]> Cypherpunks repositories - gostls13.git/log
gostls13.git
9 years agocmd/asm: delete Addr.Reg2
Josh Bleecher Snyder [Fri, 29 May 2015 17:33:39 +0000 (10:33 -0700)]
cmd/asm: delete Addr.Reg2

That which cannot happen has not happened.

No immediate changes to Addr or Prog size.

Change-Id: I4cb9315f2c9f5f92eda340bfc4abb46395fa467f
Reviewed-on: https://go-review.googlesource.com/10513
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/internal/obj: shrink Prog, remove two unused fields
Brad Fitzpatrick [Fri, 29 May 2015 16:51:04 +0000 (09:51 -0700)]
cmd/internal/obj: shrink Prog, remove two unused fields

Printed and Width were unused.

Despite only removing two bytes, due to alignment, 8 bytes are saved
on 64-bit:

Before: unsafe.Sizeof(obj.Prog{}) == 304
 After: unsafe.Sizeof(obj.Prog{}) == 296

The next size class below 320 (304=>19(320)) is 288. Still 8 bytes
away from that.

Change-Id: I8d1632dd40d387f7036c03c65ea4d64e9b6218c3
Reviewed-on: https://go-review.googlesource.com/10511
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agocmd/internal/obj: fix build breakage from making From3 a pointer
Brad Fitzpatrick [Fri, 29 May 2015 17:00:45 +0000 (10:00 -0700)]
cmd/internal/obj: fix build breakage from making From3 a pointer

Change-Id: I55a7f455ebbd6b1bd6912aae82c0fcff6f43387c
Reviewed-on: https://go-review.googlesource.com/10512
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/old5a, cmd/old8a, cmd/old9a: rebuild y.go
Russ Cox [Fri, 29 May 2015 17:35:31 +0000 (13:35 -0400)]
cmd/old5a, cmd/old8a, cmd/old9a: rebuild y.go

Missed in previous CL, causing build failures.

Change-Id: I60aae5a3139aa009cb643d871d91b9d4c47dcbb8
Reviewed-on: https://go-review.googlesource.com/10538
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agomath/big: remove (*Float).Scan, ScanFloat; more robust (*Float).Parse
Robert Griesemer [Fri, 29 May 2015 00:38:05 +0000 (17:38 -0700)]
math/big: remove (*Float).Scan, ScanFloat; more robust (*Float).Parse

- (*Float).Scan conflicted with fmt.Scanner.Scan; it was also only used
  internally. Removed it, as well as the companion ScanFloat function.

- (*Float).Parse (and thus ParseFloat) can now also parse infinities.
  As a result, more code could be simplified.

- Fixed a bug in rounding (round may implicitly be called for infinite
  values). Found via existing test cases, after simplifying some code.

- Added more test cases.

Fixes issue #10938.

Change-Id: I1df97821654f034965ba8b82b272e52e6dc427f1
Reviewed-on: https://go-review.googlesource.com/10498
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agomath/big: rename (*Float).Format to (*Float).Text
Robert Griesemer [Thu, 28 May 2015 22:25:06 +0000 (15:25 -0700)]
math/big: rename (*Float).Format to (*Float).Text

This paves the way for a fmt-compatible (*Float).Format method.
A better name then Text is still desirable (suggestions welcome).

This is partly fixing issue #10938.

Change-Id: I59c20a8cee11f5dba059fe0f38b414fe75f2ab13
Reviewed-on: https://go-review.googlesource.com/10493
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agocmd/internal/obj: make Prog.From3 a pointer
Russ Cox [Wed, 27 May 2015 19:01:44 +0000 (15:01 -0400)]
cmd/internal/obj: make Prog.From3 a pointer

It is almost never set and Addr is large, so having the full struct
in the Prog wastes memory most of the time.

Before (on a 64-bit system):

$ sizeof -p cmd/internal/obj Addr Prog
Addr 80
Prog 376
$

After:

$ sizeof -p cmd/internal/obj Addr Prog
Addr 80
Prog 304
$

Change-Id: I491f201241f87543964a7d0f48b85830759be9d0
Reviewed-on: https://go-review.googlesource.com/10457
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agoio: minor improvements to doc comment on WriteString.
David Symonds [Fri, 29 May 2015 04:29:57 +0000 (14:29 +1000)]
io: minor improvements to doc comment on WriteString.

Change-Id: Iddcd0cfb8f2c2f1c4ad7a94b50a9f65b543862c4
Reviewed-on: https://go-review.googlesource.com/10473
Reviewed-by: Minux Ma <minux@golang.org>
9 years agoarchive/tar: terminate when reading malformed sparse files
Håvard Haugen [Thu, 28 May 2015 11:48:47 +0000 (13:48 +0200)]
archive/tar: terminate when reading malformed sparse files

Fixes #10968.

Change-Id: I027bc571a71629ac49c2a0ff101b2950af6e7531
Reviewed-on: https://go-review.googlesource.com/10482
Reviewed-by: David Symonds <dsymonds@golang.org>
Run-TryBot: David Symonds <dsymonds@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/dist: only test packages and commands that have tests
Andrew Gerrand [Thu, 28 May 2015 22:14:58 +0000 (15:14 -0700)]
cmd/dist: only test packages and commands that have tests

Change-Id: I7aeb9fef3739c17c03fdaadbe00cd945ec9c0d72
Reviewed-on: https://go-review.googlesource.com/10492
Run-TryBot: Andrew Gerrand <adg@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agogo/types: fix error message for embedded non-interface types in interfaces
Robert Griesemer [Thu, 28 May 2015 18:58:42 +0000 (11:58 -0700)]
go/types: fix error message for embedded non-interface types in interfaces

Fixes #10979.

Change-Id: Iac25645ba8181a56a75ddfcd29ff6d64c15c4f57
Reviewed-on: https://go-review.googlesource.com/10466
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agotest: re-enable rotate tests
Josh Bleecher Snyder [Mon, 18 May 2015 18:09:29 +0000 (11:09 -0700)]
test: re-enable rotate tests

Memory usage has been reduced.
The tests are still slow,
but that is issue #10571.

/usr/bin/time shows significant variation
in the peak memory usage compiling with tip.
This is unsurprising, given GC.

Using Go 1.4.2, memory is stable at 410mb.
Using tip at d2ee09298,
memory ranges from 470mb (+15%) to 534mb (+30%),
with a mean of 504mb (+23%), with n=50.

Fixes #9933.

Change-Id: Id31f3ae086ec324abf70e8f1a8044c4a0c27e274
Reviewed-on: https://go-review.googlesource.com/10211
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/compile: reject p-notation floats in Go source files
David Chase [Wed, 27 May 2015 21:33:46 +0000 (17:33 -0400)]
cmd/compile: reject p-notation floats in Go source files

Use pkgimport == nil (or not) to distinguish between
parsing .go source files where "p" exponent specifier
is not allowed and parsing .a or .o export data where
it is.  Use that to control error when p-exponent is
seen.

Fixes #9036

Change-Id: I8924f09c91d4945ef3f20e80a6e544008a94a7e4
Reviewed-on: https://go-review.googlesource.com/10450
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/internal/gc: unembed Param field
Josh Bleecher Snyder [Tue, 19 May 2015 22:25:35 +0000 (15:25 -0700)]
cmd/internal/gc: unembed Param field

This is an automated follow-up to CL 10210.
It was generated with a combination of eg and gofmt -r.

No functional changes. Passes toolstash -cmp.

Change-Id: I35f5897948a270b472d8cf80612071b4b29e9a2b
Reviewed-on: https://go-review.googlesource.com/10253
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoruntime: disable GC around TestGoroutineParallelism
Austin Clements [Tue, 26 May 2015 18:32:24 +0000 (14:32 -0400)]
runtime: disable GC around TestGoroutineParallelism

TestGoroutineParallelism can deadlock if the GC runs during the
test. Currently it tries to prevent this by forcing a GC before the
test, but this is best effort and fails completely if GOGC is very low
for testing.

This change replaces this best-effort fix with simply setting GOGC to
off for the duration of the test.

