± 2%    17.8ms ± 2%  -3.06%  (p=0.016 n=5+5)
JSONDecode                105ms ± 1%     103ms ± 2%    ~     (p=0.056 n=5+5)
Mandelbrot200            5.48ms ± 0%    5.49ms ± 0%    ~     (p=0.841 n=5+5)
GoParse                  6.05ms ± 1%    6.05ms ± 2%    ~     (p=1.000 n=5+5)
RegexpMatchEasy0_32       143ns ± 1%     146ns ± 4%  +2.10%  (p=0.048 n=4+5)
RegexpMatchEasy0_1K       499ns ± 1%     492ns ± 2%    ~     (p=0.079 n=5+5)
RegexpMatchEasy1_32       137ns ± 0%     136ns ± 1%  -0.73%  (p=0.016 n=4+5)
RegexpMatchEasy1_1K       826ns ± 4%     823ns ± 2%    ~     (p=0.841 n=5+5)
RegexpMatchMedium_32      224ns ± 5%     233ns ± 8%    ~     (p=0.119 n=5+5)
RegexpMatchMedium_1K     59.6µs ± 0%    59.3µs ± 1%  -0.66%  (p=0.016 n=4+5)
RegexpMatchHard_32       3.29µs ± 3%    3.26µs ± 1%    ~     (p=0.889 n=5+5)
RegexpMatchHard_1K       98.8µs ± 2%    99.0µs ± 0%    ~     (p=0.690 n=5+5)
Revcomp                   1.02s ± 1%     1.01s ± 1%    ~     (p=0.095 n=5+5)
Template                  135ms ± 5%     131ms ± 1%    ~     (p=0.151 n=5+5)
TimeParse                 591ns ± 0%     593ns ± 0%  +0.20%  (p=0.048 n=5+5)
TimeFormat                655ns ± 2%     607ns ± 0%  -7.42%  (p=0.016 n=5+4)
[Geo mean]               93.5µs         93.8µs       +0.23%

name                   old speed      new speed      delta
GobDecode              55.1MB/s ± 4%  55.1MB/s ± 4%    ~     (p=1.000 n=5+5)
GobEncode              72.4MB/s ± 4%  72.3MB/s ± 5%    ~     (p=0.421 n=5+5)
Gzip                   34.2MB/s ± 1%  34.5MB/s ± 4%    ~     (p=0.548 n=5+5)
Gunzip                  322MB/s ± 1%   321MB/s ± 0%    ~     (p=0.056 n=5+5)
JSONEncode              106MB/s ± 2%   109MB/s ± 2%  +3.16%  (p=0.016 n=5+5)
JSONDecode             18.5MB/s ± 1%  18.8MB/s ± 2%    ~     (p=0.056 n=5+5)
GoParse                9.57MB/s ± 1%  9.57MB/s ± 2%    ~     (p=0.952 n=5+5)
RegexpMatchEasy0_32     223MB/s ± 1%   221MB/s ± 0%  -1.10%  (p=0.029 n=4+4)
RegexpMatchEasy0_1K    2.05GB/s ± 1%  2.08GB/s ± 2%    ~     (p=0.095 n=5+5)
RegexpMatchEasy1_32     232MB/s ± 0%   234MB/s ± 1%  +0.76%  (p=0.016 n=4+5)
RegexpMatchEasy1_1K    1.24GB/s ± 4%  1.24GB/s ± 2%    ~     (p=0.841 n=5+5)
RegexpMatchMedium_32   4.45MB/s ± 5%  4.20MB/s ± 1%  -5.63%  (p=0.000 n=5+4)
RegexpMatchMedium_1K   17.2MB/s ± 0%  17.3MB/s ± 1%  +0.66%  (p=0.016 n=4+5)
RegexpMatchHard_32     9.73MB/s ± 3%  9.83MB/s ± 1%    ~     (p=0.889 n=5+5)
RegexpMatchHard_1K     10.4MB/s ± 2%  10.3MB/s ± 0%    ~     (p=0.635 n=5+5)
Revcomp                 249MB/s ± 1%   252MB/s ± 1%    ~     (p=0.095 n=5+5)
Template               14.4MB/s ± 4%  14.8MB/s ± 1%    ~     (p=0.151 n=5+5)
[Geo mean]             62.1MB/s       62.3MB/s       +0.34%

Fixes #10108

Change-Id: I79038f3c4c2ff874c136053d1a2b1c8a5a9cfac5
Reviewed-on: https://go-review.googlesource.com/c/118796
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agointernal/traceparser: skip test on iOS
Elias Naur [Wed, 7 Nov 2018 09:28:38 +0000 (10:28 +0100)]
internal/traceparser: skip test on iOS

Change-Id: Ifc9581ba82a13f507c288282b517ebf8a5f93b4e
Reviewed-on: https://go-review.googlesource.com/c/148058
Run-TryBot: Elias Naur <elias.naur@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
6 years agocmd/vendor: update github.com/google/pprof
Hana (Hyang-Ah) Kim [Tue, 6 Nov 2018 22:48:08 +0000 (17:48 -0500)]
cmd/vendor: update github.com/google/pprof

Sync @ fde099a (Oct 26, 2018)

Also update misc/nacl/testzip.proto to include new testdata.
Change-Id: If41590be9f395a591056e89a417b589c4ba71b1a
Reviewed-on: https://go-review.googlesource.com/c/147979
Run-TryBot: Hyang-Ah Hana Kim <hyangah@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agogo/internal/gcimporter: ensure tests pass even if GOROOT is read-only
Diogo Pinela [Tue, 30 Oct 2018 22:58:24 +0000 (22:58 +0000)]
go/internal/gcimporter: ensure tests pass even if GOROOT is read-only

This mainly entails writing compiler output files to a temporary
directory, as well as the corrupted files in TestVersionHandling.