Change-Id: I8229310833f241b149ebcd32845870c1cb14e9f8
Reviewed-on: https://go-review.googlesource.com/10454
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/compile: avoid temporary in race mode with slice and append
Austin Clements [Wed, 27 May 2015 19:20:49 +0000 (15:20 -0400)]
cmd/compile: avoid temporary in race mode with slice and append

Currently when the race detector is enabled, orderexpr always creates
a temporary for slice and append operations. This used to be necessary
because the race detector had a different code path for slice
assignment that required this temporary. Unfortunately, creating this
temporary inhibits the optimization that eliminates write barriers
when a slice is assigned only to change its length or cap. For most
code, this is bad for performance, and in go:nowritebarrier functions
in the runtime, this can mean the difference between compiling and not
compiling.

Now the race detector uses the regular slice assignment code, so
creating this temporary is no longer necessary.

Change-Id: I296042e1edc571b77c407f709c2ff9091c4aa795
Reviewed-on: https://go-review.googlesource.com/10456
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoruntime: use stripped test environment for TestGdbPython
Austin Clements [Tue, 26 May 2015 19:21:18 +0000 (15:21 -0400)]
runtime: use stripped test environment for TestGdbPython

Most runtime tests that invoke the compiler to build a sub-test binary
do so with a special environment constructed by testEnv that strips
out environment variables that should apply to the test but not to the
build.

Fix TestGdbPython to use this test environment when invoking go build,
like other tests do.

Change-Id: Iafdf89d4765c587cbebc427a5d61cb8a7e71b326
Reviewed-on: https://go-review.googlesource.com/10455
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agomath/big: fix latent decimal conversion bug
Robert Griesemer [Wed, 27 May 2015 21:43:30 +0000 (14:43 -0700)]
math/big: fix latent decimal conversion bug

A decimal represented 0.0 with a 0-length mantissa and undefined
exponent, but the formatting code assumes a valid zero exponent
if the float value is 0.0. The code worked because we allocate a
new decimal value each time and because there's no rounding that
lead to 0.0.

Change-Id: Ifd771d7709de83b87fdbf141786286b4c3e13d4f
Reviewed-on: https://go-review.googlesource.com/10448
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agocmd/link/internal/ld: Skip combining dwarf for darwin/arm.
Ryan Brown [Wed, 8 Apr 2015 19:55:34 +0000 (12:55 -0700)]
cmd/link/internal/ld: Skip combining dwarf for darwin/arm.

Change-Id: I3a6df0a76d57db7cb6910f4179a6ce380f219a37
Reviewed-on: https://go-review.googlesource.com/10442
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agotest: remove arch char from nosplit
Josh Bleecher Snyder [Wed, 27 May 2015 19:33:43 +0000 (12:33 -0700)]
test: remove arch char from nosplit

This is dead code that was missed
during the 'go tool compile' migration.

Change-Id: Ice2af8a9ef72f8fd5f82225ee261854d93b659f1
Reviewed-on: https://go-review.googlesource.com/10430
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/internal/obj: make arm64 use RegTo2 instead of a full fledged Addr To2
Shenghou Ma [Thu, 21 May 2015 21:51:34 +0000 (17:51 -0400)]
cmd/internal/obj: make arm64 use RegTo2 instead of a full fledged Addr To2

It shrinks Prog type from 448 bytes down to 376 bytes on amd64.

It also makes sense, because I don't know of any modern architecture
that have instructions which can write to two destinations, none of
which is a register (even x86 doesn't have such instructions).

Change-Id: I3061f1c9ac93d79ee2b92ecb9049641d0e0f6300
Reviewed-on: https://go-review.googlesource.com/10330
Reviewed-by: Aram Hăvărneanu <aram@mgk.ro>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/dist: don't run go list when running a specific test
Brad Fitzpatrick [Wed, 27 May 2015 23:33:03 +0000 (16:33 -0700)]
cmd/dist: don't run go list when running a specific test

This speeds up sharded builds notably, by 1 second * the number of
tests.

Change-Id: Ib0295c31e4974f3003f72cb16c48949812b6f22b
Reviewed-on: https://go-review.googlesource.com/10460
Reviewed-by: Andrew Gerrand <adg@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agomath/big: more cleanups (msbxx, nlzxx functions)
Robert Griesemer [Tue, 26 May 2015 23:42:24 +0000 (16:42 -0700)]
math/big: more cleanups (msbxx, nlzxx functions)

Change-Id: Ibace718452b6dc029c5af5240117f5fc794c38cf
Reviewed-on: https://go-review.googlesource.com/10388
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agomath/big: removed TODO, cleanups
Robert Griesemer [Tue, 26 May 2015 21:57:05 +0000 (14:57 -0700)]
math/big: removed TODO, cleanups

- factor out handling of sign
- rename bstring, pstring to fmtB, fmtP consistent with fmtE, fmtF
- move all float-to-string conversion functions into ftoa.go
- no functional changes

Change-Id: I5970ecb874dc9c387630b59147d90bda16a5d8e6
Reviewed-on: https://go-review.googlesource.com/10387
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agostrconv: minor internal comment fix
Robert Griesemer [Wed, 27 May 2015 19:08:38 +0000 (12:08 -0700)]
strconv: minor internal comment fix

Change-Id: I590ac9e976d4044d1f4f280137ea9b38851a9fc2
Reviewed-on: https://go-review.googlesource.com/10424
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agoarchive/zip: sanity check the TOC's declared number of files
Brad Fitzpatrick [Wed, 27 May 2015 17:49:25 +0000 (10:49 -0700)]
archive/zip: sanity check the TOC's declared number of files

Fixes #10956

Change-Id: If8517094f04250c4f722e1e899a237eb6e170eb9
Reviewed-on: https://go-review.googlesource.com/10421
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agocmd/compile: propagate correct line numbers in treecopy
David Chase [Sat, 23 May 2015 02:01:01 +0000 (22:01 -0400)]
cmd/compile: propagate correct line numbers in treecopy

Added a lineno parameter to treecopy and listtreecopy
(ignored if = 0).  When nodes are copied the copy is
assigned the non-zero lineno (normally this would be
the destination).

Fixes #8183

Change-Id: Iffb767a745093fb89aa08bf8a7692c2f0122be98
Reviewed-on: https://go-review.googlesource.com/10334
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agodoc: update go1.5.txt
Brad Fitzpatrick [Wed, 27 May 2015 16:56:28 +0000 (09:56 -0700)]
doc: update go1.5.txt

Change-Id: I48b5f10d703dba48ec8e67c58d4276befafb5524
Reviewed-on: https://go-review.googlesource.com/10420
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/internal/ld: do not depend on local symbols to read a type's gcdata
Michael Hudson-Doyle [Mon, 25 May 2015 04:13:50 +0000 (16:13 +1200)]
cmd/internal/ld: do not depend on local symbols to read a type's gcdata

We already read the address of a gcmask/gcprog out of the type data, but I
didn't know how many bytes to read. But it turns out that it's easy to
calculate, so change to do that. This means that we no longer depend on the
local symbols being present, allowing me to strip the shared libraries for
distribution and make them a lot smaller.

As a bonus, this makes LSym another 24 bytes smaller, down to 296 bytes now.

Change-Id: I379d359e28d63afae6753efd23efdf1fbb716992
Reviewed-on: https://go-review.googlesource.com/10377
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/internal/ld: store the libraries a shared library was linked against in a note
Michael Hudson-Doyle [Mon, 25 May 2015 02:51:02 +0000 (14:51 +1200)]
cmd/internal/ld: store the libraries a shared library was linked against in a note

The motivation for this is the innocuous looking test case that is added. This
creates a stack exe -> libdep2.so -> libdep.so -> libruntime.so. The problem
comes from the fact that a function from libdep.so gets inlined all the way
into exe. This (unsurprisingly) means that the object file for exe references
symbols from libdep.so, which means that -ldep needs to be passed when linking
exe and it isn't. The fix is simply to pass it -- there is no harm in passing
it when it's not needed.

The thing is, it's not clear at all in the current code to see how the linker
can know that libdep2 is linked against libdep. It could look through the
DT_NEEDED entries in libdep2 and try to guess which are Go libraries, but it
feels better to be explicit. So this adds another SHT_NOTE section that lists
the shared libraries a shared library was linked against, and makes sure the
complete set of depended upon shared libraries is passed to the external
linker.

Change-Id: I79aa6f98b4db4721d657a7eb7b7f062269bf49e2
Reviewed-on: https://go-review.googlesource.com/10376
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoarchive/tar: don't panic on negative file size
Håvard Haugen [Wed, 27 May 2015 08:44:44 +0000 (10:44 +0200)]
archive/tar: don't panic on negative file size

Fixes #10959.
Fixes #10960.

Change-Id: I9a81a0e2b8275338d0d1c3f7f7265e0fd91f3de2
Reviewed-on: https://go-review.googlesource.com/10402
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Symonds <dsymonds@golang.org>
9 years agocmd/link/internal/ld: put abi hash into a note
Michael Hudson-Doyle [Mon, 25 May 2015 01:59:08 +0000 (13:59 +1200)]
cmd/link/internal/ld: put abi hash into a note

This makes for a more stable API for tools (including cmd/link itself) to
extract the abi hash from a shared library and makes it possible at all for a
library that has had the local symbol table removed.

The existing note-writing code only supports writing notes into the very start
of the object file so they are easy to find in core dumps. This doesn't apply
to the "go" notes and means that all notes have to fit into a fixed size
budget. That's annoying now we have more notes (and the next CL will add
another one) so this does a little bit of work to make adding notes that do not
have to go at the start of the file easier and moves the writing of the package
list note over to that mechanism, which lets me revert a hack that increased
the size budget mentioned above for -buildmode=shared builds.

Change-Id: I6077a68d395c8a2bc43dec8506e73c71ef77d9b9
Reviewed-on: https://go-review.googlesource.com/10375
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agomisc/cgo/testshared: do not capture output of go commands in verbose mode
Michael Hudson-Doyle [Tue, 26 May 2015 22:30:55 +0000 (10:30 +1200)]
misc/cgo/testshared: do not capture output of go commands in verbose mode

Change-Id: I8694ee5e5642c31815ae63cd414a3b1fcd9c95b0
Reviewed-on: https://go-review.googlesource.com/10411
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/link: replace interface{} fields with concrete types
Michael Hudson-Doyle [Wed, 27 May 2015 00:04:25 +0000 (12:04 +1200)]
cmd/link: replace interface{} fields with concrete types

The LSym.Section and Section.Elfsect fields were defined as interface{} but
always had the same concrete type (*Section and *ElfShdr respectively) so just
define them with that type. Reduces size of LSym from 328 to 320 bytes and
reduces best-of-10 maxresident size from 246028k to 238036k when linking
libstd.so.

Change-Id: Ie7112c53e4c2c7ce5fe233b81372aa5633f572e8
Reviewed-on: https://go-review.googlesource.com/10410
Reviewed-by: Minux Ma <minux@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/internal/objfile: Skip mach-o debug symbols.
Ryan Brown [Wed, 8 Apr 2015 19:55:34 +0000 (12:55 -0700)]
cmd/internal/objfile: Skip mach-o debug symbols.

This allows objdump to disassemble gcc generated binaries on OS X 10.6.

Change-Id: I1a5bfbf7c252e78215ef1f122520689d5ce6ddca
Reviewed-on: https://go-review.googlesource.com/10383
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoarchive/zip: verify number of File bytes read at EOF
Brad Fitzpatrick [Tue, 26 May 2015 20:49:55 +0000 (13:49 -0700)]
archive/zip: verify number of File bytes read at EOF

Fixes #10957

Change-Id: I75fe25133dfcebd1682a8058b1c354ec894cc997
Reviewed-on: https://go-review.googlesource.com/10384
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Andrew Gerrand <adg@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/link/internal/amd64: -buildmode=c-archive forces external link mode
Ian Lance Taylor [Tue, 26 May 2015 20:49:59 +0000 (13:49 -0700)]
cmd/link/internal/amd64: -buildmode=c-archive forces external link mode

At some point this code should probably move to cmd/link/internal/ld,
but at least for now just handle c-archive like c-shared.

Change-Id: Ic17656529cb0fe189a37f15e670350ab13bb5276
Reviewed-on: https://go-review.googlesource.com/10385
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agospec: removed TODOs (invisible html comment) in favor of issues
Robert Griesemer [Tue, 26 May 2015 18:44:18 +0000 (11:44 -0700)]
spec: removed TODOs (invisible html comment) in favor of issues

- no "visible" change to spec but for updated date
- retired several outdated TODO items
- filed non-urgent issues 10953, 10954, 10955 for current TODOs

Change-Id: If87ad0fb546c6955a6d4b5801e06e5c7d5695ea2
Reviewed-on: https://go-review.googlesource.com/10382
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agonet/mail: add AddressParser type
Alexandre Cesaro [Mon, 25 May 2015 16:58:19 +0000 (18:58 +0200)]
net/mail: add AddressParser type

Add the AddressParser type to allow decoding any charset in
mail addresses.

Fixes #7079

Change-Id: Ic34efb3e3d804a4e17149a6c38cfd73f5f275ab7
Reviewed-on: https://go-review.googlesource.com/10392
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/compile: don't cater to race detector in write barriers
Russ Cox [Thu, 21 May 2015 19:00:06 +0000 (15:00 -0400)]
cmd/compile: don't cater to race detector in write barriers

The new lower-level barriers work fine and don't need special handling,
because they appear to the race detector as (visible) ordinary assignments.

Change-Id: I7477d73a3deecbebf68716580678c595cc4151e3
Reviewed-on: https://go-review.googlesource.com/10316
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/link/internal/ld: if -v, display external linker output
Ian Lance Taylor [Sun, 24 May 2015 18:29:38 +0000 (11:29 -0700)]
cmd/link/internal/ld: if -v, display external linker output

It can be useful when debugging to be able to see what the external
linker is doing even when it succeeds.  In particular this permits
passing -v to the external linker to see precisely what it is doing.

Change-Id: Ifed441912d97bbebea20303fdb899e140b380215
Reviewed-on: https://go-review.googlesource.com/10363
Reviewed-by: Minux Ma <minux@golang.org>
9 years agocmd/go: permit C files if the package uses SWIG
Ian Lance Taylor [Sat, 16 May 2015 00:12:20 +0000 (17:12 -0700)]
cmd/go: permit C files if the package uses SWIG

They will be compiled and added to the archive, just as though the
package used cgo.  In effect all SWIG packages now use cgo anyhow.

Change-Id: I5d5a28ed0ec4295f24036b2834218bc980f080d0
Reviewed-on: https://go-review.googlesource.com/10146
Reviewed-by: Minux Ma <minux@golang.org>
9 years agoandroidtest.bash: clean up stale GOROOT
David Crawshaw [Mon, 25 May 2015 11:17:27 +0000 (07:17 -0400)]
androidtest.bash: clean up stale GOROOT

Fixes #10806

Change-Id: I1be1f28ad60c913105d8417c42ec1b262f101f72
Reviewed-on: https://go-review.googlesource.com/10391
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agomisc/android: cleaner to remove stale GOROOT files
David Crawshaw [Mon, 25 May 2015 11:20:07 +0000 (07:20 -0400)]
misc/android: cleaner to remove stale GOROOT files

Updates #10806

Change-Id: I734d6db026cc7c2e3099a76dc8db8e42b2b90aa7
Reviewed-on: https://go-review.googlesource.com/10390
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agoruntime: don't always block all signals on OpenBSD
Elias Naur [Sat, 23 May 2015 09:26:22 +0000 (11:26 +0200)]
runtime: don't always block all signals on OpenBSD

Implement the changes from CL 10173 on OpenBSD.

Change-Id: I2db1cd8141fd392a34753a1b8113e2e0401173b9
Reviewed-on: https://go-review.googlesource.com/10342
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agonet: adjust dual stack support on dragonfly
Mikio Hara [Thu, 14 May 2015 01:18:10 +0000 (10:18 +0900)]
net: adjust dual stack support on dragonfly

As mentioned in
http://gitweb.dragonflybsd.org/dragonfly.git/commit/727ccde8cce813911d885b7f6ed749dcea68a886,
DragonFly BSD is dropping support for IPv6 IPv4-mapped address.
Unfortunately, on some released versions we see the kernels pretend to
support the feature but actually not (unless tweaking some kernel states
via sysctl.)

To avoid unpredictable behavior, the net package assumes that all
DragonFly BSD kernels don't support IPv6 IPv4-mapped address.

Fixes #10764.

Change-Id: Ic7af3651e0372ec03774432fbb6b2eb0c455e994
Reviewed-on: https://go-review.googlesource.com/10071
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agonet: don't show verbose information when -test.v=false
Mikio Hara [Fri, 15 May 2015 03:10:10 +0000 (12:10 +0900)]
net: don't show verbose information when -test.v=false

Updates #10845.