Updates #28387

Change-Id: I6b3619a91fff27011c7d73daa4febd14a6c5c348
Reviewed-on: https://go-review.googlesource.com/c/146119
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
6 years agocmd/compile: update TestNexting golden file for Delve
David Chase [Tue, 6 Nov 2018 21:42:13 +0000 (16:42 -0500)]
cmd/compile: update TestNexting golden file for Delve

This change updates the expected output of the delve debugging session
in the TestNexting internal/ssa test, aligning it with the changes
introduced in CL 147360 and earlier.

Change-Id: I1cc788d02433624a36f4690f24201569d765e5d3
Reviewed-on: https://go-review.googlesource.com/c/147998
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
6 years agocmd/compile: update TestNexting golden file
Alberto Donizetti [Tue, 6 Nov 2018 21:08:30 +0000 (22:08 +0100)]
cmd/compile: update TestNexting golden file

This change updates the expected output of the gdb debugging session
in the TestNexting internal/ssa test, aligning it with the changes
introduced in CL 147360.

Fixes the longtest builder.

Change-Id: I5b5c22e1cf5e205967ff8359dc6c1485c815428e
Reviewed-on: https://go-review.googlesource.com/c/147957
Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com>
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/go: make 'go test -h' print two lines
Daniel Martí [Thu, 25 Oct 2018 12:42:46 +0000 (13:42 +0100)]
cmd/go: make 'go test -h' print two lines

Like every other command's -h flag. To achieve this, pass the command's
usage function to the cmdflag package, since that package is used by
multiple commands and cannot directly access *base.Command.

This also lets us get rid of testFlag1 and testFlag2, and instead have
contiguous raw strings for the test and testflag help docs.

Fixes #26999.

Change-Id: I2ebd66835ee61fa83270816a01fa312425224bb3
Reviewed-on: https://go-review.googlesource.com/c/144558
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
6 years agocmd/compile: add new format to known_formats
Alberto Donizetti [Tue, 6 Nov 2018 18:43:55 +0000 (19:43 +0100)]
cmd/compile: add new format to known_formats

This change fixes a TestFormat failure in fmt_test by adding a
recently introduced new known format (%q for syntax.Error).

Fixes #28621

Change-Id: I026ec88c334549a957a692c1652a860c57e23dae
Reviewed-on: https://go-review.googlesource.com/c/147837
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agocmd/asm: rename R18 to R18_PLATFORM on ARM64
Cherry Zhang [Fri, 2 Nov 2018 20:51:14 +0000 (16:51 -0400)]
cmd/asm: rename R18 to R18_PLATFORM on ARM64

In ARM64 ABI, R18 is the "platform register", the use of which is
OS specific. The OS could choose to reserve this register. In
practice, it seems fine to use R18 on Linux but not on darwin (iOS).

Rename R18 to R18_PLATFORM to prevent accidental use. There is no
R18 usage within the standard library (besides tests, which are
updated).

Fixes #26110

Change-Id: Icef7b9549e2049db1df307a0180a3c90a12d7a84
Reviewed-on: https://go-review.googlesource.com/c/147218
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agocmd/compile: don't deadcode eliminate labels
Keith Randall [Tue, 6 Nov 2018 18:16:17 +0000 (10:16 -0800)]
cmd/compile: don't deadcode eliminate labels

Dead-code eliminating labels is tricky because there might
be gotos that can still reach them.

Bug probably introduced with CL 91056

Fixes #28616

Change-Id: I6680465134e3486dcb658896f5172606cc51b104
Reviewed-on: https://go-review.googlesource.com/c/147817
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Iskander Sharipov <iskander.sharipov@intel.com>
6 years agocmd/vendor: add x/tools/go/analysis/cmd/vet-lite + deps
Alan Donovan [Mon, 5 Nov 2018 20:51:35 +0000 (15:51 -0500)]
cmd/vendor: add x/tools/go/analysis/cmd/vet-lite + deps

This change adds the vet-lite command (the future cmd/vet) and all its
dependencies from x/tools, but not its tests and their dependencies.
It was created with these commands:

  $ (cd $GOPATH/src/golang.org/x/tools && git checkout c76e1ad)
  $ cd GOROOT/src/cmd
  $ govendor add $(go list -deps golang.org/x/tools/go/analysis/cmd/vet-lite | grep golang.org/x/tools)
  $ rm -fr $(find vendor/golang.org/x/tools/ -name testdata)
  $ rm $(find vendor/golang.org/x/tools/ -name \*_test.go)

I feel sure I am holding govendor wrong. Please advise.

A followup CL will make cmd/vet behave like vet-lite, initially just
for users that opt in, and soon after for all users, at which point
cmd/vet will be replaced in its entirety by a copy of vet-lite's small
main.go.

In the meantime, anyone can try the new tool using these commands:

 $ go build cmd/vendor/golang.org/x/tools/go/analysis/cmd/vet-lite
 $ export GOVETTOOL=$(which vet-lite)
 $ go vet your/project/...

Change-Id: Iea168111a32ce62f82f9fb706385ca0f368bc869
Reviewed-on: https://go-review.googlesource.com/c/147444
Reviewed-by: Russ Cox <rsc@golang.org>
6 years agoruntime: improve performance of memclr, memmove on ppc64x
Lynn Boger [Fri, 5 Oct 2018 18:21:39 +0000 (14:21 -0400)]
runtime: improve performance of memclr, memmove on ppc64x

This improves the asm implementations for memmove and memclr on
ppc64x through use of vsx loads and stores when size is >= 32 bytes.
For memclr, dcbz is used when the size is >= 512 and aligned to 128.