Change-Id: I4cec670c7db88c50a6e5619e611744e161d73b3c
Reviewed-on: https://go-review.googlesource.com/10131
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agomath/big: Always print exponent sign when using 'p' exponent for Floats.
Robert Griesemer [Fri, 22 May 2015 20:58:03 +0000 (13:58 -0700)]
math/big: Always print exponent sign when using 'p' exponent for Floats.

Float.Format supports the 'b' and 'p' format, both of which print
a binary ('p') exponent. The 'b' format always printed a sign ('+'
or '-') for the exponent; the 'p' format only printed a negative
sign for the exponent. This change makes the two consistent. It
also makes the 'p' format easier to read if the exponent is >= 0.

Also:
- Comments added elsewhere.

Change-Id: Ifd2e01bdafb3043345972ca22a90248d055bd29b
Reviewed-on: https://go-review.googlesource.com/10359
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agomisc/cgo/test: fix build for CC=clang
Elias Naur [Fri, 22 May 2015 22:46:10 +0000 (00:46 +0200)]
misc/cgo/test: fix build for CC=clang

Fix build error when CL=clang introduced by CL 10173.

Change-Id: I8edf210787a9803280c0779ff710c7e634a820d6
Reviewed-on: https://go-review.googlesource.com/10341
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/compile/internal/big: update and apply vendor.bash
Robert Griesemer [Fri, 22 May 2015 22:21:56 +0000 (15:21 -0700)]
cmd/compile/internal/big: update and apply vendor.bash

Package-external tests must use the vendored math/big package, not
the original one, otherwise tests may fail if there are discrepancies
in the implementation.

Change-Id: Ic5f0489aa6420ffea1f488633453f871ce1f0f66
Reviewed-on: https://go-review.googlesource.com/10380
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/compile/internal/gc: simplify mpgetflt (remove switch and indirection)
Robert Griesemer [Fri, 22 May 2015 18:31:47 +0000 (11:31 -0700)]
cmd/compile/internal/gc: simplify mpgetflt (remove switch and indirection)

Change-Id: I6ae3534defdae9367e1b856dbb8e846c3263a758
Reviewed-on: https://go-review.googlesource.com/10358
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agocmd/compile/internal/gc: correctly use Float32 in mpgetfltN
Robert Griesemer [Fri, 22 May 2015 01:23:51 +0000 (18:23 -0700)]
cmd/compile/internal/gc: correctly use Float32 in mpgetfltN

This resolves the compiler part of issue #10321.

Change-Id: I44b9909f992b37dd34b1c5292decd12de3d3a65e
Reviewed-on: https://go-review.googlesource.com/10355
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agogo/build: make ArchChar always return "?" to match docs
Shenghou Ma [Fri, 22 May 2015 21:09:36 +0000 (17:09 -0400)]
go/build: make ArchChar always return "?" to match docs

Change-Id: I56f825f81aead9ded7af07a02188a52d3650ccf8
Reviewed-on: https://go-review.googlesource.com/10333
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/compile/internal/big: update to latest version (run sh vendor.bash)
Robert Griesemer [Fri, 22 May 2015 01:13:44 +0000 (18:13 -0700)]
cmd/compile/internal/big: update to latest version (run sh vendor.bash)

No manual code changes.

This will permit addressing the compiler aspect of issue #10321 in a
subsequent change.

Change-Id: I3376dc38cafa0ec98bf54de33293015d0183cc82
Reviewed-on: https://go-review.googlesource.com/10354
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agomath/big: fix Float.Float64 conversion for denormal corner cases
Robert Griesemer [Fri, 22 May 2015 01:00:54 +0000 (18:00 -0700)]
math/big: fix Float.Float64 conversion for denormal corner cases

- This change uses the same code as for Float32 and fixes the case
  of a number that gets rounded up to the smallest denormal.

- Enabled correspoding test case.

Change-Id: I8aac874a566cd727863a82717854f603fbdc26c6
Reviewed-on: https://go-review.googlesource.com/10352
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agomath/big: add more Float.Float64 conversion tests
Robert Griesemer [Fri, 22 May 2015 00:52:49 +0000 (17:52 -0700)]
math/big: add more Float.Float64 conversion tests

- structure the Float64 conversion tests the same way as for Float32
- add additional test cases, including one that exposes a current issue
  (currently disabled, same issue as was fixed for Float32)

The Float64 fix will be in a subsequent change for easier reviewing.

Change-Id: I95dc9e8d1f6b6073a98c7bc2289e6d3248fc3420
Reviewed-on: https://go-review.googlesource.com/10351
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agomath/big: fix Float.Float32 conversion for denormal corner cases
Robert Griesemer [Fri, 22 May 2015 00:00:37 +0000 (17:00 -0700)]
math/big: fix Float.Float32 conversion for denormal corner cases

The existing code was incorrect for numbers that after rounding would
become the smallest denormal float32 (instead the result was 0). This
caused all.bash to fail if Float32() were used in the compiler for
constant arithmetic (there's currently a work-around - see also issue
10321.

This change fixes the implementation of Float.Float32 and adds
corresponding test cases. Float32 and Float64 diverge at this point.
For ease of review, this change only fixes Float32. Float64 will be
made to match in a subsequent change.

Fixes #10321.

Change-Id: Iccafe37c1593a4946bc552e4ad2045f69be62d80
Reviewed-on: https://go-review.googlesource.com/10350
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agoruntime: don't always unblock all signals
Elias Naur [Mon, 18 May 2015 09:00:24 +0000 (11:00 +0200)]
runtime: don't always unblock all signals

Ian proposed an improved way of handling signals masks in Go, motivated
by a problem where the Android java runtime expects certain signals to
be blocked for all JVM threads. Discussion here

https://groups.google.com/forum/#!topic/golang-dev/_TSCkQHJt6g

Ian's text is used in the following:

A Go program always needs to have the synchronous signals enabled.
These are the signals for which _SigPanic is set in sigtable, namely
SIGSEGV, SIGBUS, SIGFPE.

A Go program that uses the os/signal package, and calls signal.Notify,
needs to have at least one thread which is not blocking that signal,
but it doesn't matter much which one.

Unix programs do not change signal mask across execve.  They inherit
signal masks across fork.  The shell uses this fact to some extent;
for example, the job control signals (SIGTTIN, SIGTTOU, SIGTSTP) are
blocked for commands run due to backquote quoting or $().

Our current position on signal masks was not thought out.  We wandered
into step by step, e.g., http://golang.org/cl/7323067 .

This CL does the following:

Introduce a new platform hook, msigsave, that saves the signal mask of
the current thread to m.sigsave.

Call msigsave from needm and newm.

In minit grab set up the signal mask from m.sigsave and unblock the
essential synchronous signals, and SIGILL, SIGTRAP, SIGPROF, SIGSTKFLT
(for systems that have it).

In unminit, restore the signal mask from m.sigsave.

The first time that os/signal.Notify is called, start a new thread whose
only purpose is to update its signal mask to make sure signals for
signal.Notify are unblocked on at least one thread.

The effect on Go programs will be that if they are invoked with some
non-synchronous signals blocked, those signals will normally be
ignored.  Previously, those signals would mostly be ignored.  A change
in behaviour will occur for programs started with any of these signals
blocked, if they receive the signal: SIGHUP, SIGINT, SIGQUIT, SIGABRT,
SIGTERM.  Previously those signals would always cause a crash (unless
using the os/signal package); with this change, they will be ignored
if the program is started with the signal blocked (and does not use
the os/signal package).

./all.bash completes successfully on linux/amd64.

OpenBSD is missing the implementation.

Change-Id: I188098ba7eb85eae4c14861269cc466f2aa40e8c
Reviewed-on: https://go-review.googlesource.com/10173
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agonet: fix panic in TestDialerDualStack
Dave Cheney [Fri, 22 May 2015 02:55:47 +0000 (12:55 +1000)]
net: fix panic in TestDialerDualStack

This change ensures that the test does not try to close dual stack
listeners which have not yet been opened.

Spotted in crash here
http://build.golang.org/log/e5843777df400868ce708b7f00c50dc32c2ec478

Change-Id: I79d513e166effb3e018e2b9dfc23751d92fcbe4b
Reviewed-on: https://go-review.googlesource.com/10371
Reviewed-by: Mikio Hara <mikioh.mikioh@gmail.com>
Run-TryBot: Dave Cheney <dave@cheney.net>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/internal/gc: move check for large-hence-heap-allocated types into escape analysis
David Chase [Wed, 20 May 2015 19:16:34 +0000 (15:16 -0400)]
cmd/internal/gc: move check for large-hence-heap-allocated types into escape analysis

Before this change, the check for too-large arrays (and other large
types) occurred after escape analysis.  If the data moved off stack
and onto the heap contained any pointers, it would therefore escape,
but because the too-large check occurred after escape analysis this
would not be recorded and a stack pointer would leak to the heap
(see the modified escape_array.go for an example).

Some of these appear to remain, in calls to typecheck from within walk.

Also corrected a few comments in escape_array.go about "BAD"
analysis that is now done correctly.

Enhanced to move aditional EscNone-but-large-so-heap checks into esc.c.

Change-Id: I770c111baff28a9ed5f8beb601cf09dacc561b83
Reviewed-on: https://go-review.googlesource.com/10268
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/internal/gc: make indirect calls properly escape-y
David Chase [Thu, 21 May 2015 16:40:25 +0000 (12:40 -0400)]
cmd/internal/gc: make indirect calls properly escape-y

Indirect function and method calls should leak everything,
but they didn't.

This fix had no particular effect on the cost of running the
compiler on html/template/*.go and added a single new "escape"
to the standard library:

    syscall/syscall_unix.go:85: &b[0] escapes to heap
in
if errno := m.munmap(uintptr(unsafe.Pointer(&b[0])),
                     uintptr(len(b))); errno != nil {

Added specific escape testing to escape_calls.go
(and verified that it fails without this patch)

I also did a little code cleanup around the changes in esc.c.

Fixes #10925

Change-Id: I9984b701621ad4c49caed35b01e359295c210033
Reviewed-on: https://go-review.googlesource.com/10295
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agomath/big: gofmt nat_test.go
Robert Griesemer [Fri, 22 May 2015 01:06:34 +0000 (18:06 -0700)]
math/big: gofmt nat_test.go

Was apparently checked in by https://go-review.googlesource.com/#/c/9253/
without being gofmt-ed.

TBR: agl

Change-Id: I4d011dbaa15b7c5e73ca71f724f32951a0302dae
Reviewed-on: https://go-review.googlesource.com/10353
Reviewed-by: Robert Griesemer <gri@golang.org>
9 years agomath/big: Simple Montgomery Multiplication to accelerate Mod-Exp
Vlad Krasnov [Wed, 22 Apr 2015 22:03:59 +0000 (15:03 -0700)]
math/big: Simple Montgomery Multiplication to accelerate Mod-Exp

On Haswell I measure anywhere between 2X to 3.5X speedup for RSA.
I believe other architectures will also greatly improve.
In the future may be upgraded by dedicated assembly routine.

Built-in benchmarks i5-4278U turbo off:

benchmark                         old ns/op     new ns/op     delta
BenchmarkRSA2048Decrypt           6696649       3073769       -54.10%
Benchmark3PrimeRSA2048Decrypt     4472340       1669080       -62.68%

Change-Id: I17df84f85e34208f990665f9f90ea671695b2add
Reviewed-on: https://go-review.googlesource.com/9253
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Adam Langley <agl@golang.org>
Reviewed-by: Vlad Krasnov <vlad@cloudflare.com>
Run-TryBot: Adam Langley <agl@golang.org>

9 years agoruntime: fix callwritebarrier
Russ Cox [Wed, 20 May 2015 02:58:10 +0000 (22:58 -0400)]
runtime: fix callwritebarrier

Given a call frame F of size N where the return values start at offset R,
callwritebarrier was instructing heapBitsBulkBarrier to scan the block
of memory [F+R, F+R+N). It should only scan [F+R, F+N). The extra N-R
bytes scanned might lead into the next allocated block in memory.
Because the scan was consulting the heap bitmap for type information,
scanning into the next block normally "just worked" in the sense of
not crashing.

Scanning the extra N-R bytes of memory is a problem mainly because
it causes the GC to consider pointers that might otherwise not be
considered, leading it to retain objects that should actually be freed.
This is very difficult to detect.

Luckily, juju turned up a case where the heap bitmap and the memory
were out of sync for the block immediately after the call frame, so that
heapBitsBulkBarrier saw an obvious non-pointer where it expected a
pointer, causing a loud crash.

Why is there a non-pointer in memory that the heap bitmap records as
a pointer? That is more difficult to answer. At least one way that it
could happen is that allocations containing no pointers at all do not
update the heap bitmap. So if heapBitsBulkBarrier walked out of the
current object and into a no-pointer object and consulted those bitmap
bits, it would be misled. This doesn't happen in general because all
the paths to heapBitsBulkBarrier first check for the no-pointer case.
This may or may not be what happened, but it's the only scenario
I've been able to construct.

I tried for quite a while to write a simple test for this and could not.
It does fix the juju crash, and it is clearly an improvement over the
old code.

Fixes #10844.

Change-Id: I53982c93ef23ef93155c4086bbd95a4c4fdaac9a
Reviewed-on: https://go-review.googlesource.com/10317
Reviewed-by: Austin Clements <austin@google.com>
9 years agoruntime: eliminate write barrier from adjustpointers
Austin Clements [Wed, 20 May 2015 15:57:02 +0000 (11:57 -0400)]
runtime: eliminate write barrier from adjustpointers

Currently adjustpointers invokes a write barrier for every stack slot
it updates. This is safe---the write barrier always does nothing
because the new value is never a heap pointer---but it's unnecessary
overhead in performance and complexity.

Fix this by rewriting adjustpointers to work with *uintptrs instead of
*unsafe.Pointers. As an added bonus, this makes the code cleaner.

name                   old mean              new mean              delta
BinaryTree17            3.35s × (0.98,1.01)   3.33s × (0.99,1.02)    ~    (p=0.095 n=20+19)
Fannkuch11              2.49s × (1.00,1.01)   2.52s × (0.99,1.01)  +1.23% (p=0.000 n=19+20)
FmtFprintfEmpty        52.2ns × (0.99,1.02)  52.2ns × (0.99,1.02)    ~    (p=0.766 n=19+19)
FmtFprintfString        181ns × (0.99,1.02)   179ns × (0.99,1.01)  -1.06% (p=0.000 n=20+19)
FmtFprintfInt           177ns × (0.99,1.01)   173ns × (0.99,1.02)  -2.26% (p=0.000 n=17+20)
FmtFprintfIntInt        300ns × (0.99,1.01)   302ns × (0.99,1.01)  +0.76% (p=0.000 n=19+20)
FmtFprintfPrefixedInt   253ns × (0.99,1.02)   256ns × (0.99,1.01)  +0.96% (p=0.000 n=20+19)
FmtFprintfFloat         334ns × (0.99,1.02)   334ns × (1.00,1.01)    ~    (p=0.243 n=20+19)
FmtManyArgs            1.16µs × (0.99,1.01)  1.17µs × (0.99,1.02)  +0.88% (p=0.000 n=20+20)
GobDecode              9.16ms × (0.99,1.02)  9.18ms × (1.00,1.00)  +0.21% (p=0.048 n=20+17)
GobEncode              7.03ms × (0.99,1.01)  7.05ms × (0.99,1.01)    ~    (p=0.091 n=19+19)
Gzip                    374ms × (0.99,1.01)   372ms × (0.99,1.02)  -0.50% (p=0.008 n=18+20)
Gunzip                 92.9ms × (0.99,1.01)  92.5ms × (1.00,1.01)  -0.47% (p=0.002 n=19+19)
HTTPClientServer       53.1µs × (0.98,1.01)  52.5µs × (0.99,1.01)  -0.98% (p=0.000 n=20+19)
JSONEncode             17.4ms × (0.99,1.02)  17.5ms × (0.99,1.01)    ~    (p=0.061 n=19+20)
JSONDecode             66.0ms × (0.99,1.02)  64.7ms × (0.99,1.01)  -1.87% (p=0.000 n=20+20)
Mandelbrot200          3.94ms × (1.00,1.01)  3.95ms × (1.00,1.01)    ~    (p=0.799 n=18+19)
GoParse                3.89ms × (0.99,1.02)  3.86ms × (0.99,1.01)  -0.70% (p=0.016 n=20+19)
RegexpMatchEasy0_32     102ns × (0.99,1.02)   102ns × (1.00,1.01)    ~    (p=0.557 n=20+18)
RegexpMatchEasy0_1K     353ns × (0.99,1.02)   341ns × (0.99,1.01)  -3.38% (p=0.000 n=20+20)
RegexpMatchEasy1_32    85.0ns × (0.99,1.02)  85.0ns × (0.99,1.01)    ~    (p=0.851 n=19+20)
RegexpMatchEasy1_1K     521ns × (0.99,1.02)   506ns × (1.00,1.01)  -2.85% (p=0.000 n=20+18)
RegexpMatchMedium_32    142ns × (0.99,1.02)   141ns × (1.00,1.01)  -1.17% (p=0.000 n=20+19)
RegexpMatchMedium_1K   42.8µs × (0.99,1.01)  42.3µs × (0.99,1.01)  -1.07% (p=0.000 n=20+19)
RegexpMatchHard_32     2.17µs × (0.99,1.01)  2.16µs × (1.00,1.01)  -0.51% (p=0.042 n=20+18)
RegexpMatchHard_1K     65.6µs × (0.99,1.01)  64.8µs × (1.00,1.00)  -1.21% (p=0.000 n=20+17)
Revcomp                 581ms × (0.99,1.04)   536ms × (1.00,1.01)  -7.71% (p=0.000 n=20+18)
Template               77.2ms × (0.99,1.01)  76.8ms × (0.99,1.01)    ~    (p=0.426 n=20+18)
TimeParse               369ns × (0.99,1.02)   371ns × (1.00,1.01)    ~    (p=0.117 n=20+19)
TimeFormat              371ns × (0.99,1.02)   391ns × (0.99,1.01)  +5.33% (p=0.000 n=20+19)