Memclr/64       13.3ns ± 0%     10.7ns ± 0%   -19.55%  (p=0.000 n=8+7)
Memclr/96       14.9ns ± 0%     11.4ns ± 0%   -23.49%  (p=0.000 n=8+8)
Memclr/128      16.3ns ± 0%     12.3ns ± 0%   -24.54%  (p=0.000 n=8+8)
Memclr/160      17.3ns ± 0%     13.0ns ± 0%   -24.86%  (p=0.000 n=8+8)
Memclr/256      20.0ns ± 0%     15.3ns ± 0%   -23.62%  (p=0.000 n=8+8)
Memclr/512      34.2ns ± 0%     10.2ns ± 0%   -70.20%  (p=0.000 n=8+8)
Memclr/4096      178ns ± 0%       23ns ± 0%   -87.13%  (p=0.000 n=8+8)
Memclr/65536    2.67µs ± 0%     0.30µs ± 0%   -88.89%  (p=0.000 n=7+8)
Memclr/1M       43.2µs ± 0%     10.0µs ± 0%   -76.85%  (p=0.000 n=8+8)
Memclr/4M        173µs ± 0%       40µs ± 0%   -76.88%  (p=0.000 n=8+8)
Memclr/8M        349µs ± 0%       82µs ± 0%   -76.58%  (p=0.000 n=8+8)
Memclr/16M       701µs ± 7%      672µs ± 0%    -4.05%  (p=0.040 n=8+7)
Memclr/64M      2.70ms ± 0%     2.67ms ± 0%    -0.96%  (p=0.000 n=8+7)

Memmove/32      6.59ns ± 0%    5.84ns ± 0%  -11.34%  (p=0.029 n=4+4)
Memmove/64      7.91ns ± 0%    6.97ns ± 0%  -11.92%  (p=0.029 n=4+4)
Memmove/128     10.5ns ± 0%     8.8ns ± 0%  -16.24%  (p=0.029 n=4+4)
Memmove/256     21.0ns ± 0%    12.9ns ± 0%  -38.57%  (p=0.029 n=4+4)
Memmove/512     28.4ns ± 0%    26.2ns ± 0%   -7.75%  (p=0.029 n=4+4)
Memmove/1024    48.2ns ± 1%    39.4ns ± 0%  -18.26%  (p=0.029 n=4+4)
Memmove/2048    85.4ns ± 0%    69.0ns ± 0%  -19.20%  (p=0.029 n=4+4)
Memmove/4096     159ns ± 0%     128ns ± 0%  -19.50%  (p=0.029 n=4+4)

Change-Id: I8c1adf88790845bf31444a15249456006eb5bf8b
Reviewed-on: https://go-review.googlesource.com/c/141217
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Munday <mike.munday@ibm.com>
6 years agocmd/compile/internal/gc: update cgo_import_dynamic for AIX
Clément Chigot [Fri, 2 Nov 2018 12:14:07 +0000 (13:14 +0100)]
cmd/compile/internal/gc: update cgo_import_dynamic for AIX

On AIX, cmd/link needs two information in order to generate a dynamic
import, the library and its object needed. Currently, cmd/link isn't
able to retrieve this object only with the name of the library.
Therefore, the library pattern in cgo_import_dynamic must be
"lib.a/obj.o".

Change-Id: Ib8b8aaa9807c9fa6af46ece4e312d58073ed6ec1
Reviewed-on: https://go-review.googlesource.com/c/146957
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agoreflect: fix StructOf panics from too many methods in embedded fields
Raghavendra Nagaraj [Tue, 6 Nov 2018 09:02:03 +0000 (09:02 +0000)]
reflect: fix StructOf panics from too many methods in embedded fields

Previously we panicked if the number of methods present for an embedded
field was >= 32. This change removes that limit and now StructOf
dynamically calls itself to create space for the number of methods.

Fixes #25402

Change-Id: I3b1deb119796d25f7e6eee1cdb126327b49a0b5e
GitHub-Last-Rev: 16da71ad6b23563f3ed26f1914adf41e3d42de69
GitHub-Pull-Request: golang/go#26865
Reviewed-on: https://go-review.googlesource.com/c/128479
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agonet/http: update bundled SOCKS client
Mikio Hara [Tue, 6 Nov 2018 03:48:17 +0000 (12:48 +0900)]
net/http: update bundled SOCKS client

Updates socks_bundle.go to git rev 26e67e7 for:
26e67e7 internal/socks: fix socket descriptor leakage in Dialer.Dial

Change-Id: I9ab27a85504d77f1ca2e97cb005f5e37fd3c3ff4
Reviewed-on: https://go-review.googlesource.com/c/147717
Run-TryBot: Mikio Hara <mikioh.public.networking@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/compile: loop in disjoint OpOffPtr check
Josh Bleecher Snyder [Sun, 28 Oct 2018 18:19:33 +0000 (11:19 -0700)]
cmd/compile: loop in disjoint OpOffPtr check

We collapse OpOffPtrs during generic rewrites.
However, we also use disjoint at the same time.
Instead of waiting for all OpOffPtrs to be collapsed
before the disjointness rules can kick in,
burrow through all OpOffPtrs immediately.

Change-Id: I60d0a70a9b4605b1817db7c4aab0c0d789651c90
Reviewed-on: https://go-review.googlesource.com/c/145206
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Michael Munday <mike.munday@ibm.com>
Reviewed-by: Keith Randall <khr@golang.org>
6 years agonet: update zoneCache on cache misses to cover appearing interfaces
Michael Stapelberg [Fri, 2 Nov 2018 10:23:54 +0000 (11:23 +0100)]
net: update zoneCache on cache misses to cover appearing interfaces

performance differences are in measurement noise as per benchcmp:

benchmark                            old ns/op     new ns/op     delta
BenchmarkUDP6LinkLocalUnicast-12     5012          5009          -0.06%

Fixes #28535

Change-Id: Id022e2ed089ce8388a2398e755848ec94e77e653
Reviewed-on: https://go-review.googlesource.com/c/146941
Run-TryBot: Mikio Hara <mikioh.public.networking@gmail.com>
Reviewed-by: Mikio Hara <mikioh.public.networking@gmail.com>
6 years agocmd/compile: shrink specialized convT2x call sites
Josh Bleecher Snyder [Mon, 29 Oct 2018 20:54:24 +0000 (13:54 -0700)]
cmd/compile: shrink specialized convT2x call sites

convT2E16 and other specialized type-to-interface routines
accept a type/itab argument and return a complete interface value.
However, we know enough in the routine to do without the type.
And the caller can construct the interface value using the type.