Change-Id: I5b952ba577ac4365c8c87db837c5804a1e30b7be
Reviewed-on: https://go-review.googlesource.com/10293
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/internal/obj: remove F3t field from Prog
Josh Bleecher Snyder [Thu, 21 May 2015 00:35:43 +0000 (17:35 -0700)]
cmd/internal/obj: remove F3t field from Prog

F3t was effectively a local variable.
Remove it.

This shrinks obj.Prog from 456 to 448 bytes,
which places it in a smaller malloc class.

This reduces the memory usage of the compiler
while compiling the rotate tests by ~2.75%.

Change-Id: I31cc9dd67269851a430b56bcc7d255c9349eb522
Reviewed-on: https://go-review.googlesource.com/10255
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/go: set correct install location for cmd/compile and cmd/link
Russ Cox [Thu, 21 May 2015 17:40:17 +0000 (13:40 -0400)]
cmd/go: set correct install location for cmd/compile and cmd/link

Without this, they install to $GOROOT/bin.

Change-Id: Iae4b8f59c8392f6abd841490e56922738089f8d4
Reviewed-on: https://go-review.googlesource.com/10297
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoall: retire architecture letter in file names, public API
Russ Cox [Thu, 21 May 2015 17:28:17 +0000 (13:28 -0400)]
all: retire architecture letter in file names, public API

This CL removes the remaining visible uses of the "architecture letter" concept.
(They are no longer in tool names nor in source directory names.)

Because the architecture letter concept is now gone, delete GOCHAR
from "go env" output, and change go/build.ArchChar to return an
error always.

The architecture letter is still used in the compiler and linker sources
as a clumsy architecture enumeration, but that use is not visible to
Go users and can be cleaned up separately.

Change-Id: I4d97a38f372003fb610c9c5241bea440d9dbeb8d
Reviewed-on: https://go-review.googlesource.com/10289
Reviewed-by: Rob Pike <r@golang.org>
9 years agoall: build and use go tool compile, go tool link
Russ Cox [Thu, 21 May 2015 17:28:13 +0000 (13:28 -0400)]
all: build and use go tool compile, go tool link

This CL fixes the build to use the newly created go tool compile
and go tool link in place of go tool 5g, go tool 5l, and so on.

See golang-dev thread titled "go tool compile, etc" for background.

Although it was not a primary motivation, this conversion does
reduce the wall clock time and cpu time required for make.bash
by about 10%.

Change-Id: I79cbbdb676cab029db8aeefb99a53178ff55f98d
Reviewed-on: https://go-review.googlesource.com/10288
Reviewed-by: Rob Pike <r@golang.org>
9 years agocmd/compile, cmd/link: create from 5g, 5l, etc
Russ Cox [Thu, 21 May 2015 17:28:10 +0000 (13:28 -0400)]
cmd/compile, cmd/link: create from 5g, 5l, etc

Trivial merging of 5g, 6g, ... into go tool compile,
and similarlly 5l, 6l, ... into go tool link.
The files compile/main.go and link/main.go are new.
Everything else in those directories is a move followed by
change of imports and package name.

This CL breaks the build. Manual fixups are in the next CL.

See golang-dev thread titled "go tool compile, etc" for background.

Change-Id: Id35ff5a5859ad9037c61275d637b1bd51df6828b
Reviewed-on: https://go-review.googlesource.com/10287
Reviewed-by: Dave Cheney <dave@cheney.net>
Reviewed-by: Rob Pike <r@golang.org>
9 years agocmd/link: move to cmd/newlink
Russ Cox [Thu, 21 May 2015 17:28:06 +0000 (13:28 -0400)]
cmd/link: move to cmd/newlink

In preparation for making the current linker cmd/link.
If cmd/newlink is ever completed, it can be moved back.

See golang-dev thread titled "go tool compile, etc" for background.

Change-Id: I4029580f470038240c5181a37ea4202ba971f9ef
Reviewed-on: https://go-review.googlesource.com/10286
Reviewed-by: Rob Pike <r@golang.org>
9 years agocmd/internal/ld: output dwarf in external link mode on darwin
Ryan Brown [Wed, 8 Apr 2015 19:55:34 +0000 (12:55 -0700)]
cmd/internal/ld: output dwarf in external link mode on darwin

Fixes #8973

Change-Id: I746fae430db6d8f9ebd33586b8cffcb31d688cc8
Reviewed-on: https://go-review.googlesource.com/10284
Run-TryBot: Minux Ma <minux@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoruntime: turn work buffer tracing off by default
Rick Hudson [Thu, 21 May 2015 13:09:35 +0000 (09:09 -0400)]
runtime: turn work buffer tracing off by default

During development we ran with monitoring code turned
on by default. This CL turns the work buffer monitoring
off. Performance change on most go1 benchmarks is small
or insignificant.