Doing so shrinks the call sites of ten of the specialized convT2x routines.
It also lets us unify the empty and non-empty interface routines.

Cuts 12k off cmd/go.

name                         old time/op  new time/op  delta
ConvT2ESmall-8               2.96ns ± 2%  2.34ns ± 4%  -21.01%  (p=0.000 n=175+189)
ConvT2EUintptr-8             3.00ns ± 3%  2.34ns ± 4%  -22.02%  (p=0.000 n=189+187)
ConvT2ELarge-8               21.3ns ± 7%  21.5ns ± 5%   +1.02%  (p=0.000 n=200+197)
ConvT2ISmall-8               2.99ns ± 4%  2.33ns ± 3%  -21.95%  (p=0.000 n=193+184)
ConvT2IUintptr-8             3.02ns ± 3%  2.33ns ± 3%  -22.82%  (p=0.000 n=198+190)
ConvT2ILarge-8               21.7ns ± 5%  22.2ns ± 4%   +2.31%  (p=0.000 n=199+198)
ConvT2Ezero/zero/16-8        2.96ns ± 2%  2.33ns ± 3%  -21.11%  (p=0.000 n=174+187)
ConvT2Ezero/zero/32-8        2.96ns ± 1%  2.35ns ± 4%  -20.62%  (p=0.000 n=163+193)
ConvT2Ezero/zero/64-8        2.99ns ± 2%  2.34ns ± 4%  -21.78%  (p=0.000 n=183+188)
ConvT2Ezero/zero/str-8       3.27ns ± 3%  2.54ns ± 3%  -22.32%  (p=0.000 n=195+192)
ConvT2Ezero/zero/slice-8     3.46ns ± 4%  2.81ns ± 3%  -18.96%  (p=0.000 n=197+164)
ConvT2Ezero/zero/big-8       88.4ns ±20%  90.0ns ±20%   +1.84%  (p=0.000 n=196+198)
ConvT2Ezero/nonzero/16-8     12.6ns ± 3%  12.3ns ± 3%   -2.34%  (p=0.000 n=167+196)
ConvT2Ezero/nonzero/32-8     12.3ns ± 4%  11.9ns ± 3%   -2.95%  (p=0.000 n=187+193)
ConvT2Ezero/nonzero/64-8     14.2ns ± 6%  13.8ns ± 5%   -2.94%  (p=0.000 n=198+199)
ConvT2Ezero/nonzero/str-8    27.2ns ± 5%  26.8ns ± 5%   -1.33%  (p=0.000 n=200+198)
ConvT2Ezero/nonzero/slice-8  33.3ns ± 8%  33.1ns ± 6%   -0.82%  (p=0.000 n=199+200)
ConvT2Ezero/nonzero/big-8    88.8ns ±22%  90.2ns ±18%   +1.58%  (p=0.000 n=200+199)

Neligible toolspeed impact.

name        old alloc/op      new alloc/op      delta
Template         35.4MB ± 0%       35.3MB ± 0%  -0.06%  (p=0.008 n=5+5)
Unicode          29.1MB ± 0%       29.1MB ± 0%    ~     (p=0.310 n=5+5)
GoTypes           122MB ± 0%        122MB ± 0%  -0.08%  (p=0.008 n=5+5)
Compiler          514MB ± 0%        513MB ± 0%  -0.02%  (p=0.008 n=5+5)
SSA              1.94GB ± 0%       1.94GB ± 0%  -0.01%  (p=0.008 n=5+5)
Flate            24.2MB ± 0%       24.2MB ± 0%    ~     (p=0.548 n=5+5)
GoParser         28.5MB ± 0%       28.5MB ± 0%  -0.05%  (p=0.016 n=5+5)
Reflect          86.3MB ± 0%       86.2MB ± 0%  -0.02%  (p=0.008 n=5+5)
Tar              34.9MB ± 0%       34.9MB ± 0%    ~     (p=0.095 n=5+5)
XML              47.1MB ± 0%       47.1MB ± 0%  -0.05%  (p=0.008 n=5+5)
[Geo mean]       81.0MB            81.0MB       -0.03%

name        old allocs/op     new allocs/op     delta
Template           349k ± 0%         349k ± 0%  -0.08%  (p=0.008 n=5+5)
Unicode            340k ± 0%         340k ± 0%    ~     (p=0.111 n=5+5)
GoTypes           1.28M ± 0%        1.28M ± 0%  -0.09%  (p=0.008 n=5+5)
Compiler          4.92M ± 0%        4.92M ± 0%  -0.08%  (p=0.008 n=5+5)
SSA               15.3M ± 0%        15.3M ± 0%  -0.03%  (p=0.008 n=5+5)
Flate              233k ± 0%         233k ± 0%    ~     (p=0.500 n=5+5)
GoParser           292k ± 0%         292k ± 0%  -0.06%  (p=0.008 n=5+5)
Reflect           1.05M ± 0%        1.05M ± 0%  -0.02%  (p=0.008 n=5+5)
Tar                344k ± 0%         343k ± 0%  -0.06%  (p=0.008 n=5+5)
XML                430k ± 0%         429k ± 0%  -0.08%  (p=0.008 n=5+5)
[Geo mean]         809k              809k       -0.05%

name        old object-bytes  new object-bytes  delta
Template          507kB ± 0%        507kB ± 0%  -0.04%  (p=0.008 n=5+5)
Unicode           225kB ± 0%        225kB ± 0%    ~     (all equal)
GoTypes          1.85MB ± 0%       1.85MB ± 0%  -0.08%  (p=0.008 n=5+5)
Compiler         6.75MB ± 0%       6.75MB ± 0%  +0.01%  (p=0.008 n=5+5)
SSA              21.4MB ± 0%       21.4MB ± 0%  -0.02%  (p=0.008 n=5+5)
Flate             328kB ± 0%        328kB ± 0%  -0.03%  (p=0.008 n=5+5)
GoParser          403kB ± 0%        402kB ± 0%  -0.06%  (p=0.008 n=5+5)
Reflect          1.41MB ± 0%       1.41MB ± 0%  -0.03%  (p=0.008 n=5+5)
Tar               457kB ± 0%        457kB ± 0%  -0.05%  (p=0.008 n=5+5)
XML               601kB ± 0%        600kB ± 0%  -0.16%  (p=0.008 n=5+5)
[Geo mean]       1.05MB            1.04MB       -0.05%

Change-Id: I677a4108c0ecd32617549294036aa84f9214c4fe
Reviewed-on: https://go-review.googlesource.com/c/147360
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Martin Möhrmann <moehrmann@google.com>
6 years agocmd/compile: copy all fields during SubstAny
Josh Bleecher Snyder [Sat, 3 Nov 2018 16:09:28 +0000 (09:09 -0700)]
cmd/compile: copy all fields during SubstAny

Consider these functions:

func f(a any) int
func g(a any) int

Prior to this change, since f and g have identical signatures,
they would share a single generated func type.

types.SubstAny makes a shallow type copy, even after instantiation,
f and g share a single generated Result type.
So if you instantiate f with any=T, call dowidth,
instantiate g with any=U, and call dowidth,
and if sizeof(T) != sizeof(U),
then the Offset of the result for f is now wrong.

I don't believe this happens at all right now, but it bit me hard when
experimenting with some other compiler changes.
And it's hard to debug. It results in rare stack corruption, causing
problems far from the actual source of the problem.

To fix this, change SubstAny to make deep copies of TSTRUCTs.

name        old alloc/op      new alloc/op      delta
Template         35.3MB ± 0%       35.4MB ± 0%  +0.23%  (p=0.008 n=5+5)
Unicode          29.1MB ± 0%       29.1MB ± 0%  +0.16%  (p=0.008 n=5+5)
GoTypes           122MB ± 0%        122MB ± 0%  +0.16%  (p=0.008 n=5+5)
Compiler          513MB ± 0%        514MB ± 0%  +0.19%  (p=0.008 n=5+5)
SSA              1.94GB ± 0%       1.94GB ± 0%  +0.01%  (p=0.008 n=5+5)
Flate            24.2MB ± 0%       24.2MB ± 0%  +0.08%  (p=0.008 n=5+5)
GoParser         28.5MB ± 0%       28.5MB ± 0%  +0.24%  (p=0.008 n=5+5)
Reflect          86.2MB ± 0%       86.3MB ± 0%  +0.09%  (p=0.008 n=5+5)
Tar              34.9MB ± 0%       34.9MB ± 0%  +0.13%  (p=0.008 n=5+5)
XML              47.0MB ± 0%       47.1MB ± 0%  +0.18%  (p=0.008 n=5+5)
[Geo mean]       80.9MB            81.0MB       +0.15%

name        old allocs/op     new allocs/op     delta
Template           348k ± 0%         349k ± 0%  +0.38%  (p=0.008 n=5+5)
Unicode            340k ± 0%         340k ± 0%  +0.21%  (p=0.008 n=5+5)
GoTypes           1.27M ± 0%        1.28M ± 0%  +0.27%  (p=0.008 n=5+5)
Compiler          4.90M ± 0%        4.92M ± 0%  +0.36%  (p=0.008 n=5+5)
SSA               15.3M ± 0%        15.3M ± 0%  +0.03%  (p=0.008 n=5+5)
Flate              232k ± 0%         233k ± 0%  +0.14%  (p=0.008 n=5+5)
GoParser           291k ± 0%         292k ± 0%  +0.42%  (p=0.008 n=5+5)
Reflect           1.05M ± 0%        1.05M ± 0%  +0.14%  (p=0.008 n=5+5)
Tar                343k ± 0%         344k ± 0%  +0.22%  (p=0.008 n=5+5)
XML                428k ± 0%         430k ± 0%  +0.36%  (p=0.008 n=5+5)
[Geo mean]         807k              809k       +0.25%

Change-Id: I62134db642206cded01920dc1d8a7da61f7ca0ac
Reviewed-on: https://go-review.googlesource.com/c/147038
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
6 years agoruntime: clean up MSpan* MCache* MCentral* in docs
Michael Anthony Knyszek [Mon, 5 Nov 2018 19:26:25 +0000 (19:26 +0000)]
runtime: clean up MSpan* MCache* MCentral* in docs

This change cleans up references to MSpan, MCache, and MCentral in the
docs via a bunch of sed invocations to better reflect the Go names for
the equivalent structures (i.e. mspan, mcache, mcentral) and their
methods (i.e. MSpan_Sweep -> mspan.sweep).

Change-Id: Ie911ac975a24bd25200a273086dd835ab78b1711
Reviewed-on: https://go-review.googlesource.com/c/147557
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agoruntime: deflake TestTracebackAncestors
Austin Clements [Mon, 5 Nov 2018 17:36:42 +0000 (12:36 -0500)]
runtime: deflake TestTracebackAncestors

TestTracebackAncestors has a ~0.1% chance of failing with more
goroutines in the traceback than expected. This happens because
there's a window between each goroutine starting its child and that
goroutine actually exiting. The test captures its own stack trace
after everything is "done", but if this happens during that window, it
will include the goroutine that's in the process of being torn down.
Here's an example of such a failure:
https://build.golang.org/log/fad10d0625295eb79fa879f53b8b32b9d0596af8

This CL fixes this by recording the goroutines that are expected to
exit and removing them from the stack trace. With this fix, this test
passed 15,000 times with no failures.

Change-Id: I71e7c6282987a15e8b74188b9c585aa2ca97cbcd
Reviewed-on: https://go-review.googlesource.com/c/147517
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agogo/doc: inspect function signature for building playground examples
Yury Smolsky [Tue, 30 Oct 2018 22:19:35 +0000 (00:19 +0200)]
go/doc: inspect function signature for building playground examples

This documentation example was broken:
https://golang.org/pkg/image/png/#example_Decode.
It did not have the "io" package imported,
The package was referenced in the result type of the function.

The "playExample" function did not inspect
the result types of declared functions.

This CL adds inspecting of parameters and result types of functions.