name                   old mean              new mean              delta
BinaryTree17            3.35s × (0.99,1.01)   3.35s × (0.99,1.01)    ~    (p=0.841 n=5+5)
Fannkuch11              2.59s × (1.00,1.01)   2.55s × (1.00,1.00)  -1.65% (p=0.008 n=5+5)
FmtFprintfEmpty        52.5ns × (0.99,1.02)  53.2ns × (0.98,1.01)    ~    (p=0.063 n=5+5)
FmtFprintfString        181ns × (1.00,1.00)   180ns × (1.00,1.00)  -0.55% (p=0.029 n=4+4)
FmtFprintfInt           176ns × (1.00,1.01)   174ns × (1.00,1.00)  -0.91% (p=0.000 n=5+4)
FmtFprintfIntInt        298ns × (1.00,1.00)   299ns × (1.00,1.00)    ~    (p=0.143 n=4+4)
FmtFprintfPrefixedInt   250ns × (1.00,1.01)   246ns × (1.00,1.00)  -1.68% (p=0.000 n=5+4)
FmtFprintfFloat         340ns × (1.00,1.00)   340ns × (1.00,1.01)    ~    (p=0.643 n=5+5)
FmtManyArgs            1.16µs × (1.00,1.00)  1.15µs × (1.00,1.00)  -0.47% (p=0.016 n=5+5)
GobDecode              9.22ms × (1.00,1.00)  9.23ms × (1.00,1.00)    ~    (p=0.841 n=5+5)
GobEncode              7.00ms × (1.00,1.01)  7.09ms × (0.99,1.01)  +1.26% (p=0.016 n=5+5)
Gzip                    387ms × (1.00,1.00)   389ms × (0.99,1.02)    ~    (p=1.000 n=5+5)
Gunzip                 97.8ms × (1.00,1.00)  98.3ms × (1.00,1.00)  +0.51% (p=0.016 n=5+4)
HTTPClientServer       52.6µs × (1.00,1.01)  52.7µs × (1.00,1.01)    ~    (p=1.000 n=5+5)
JSONEncode             18.0ms × (0.99,1.02)  17.9ms × (1.00,1.00)    ~    (p=0.310 n=5+5)
JSONDecode             64.8ms × (0.99,1.02)  63.6ms × (1.00,1.00)  -1.94% (p=0.008 n=5+5)
Mandelbrot200          4.05ms × (1.00,1.00)  4.05ms × (1.00,1.00)    ~    (p=0.421 n=5+5)
GoParse                3.86ms × (1.00,1.01)  3.84ms × (0.99,1.01)    ~    (p=0.421 n=5+5)
RegexpMatchEasy0_32     101ns × (1.00,1.00)   102ns × (0.99,1.02)    ~    (p=0.238 n=4+5)
RegexpMatchEasy0_1K     346ns × (1.00,1.01)   345ns × (1.00,1.00)    ~    (p=0.333 n=5+4)
RegexpMatchEasy1_32    87.3ns × (0.99,1.02)  87.4ns × (1.00,1.00)    ~    (p=0.190 n=5+4)
RegexpMatchEasy1_1K     520ns × (1.00,1.00)   520ns × (1.00,1.01)    ~    (p=1.000 n=4+5)
RegexpMatchMedium_32    143ns × (1.00,1.00)   142ns × (1.00,1.00)  -0.70% (p=0.029 n=4+4)
RegexpMatchMedium_1K   43.2µs × (1.00,1.01)  43.2µs × (1.00,1.00)    ~    (p=0.841 n=5+5)
RegexpMatchHard_32     2.24µs × (1.00,1.01)  2.23µs × (1.00,1.01)  -0.63% (p=0.048 n=5+5)
RegexpMatchHard_1K     68.7µs × (1.00,1.00)  68.3µs × (1.00,1.00)  -0.56% (p=0.008 n=5+5)
Revcomp                 577ms × (1.00,1.01)   579ms × (1.00,1.00)    ~    (p=0.151 n=5+5)
Template               74.9ms × (1.00,1.00)  76.5ms × (1.00,1.00)  +2.11% (p=0.008 n=5+5)
TimeParse               359ns × (1.00,1.00)   362ns × (1.00,1.00)  +0.72% (p=0.008 n=5+5)
TimeFormat              369ns × (1.00,1.00)   371ns × (1.00,1.01)    ~    (p=0.071 n=5+5)

Change-Id: I4206a3f77a3d1450966b7a62ea7597aec44cb72f
Reviewed-on: https://go-review.googlesource.com/10294
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
9 years agoruntime: make runtime.callers walk calling G, not g0
Austin Clements [Wed, 20 May 2015 15:50:48 +0000 (11:50 -0400)]
runtime: make runtime.callers walk calling G, not g0

Currently runtime.callers invokes gentraceback with the pc and sp of
the G it is called from, but always passes g0 even if it was called
from a regular g. Right now this has no ill effects because
runtime.callers does not use either callback argument or the
_TraceJumpStack flag, but it makes the code fragile and will break
some upcoming changes.

Fix this by lifting the getg() call outside of the systemstack in
runtime.callers.

Change-Id: I4e1e927961c0e0cd4dcf28693be47df7bae9e122
Reviewed-on: https://go-review.googlesource.com/10292
Reviewed-by: Daniel Morsing <daniel.morsing@gmail.com>
Reviewed-by: Rick Hudson <rlh@golang.org>
9 years agoencoding/gob: fix docs
Shenghou Ma [Tue, 19 May 2015 04:19:04 +0000 (00:19 -0400)]
encoding/gob: fix docs

Fixes #10908.

Change-Id: I5ac4bd90204bc230610dcced47ce5b2253e5a004
Reviewed-on: https://go-review.googlesource.com/10250
Reviewed-by: Rob Pike <r@golang.org>
9 years agonet: document that ListenMulticastUDP is for simple applications
Mikio Hara [Thu, 7 May 2015 10:10:29 +0000 (19:10 +0900)]
net: document that ListenMulticastUDP is for simple applications

Also mentions golang.org/x/net/ipv4 and golang.org/x/net/ipv6.

Change-Id: I653deac7a5e2b129237655a72d6c91207f1b1685
Reviewed-on: https://go-review.googlesource.com/9779
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agogo/parser: parse incomplete selection "fmt." as a blank selection "fmt._"
Alan Donovan [Wed, 20 May 2015 19:49:23 +0000 (15:49 -0400)]
go/parser: parse incomplete selection "fmt." as a blank selection "fmt._"

Formerly it would return a BadExpr.

This prevents partial syntax from being discarded, and makes the error
recovery logic more consistent with other places where an identifier
was expected but not found.

+ test

Change-Id: I223c0c0589e7ceb7207ae951b8f71b9275a1eb73
Reviewed-on: https://go-review.googlesource.com/10269
Reviewed-by: Robert Griesemer <gri@golang.org>
9 years agocmd/doc: fix handling of paths like ./fmt
Rob Pike [Wed, 20 May 2015 20:26:02 +0000 (13:26 -0700)]
cmd/doc: fix handling of paths like ./fmt

An error in string slice offsets caused the loop to run forever if the
first character in the argument was a period.

Fixes #10833.

Change-Id: Iefb6aac5cff8864fe93d08e2600cb07d82c6f6df
Reviewed-on: https://go-review.googlesource.com/10285
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoruntime: remove unused quiesce code
Rick Hudson [Wed, 20 May 2015 15:40:17 +0000 (11:40 -0400)]
runtime: remove unused quiesce code

This is dead code. If you want to quiesce the system the
preferred way is to use forEachP(func(*p){}).

Change-Id: Ic7677a5dd55e3639b99e78ddeb2c71dd1dd091fa
Reviewed-on: https://go-review.googlesource.com/10267
Reviewed-by: Austin Clements <austin@google.com>
9 years agocmd/go: change Package.Shlib to be the absolute path of the shared library
Michael Hudson-Doyle [Mon, 18 May 2015 01:03:22 +0000 (13:03 +1200)]
cmd/go: change Package.Shlib to be the absolute path of the shared library

Makes little difference internally but makes go list output more useful.

Change-Id: I1fa1f839107de08818427382b2aef8dc4d765b36
Reviewed-on: https://go-review.googlesource.com/10192
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>

9 years agonet: fix the series of TestLookup and external tests
Mikio Hara [Wed, 13 May 2015 03:44:45 +0000 (12:44 +0900)]
net: fix the series of TestLookup and external tests

On Windows, we need to make sure that the node under test has external
connectivity.

Fixes #10795.

Change-Id: I99f2336180c7b56474fa90a4a6cdd5a6c4dd3805
Reviewed-on: https://go-review.googlesource.com/10006
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agonet: fix data race in TestSocket{Conn,PacketConn}
Mikio Hara [Mon, 18 May 2015 08:27:33 +0000 (17:27 +0900)]
net: fix data race in TestSocket{Conn,PacketConn}

Fixes #10891.

Change-Id: Ie432c9c5520ac29cea8fe6452628ec467567eea5
Reviewed-on: https://go-review.googlesource.com/10194
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agoRevert "cmd/internal/ld: output dwarf in external link mode on darwin"
Josh Bleecher Snyder [Tue, 19 May 2015 23:49:24 +0000 (23:49 +0000)]
Revert "cmd/internal/ld: output dwarf in external link mode on darwin"

This reverts commit 8b83306cf20abed54d7cb23a3f3091b7e6202056.

Change-Id: I3fb998bdf11eceef13e3997e336d86e7c5d47a60
Reviewed-on: https://go-review.googlesource.com/10254
Reviewed-by: Minux Ma <minux@golang.org>
9 years agomisc/cgo/testshared: when checking for RPATHs also look for DT_RUNPATH
Shenghou Ma [Tue, 19 May 2015 06:48:15 +0000 (02:48 -0400)]
misc/cgo/testshared: when checking for RPATHs also look for DT_RUNPATH

On my systems, ld -rpath sets DT_RUNPATH instead of DT_RPATH.