Fixes #28492
Updates #9679

Change-Id: I6d8b11bad2db8ea8ba69039cfaa914093bdd5132
Reviewed-on: https://go-review.googlesource.com/c/146118
Run-TryBot: Yury Smolsky <yury@smolsky.by>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
6 years agotest: add test that gccgo failed to compile
Ian Lance Taylor [Mon, 5 Nov 2018 18:33:28 +0000 (10:33 -0800)]
test: add test that gccgo failed to compile

Updates #28601

Change-Id: I734fc5ded153126d384f0df912ecd4d208005e49
Reviewed-on: https://go-review.googlesource.com/c/147537
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
6 years agocmd/compile: reintroduce work-around for cyclic alias declarations
Robert Griesemer [Sat, 3 Nov 2018 06:28:26 +0000 (23:28 -0700)]
cmd/compile: reintroduce work-around for cyclic alias declarations

This change re-introduces (temporarily) a work-around for recursive
alias type declarations, originally in https://golang.org/cl/35831/
(intended as fix for #18640). The work-around was removed later
for a more comprehensive cycle detection check. That check
contained a subtle error which made the code appear to work,
while in fact creating incorrect types internally. See #25838
for details.

By re-introducing the original work-around, we eliminate problems
with many simple recursive type declarations involving aliases;
specifically cases such as #27232 and #27267. However, the more
general problem remains.

This CL also fixes the subtle error (incorrect variable use when
analyzing a type cycle) mentioned above and now issues a fatal
error with a reference to the relevant issue (rather than crashing
later during the compilation). While not great, this is better
than the current status. The long-term solution will need to
address these cycles (see #25838).

As a consequence, several old test cases are not accepted anymore
by the compiler since they happened to work accidentally only.
This CL disables parts or all code of those test cases. The issues
are: #18640, #23823, and #24939.

One of the new test cases (fixedbugs/issue27232.go) exposed a
go/types issue. The test case is excluded from the go/types test
suite and an issue was filed (#28576).

Updates #18640.
Updates #23823.
Updates #24939.
Updates #25838.
Updates #28576.

Fixes #27232.
Fixes #27267.

Change-Id: I6c2d10da98bfc6f4f445c755fcaab17fc7b214c5
Reviewed-on: https://go-review.googlesource.com/c/147286
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
6 years agocmd/compile: encapsulate and document two types.Type internal fields
Robert Griesemer [Sat, 3 Nov 2018 04:41:41 +0000 (21:41 -0700)]
cmd/compile: encapsulate and document two types.Type internal fields

Change-Id: I5f7d2155c2c3a47dabdf16fe46b122ede81de4fc
Reviewed-on: https://go-review.googlesource.com/c/147284
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
6 years agoruntime: fix stale comments about mheap and mspan
Michael Anthony Knyszek [Wed, 17 Oct 2018 20:16:45 +0000 (20:16 +0000)]
runtime: fix stale comments about mheap and mspan

As of 07e738e all spans are allocated out of a treap, and not just
large spans or spans for large objects. Also, now we have a separate
treap for spans that have been scavenged.

Change-Id: I9c2cb7b6798fc536bbd34835da2e888224fd7ed4
Reviewed-on: https://go-review.googlesource.com/c/142958
Reviewed-by: Austin Clements <austin@google.com>
6 years agocmd/trace: notes on MMU view improvements
Austin Clements [Tue, 25 Jul 2017 19:03:44 +0000 (15:03 -0400)]
cmd/trace: notes on MMU view improvements

Change-Id: Ib9dcdc76095f6718f1cdc83349503f52567c76d4
Reviewed-on: https://go-review.googlesource.com/c/60801
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
6 years agocmd/trace: display p99.9, p99 and p95 MUT
Austin Clements [Thu, 17 Aug 2017 22:11:01 +0000 (18:11 -0400)]
cmd/trace: display p99.9, p99 and p95 MUT

This uses the mutator utilization distribution to compute the p99.9,
p99, and p95 mutator utilization topograph lines and display them
along with the MMU.

Change-Id: I8c7e0ec326aa4bc00619ec7562854253f01cc802
Reviewed-on: https://go-review.googlesource.com/c/60800
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
6 years agointernal/trace: support for mutator utilization distributions
Austin Clements [Thu, 17 Aug 2017 15:31:03 +0000 (11:31 -0400)]
internal/trace: support for mutator utilization distributions

This adds support for computing the quantiles of a mutator utilization
distribution.

Change-Id: Ia8b3ed14bf415c234e2f567360fd1b361d28bd40
Reviewed-on: https://go-review.googlesource.com/c/60799
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
6 years agocmd/trace: expose MMU analysis flags in web UI
Austin Clements [Fri, 28 Jul 2017 20:30:05 +0000 (16:30 -0400)]
cmd/trace: expose MMU analysis flags in web UI

Change-Id: I672240487172380c9eef61837b41698021aaf834
Reviewed-on: https://go-review.googlesource.com/c/60798
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
6 years agointernal/trace: add "per-P" MMU analysis
Austin Clements [Fri, 28 Jul 2017 20:26:51 +0000 (16:26 -0400)]
internal/trace: add "per-P" MMU analysis

The current MMU analysis considers all Ps together, so if, for
example, one of four Ps is blocked, mutator utilization is 75%.
However, this is less useful for understanding the impact on
individual goroutines because that one blocked goroutine could be
blocked for a very long time, but we still appear to have good
utilization.

Hence, this introduces a new flag that does a "per-P" analysis where
the utilization of each P is considered independently. The MMU is then
the combination of the MMU for each P's utilization function.

Change-Id: Id67b980d4d82b511d28300cdf92ccbb5ae8f0c78
Reviewed-on: https://go-review.googlesource.com/c/60797
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
6 years agointernal/trace: flags for what to include in GC utilization
Austin Clements [Fri, 28 Jul 2017 17:51:58 +0000 (13:51 -0400)]
internal/trace: flags for what to include in GC utilization

Change-Id: I4ba963b003cb25b39d7575d423f17930d84f3f69
Reviewed-on: https://go-review.googlesource.com/c/60796
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
6 years agocmd/trace: list and link to worst mutator utilization windows
Austin Clements [Fri, 28 Jul 2017 14:24:39 +0000 (10:24 -0400)]
cmd/trace: list and link to worst mutator utilization windows

This adds the ability to click a point on the MMU graph to show a list
of the worst 10 mutator utilization windows of the selected size. This
list in turn links to the trace viewer to drill down on specifically
what happened in each specific window.

Change-Id: Ic1b72d8b37fbf2212211c513cf36b34788b30133
Reviewed-on: https://go-review.googlesource.com/c/60795
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agointernal/trace: track worst N mutator utilization windows
Austin Clements [Wed, 26 Jul 2017 16:21:15 +0000 (12:21 -0400)]
internal/trace: track worst N mutator utilization windows

This will let the trace viewer show specifically when poor utilization
happened and link to specific instances in the trace.

Change-Id: I1f03a0f9d9a7570009bb15762e7b8b6f215e9423
Reviewed-on: https://go-review.googlesource.com/c/60793
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
6 years agointernal/trace: use banding to optimize MMU computation
Austin Clements [Tue, 8 Aug 2017 22:15:32 +0000 (18:15 -0400)]
internal/trace: use banding to optimize MMU computation

This further optimizes MMU construction by first computing a
low-resolution summary of the utilization curve. This "band" summary
lets us compute the worst-possible window starting in each of these
low-resolution bands (even without knowing where in the band the
window falls). This in turn lets us compute precise minimum mutator
utilization only in the worst low-resolution bands until we can show
that any remaining bands can't possibly contain a worse window.

This slows down MMU construction for small traces, but these are
reasonably fast to compute either way. For large traces (e.g.,
150,000+ utilization changes) it's significantly faster.

Change-Id: Ie66454e71f3fb06be3f6173b6d91ad75c61bda48
Reviewed-on: https://go-review.googlesource.com/c/60792
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
6 years agointernal/trace: use MU slope to optimize MMU
Austin Clements [Mon, 28 Aug 2017 16:29:07 +0000 (12:29 -0400)]
internal/trace: use MU slope to optimize MMU

This commit speeds up MMU construction by ~10X (and reduces the number
of windows considered by ~20X) by using an observation about the
maximum slope of the windowed mutator utilization function to advance
the window time in jumps if the window's current mean mutator
utilization is much larger than the current minimum.

Change-Id: If3cba5da0c4adc37b568740f940793e491e96a51
Reviewed-on: https://go-review.googlesource.com/c/60791
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
6 years agocmd/trace: add minimum mutator utilization (MMU) plot
Austin Clements [Mon, 24 Jul 2017 15:37:59 +0000 (11:37 -0400)]
cmd/trace: add minimum mutator utilization (MMU) plot

This adds an endpoint to the trace tool that plots the minimum mutator
utilization curve using information on mark assists and GC pauses from
the trace.

This commit implements a fairly straightforward O(nm) algorithm for
computing the MMU (and tests against an even more direct but slower
algorithm). Future commits will extend and optimize this algorithm.

This should be useful for debugging and understanding mutator
utilization issues like #14951, #14812, #18155. #18534, #21107,
particularly once follow-up CLs add trace cross-referencing.

Change-Id: Ic2866869e7da1e6c56ba3e809abbcb2eb9c4923a
Reviewed-on: https://go-review.googlesource.com/c/60790
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
6 years agocmd/addr2line: defer closing objfile
Muhammad Falak R Wani [Mon, 5 Nov 2018 16:34:46 +0000 (22:04 +0530)]
cmd/addr2line: defer closing objfile

Change-Id: I19ff9d231c4cc779b0737802c3c40ee2e00934dd
Reviewed-on: https://go-review.googlesource.com/c/147477
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agoruntime/internal/sys: regenerate zgoos_*.go files
Tobias Klauser [Mon, 5 Nov 2018 11:39:11 +0000 (12:39 +0100)]
runtime/internal/sys: regenerate zgoos_*.go files

zgoos_aix.go is missing GoosJs, the order of GoosAndroid and GoosAix is
mixed up in all files and GoosHurd was added after CL 146023 introduced
GOOS=hurd.

Change-Id: I7e2f5a15645272e9020cfca86e44c364fc072a2b
Reviewed-on: https://go-review.googlesource.com/c/147397
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agoruntime: avoid arm64 8.1 atomics on Android
Elias Naur [Sun, 4 Nov 2018 08:36:25 +0000 (09:36 +0100)]
runtime: avoid arm64 8.1 atomics on Android

The kernel on some Samsung S9+ models reports support for arm64 8.1
atomics, but in reality only some of the cores support them. Go
programs scheduled to cores without support will crash with SIGILL.

This change unconditionally disables the optimization on Android.
A better fix is to precisely detect the offending chipset.

Fixes #28431

Change-Id: I35a1273e5660603824d30ebef2ce7e429241bf1f
Reviewed-on: https://go-review.googlesource.com/c/147377
Run-TryBot: Elias Naur <elias.naur@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
6 years agostrings: lower running time of TestCompareStrings
Keith Randall [Sun, 4 Nov 2018 05:57:52 +0000 (22:57 -0700)]
strings: lower running time of TestCompareStrings

At each comparison, we're making a copy of the whole string.
Instead, use unsafe to share the string backing store with a []byte.

It reduces the test time from ~4sec to ~1sec on my machine
(darwin/amd64).  Some builders were having much more trouble with this
test (>3min), it may help more there.

Fixes #26174
Fixes #28573
Fixes #26155
Update #26473

Change-Id: Id5856fd26faf6ff46e763a088f039230556a4116
Reviewed-on: https://go-review.googlesource.com/c/147358
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agocmd/compile/internal/gc: remove isforw predicate table (cleanup)
Robert Griesemer [Sat, 3 Nov 2018 05:21:23 +0000 (22:21 -0700)]
cmd/compile/internal/gc: remove isforw predicate table (cleanup)

Was only ever filled with one Etype (TFORW) and only used
in one place. Easier to just check for TFORW.

Change-Id: Icc96da3a22b0af1d7e60bc5841c744916c53341e
Reviewed-on: https://go-review.googlesource.com/c/147285
Reviewed-by: Martin Möhrmann <moehrmann@google.com>
6 years agocmd/compile/internal/gc: don't print "internal error" twice
Robert Griesemer [Sat, 3 Nov 2018 16:33:31 +0000 (09:33 -0700)]
cmd/compile/internal/gc: don't print "internal error" twice

The Fatalf mechanism already prints "compiler internal error:"
when reporting an error. There's no need to have "internal error"
in the error message passed to Fatalf calls. Removed them.

Fixes #28575.

Change-Id: I12b1bea37bc839780f257c27ef9e2005bf334925
Reviewed-on: https://go-review.googlesource.com/c/147287
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Martin Möhrmann <moehrmann@google.com>
6 years agocmd/compile: fix fmt_test (fix long test build)
Robert Griesemer [Sat, 3 Nov 2018 04:51:01 +0000 (21:51 -0700)]
cmd/compile: fix fmt_test (fix long test build)

Follow-up on https://golang.org/cl/146782.

TBR=bradfitz

Change-Id: Idaf5488fedfc05d6ff71706fa0bcd70bf98ab25a
Reviewed-on: https://go-review.googlesource.com/c/147283
Reviewed-by: Robert Griesemer <gri@golang.org>
6 years agocmd/internal/obj: don't dedup symbols in WriteObjFile
Austin Clements [Mon, 29 Oct 2018 22:27:51 +0000 (18:27 -0400)]
cmd/internal/obj: don't dedup symbols in WriteObjFile

Currently, WriteObjFile deduplicates symbols by name. This is a
strange and unexpected place to do this. But, worse, there's no
checking that it's reasonable to deduplicate two symbols, so this
makes it incredibly easy to mask errors involving duplicate symbols.
Dealing with duplicate symbols is better left to the linker. We're
also about to introduce multiple symbols with the same name but
different ABIs/versions, which would make this deduplication more
complicated. We just removed the only part of the compiler that
actually depended on this behavior.

This CL removes symbol deduplication from WriteObjFile, since it is no
longer needed.

For #27539.

Change-Id: I650c550e46e83f95c67cb6c6646f9b2f7f10df30
Reviewed-on: https://go-review.googlesource.com/c/146558
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
6 years agocmd/compile: avoid duplicate GC bitmap symbols
Austin Clements [Mon, 29 Oct 2018 22:21:00 +0000 (18:21 -0400)]
cmd/compile: avoid duplicate GC bitmap symbols

Currently, liveness produces a distinct obj.LSym for each GC bitmap
for each function. These are then named by content hash and only
ultimately deduplicated by WriteObjFile.

For various reasons (see next commit), we want to remove this
deduplication behavior from WriteObjFile. Furthermore, it's
inefficient to produce these duplicate symbols in the first place.

GC bitmaps are the only source of duplicate symbols in the compiler.
This commit eliminates these duplicate symbols by declaring them in
the Ctxt symbol hash just like every other obj.LSym. As a result, all
GC bitmaps with the same content now refer to the same obj.LSym.

The next commit will remove deduplication from WriteObjFile.

For #27539.

Change-Id: I4f15e3d99530122cdf473b7a838c69ef5f79db59
Reviewed-on: https://go-review.googlesource.com/c/146557
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
6 years agocmd/internal/obj/arm64: fix encoding of 32-bit negated logical instructions
Cherry Zhang [Fri, 2 Nov 2018 16:03:31 +0000 (12:03 -0400)]
cmd/internal/obj/arm64: fix encoding of 32-bit negated logical instructions

32-bit negated logical instructions (BICW, ORNW, EONW) with
constants were mis-encoded, because they were missing in the
cases where we handle 32-bit logical instructions. This CL
adds the missing cases.

Fixes #28548

Change-Id: I3d6acde7d3b72bb7d3d5d00a9df698a72c806ad5
Reviewed-on: https://go-review.googlesource.com/c/147077
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Ben Shi <powerman1st@163.com>
Reviewed-by: Ben Shi <powerman1st@163.com>
6 years agonet/http: speed up ServeMux matching
Carl Mastrangelo [Fri, 26 Oct 2018 05:03:43 +0000 (22:03 -0700)]
net/http: speed up ServeMux matching

Scanning through all path patterns is not necessary, since the
paths do not change frequently.  Instead, maintain a sorted list
of path prefixes and return the first match.

name            old time/op  new time/op  delta
ServerMatch-12   134ns ± 3%    17ns ± 4%  -86.95%  (p=0.000 n=19+20)

Change-Id: I15b4483dc30db413321435ee6815fc9bf2bcc546
Reviewed-on: https://go-review.googlesource.com/c/144937
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agoall: use "reports whether" consistently in the few places that didn't
Brad Fitzpatrick [Fri, 2 Nov 2018 15:18:43 +0000 (15:18 +0000)]
all: use "reports whether" consistently in the few places that didn't

Go documentation style for boolean funcs is to say:

    // Foo reports whether ...
    func Foo() bool

(rather than "returns true if")

This CL also replaces 4 uses of "iff" with the same "reports whether"
wording, which doesn't lose any meaning, and will prevent people from
sending typo fixes when they don't realize it's "if and only if". In
the past I think we've had the typo CLs updated to just say "reports
whether". So do them all at once.

(Inspired by the addition of another "returns true if" in CL 146938
in fd_plan9.go)

Created with:

$ perl -i -npe 's/returns true if/reports whether/' $(git grep -l "returns true iff" | grep -v vendor)
$ perl -i -npe 's/returns true if/reports whether/' $(git grep -l "returns true if" | grep -v vendor)

Change-Id: Ided502237f5ab0d25cb625dbab12529c361a8b9f
Reviewed-on: https://go-review.googlesource.com/c/147037
Reviewed-by: Ian Lance Taylor <iant@golang.org>