Change-Id: I5047e795fb7ef9336f5fa13ba24bb6245c0b0582
Reviewed-on: https://go-review.googlesource.com/10260
Reviewed-by: Michael Hudson-Doyle <michael.hudson@canonical.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/internal/gc: handle 64-bit const i/j/k in cgen_slice on ARM
Shenghou Ma [Tue, 19 May 2015 07:39:30 +0000 (03:39 -0400)]
cmd/internal/gc: handle 64-bit const i/j/k in cgen_slice on ARM

386 is not affected because it doesn't use ginscmp.

Fixes #10843.

Change-Id: I1b3a133bd1e5fabc85236f15d060dbaa4c391cf3
Reviewed-on: https://go-review.googlesource.com/10116
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agohtml/template: fix string iteration in replacement operations
Didier Spezia [Thu, 14 May 2015 22:36:59 +0000 (22:36 +0000)]
html/template: fix string iteration in replacement operations

In css, js, and html, the replacement operations are implemented
by iterating on strings (rune by rune). The for/range
statement is used. The length of the rune is required
and added to the index to properly slice the string.

This is potentially wrong because there is a discrepancy between
the result of utf8.RuneLen and the increment of the index
(set by the for/range statement). For invalid strings,
utf8.RuneLen('\ufffd') == 3, while the index is incremented
only by 1 byte.

htmlReplacer triggers a panic at slicing time for some
invalid strings.

Use a more robust iteration mechanism based on
utf8.DecodeRuneInString, and make sure the same
pattern is used for all similar functions in this
package.

Fixes #10799

Change-Id: Ibad3857b2819435d9fa564f06fc2ca8774102841
Reviewed-on: https://go-review.googlesource.com/10105
Reviewed-by: Rob Pike <r@golang.org>
9 years agocmd/go: fix build
Shenghou Ma [Tue, 19 May 2015 22:25:22 +0000 (18:25 -0400)]
cmd/go: fix build

Change-Id: Ib6c121414c74f8a40eb87a52af8737502ce7216d
Reviewed-on: https://go-review.googlesource.com/10265
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
9 years agocmd/internal/gc: remove incorrect "write barrier prohibited" error
Austin Clements [Tue, 19 May 2015 20:58:14 +0000 (16:58 -0400)]
cmd/internal/gc: remove incorrect "write barrier prohibited" error

Commit 9c9e36b pushed these errors down to where the write barriers
are actually emitted, but forgot to remove the original error that was
being pushed down.

Change-Id: I751752a896e78fb9e63d69f88e7fb8d1ff5d344c
Reviewed-on: https://go-review.googlesource.com/10264
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/go: run gofmt from current GOROOT
Alexander Zolotov [Sat, 9 May 2015 13:41:32 +0000 (16:41 +0300)]
cmd/go: run gofmt from current GOROOT

The existing implementation executes `gofmt` binary from PATH
environment variable on invocation `go fmt` command.
Relying on PATH might lead to confusions for users with several Go installations.
It's more appropriate to run `gofmt` from GOBIN (if defined) or GOROOT.

Fixes #10755

Change-Id: I56d42a747319c766f2911508fab3994c3a366d12
Reviewed-on: https://go-review.googlesource.com/9900
Reviewed-by: Rob Pike <r@golang.org>
9 years agocmd/internal/ld: output dwarf in external link mode on darwin
Ryan Brown [Wed, 8 Apr 2015 19:55:34 +0000 (12:55 -0700)]
cmd/internal/ld: output dwarf in external link mode on darwin

Fixes #8973

Change-Id: Idd53fc6d9e6971ae31ed72a3df3cfdce0bfbc1fd
Reviewed-on: https://go-review.googlesource.com/8661
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>

9 years agocmd/internal/gc: sync nowritebarrier checks and write barrier insertion
Russ Cox [Tue, 19 May 2015 19:15:52 +0000 (15:15 -0400)]
cmd/internal/gc: sync nowritebarrier checks and write barrier insertion

Change-Id: I348223d0336e28d95b8e68d7653aa547acc7c9c3
Reviewed-on: https://go-review.googlesource.com/10262
Reviewed-by: Austin Clements <austin@google.com>
9 years agocmd/internal/gc: type of str[i] is byte, not uint8
Russ Cox [Tue, 19 May 2015 18:00:27 +0000 (14:00 -0400)]
cmd/internal/gc: type of str[i] is byte, not uint8

Fixes #8745.

Change-Id: Id0641e3c0f259812b41ed871e83c68740feb2b19
Reviewed-on: https://go-review.googlesource.com/10261
Reviewed-by: Austin Clements <austin@google.com>
9 years agoruntime: run background mark helpers only if work is available
Rick Hudson [Mon, 18 May 2015 20:02:37 +0000 (16:02 -0400)]
runtime: run background mark helpers only if work is available

Prior to this CL whenever the GC marking was enabled and
a P was looking for work we supplied a G to help
the GC do its marking tasks. Once this G finished all
the marking available it would release the P to find another
available G. In the case where there was no work the P would drop
into findrunnable which would execute the mark helper G which would
immediately return and the P would drop into findrunnable again repeating
the process. Since the P was always given a G to run it never blocks.
This CL first checks if the GC mark helper G has available work and if
not the P immediately falls through to its blocking logic.

Fixes #10901

Change-Id: I94ac9646866ba64b7892af358888bc9950de23b5
Reviewed-on: https://go-review.googlesource.com/10189
Reviewed-by: Austin Clements <austin@google.com>
9 years agoruntime: minor clean up to heapminimum
Austin Clements [Sun, 17 May 2015 01:14:37 +0000 (21:14 -0400)]
runtime: minor clean up to heapminimum

Currently setGCPercent sets heapminimum to heapminimum*GOGC/100. The
real intent is to set heapminimum to a scaled multiple of a fixed
default heap minimum, not to scale heapminimum based on its current
value. This turns out to be okay because setGCPercent is only called
once and heapminimum is initially set to this default heap minimum.
However, the code as written is confusing, especially since
setGCPercent is otherwise written so it could be called again to
change GOGC. Fix this by introducing a defaultHeapMinimum constant and
using this instead of the current value of heapminimum to compute the
scaled heap minimum.

As part of this, this commit improves the documentation on
heapminimum.

Change-Id: I4eb82c73dc2eb44a6e5a17c780a747a2e73d7493
Reviewed-on: https://go-review.googlesource.com/10181
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoruntime: add fast check for self-loop pointer in scanobject
Russ Cox [Mon, 18 May 2015 15:40:29 +0000 (11:40 -0400)]
runtime: add fast check for self-loop pointer in scanobject

Addresses a problem reported on the mailing list.

This will come up mainly in programs custom allocators that batch allocations,
but it still helps in our programs, which mainly do not have such allocations.

name                   old mean              new mean              delta
BinaryTree17            5.95s × (0.97,1.03)   5.93s × (0.97,1.04)    ~    (p=0.613)
Fannkuch11              4.46s × (0.98,1.04)   4.33s × (0.99,1.01)  -2.93% (p=0.000)
FmtFprintfEmpty        86.6ns × (0.98,1.03)  86.8ns × (0.98,1.02)    ~    (p=0.523)
FmtFprintfString        290ns × (0.98,1.05)   287ns × (0.98,1.03)    ~    (p=0.061)
FmtFprintfInt           271ns × (0.98,1.04)   286ns × (0.99,1.01)  +5.54% (p=0.000)
FmtFprintfIntInt        495ns × (0.98,1.04)   489ns × (0.99,1.01)  -1.24% (p=0.015)
FmtFprintfPrefixedInt   391ns × (0.99,1.02)   407ns × (0.99,1.01)  +4.00% (p=0.000)
FmtFprintfFloat         578ns × (0.99,1.01)   559ns × (0.99,1.01)  -3.35% (p=0.000)
FmtManyArgs            1.96µs × (0.98,1.05)  1.94µs × (0.99,1.01)  -1.33% (p=0.030)
GobDecode              15.9ms × (0.97,1.05)  15.7ms × (0.99,1.01)  -1.35% (p=0.044)
GobEncode              11.4ms × (0.97,1.05)  11.3ms × (0.98,1.03)    ~    (p=0.141)
Gzip                    658ms × (0.98,1.05)   648ms × (0.99,1.01)  -1.59% (p=0.009)
Gunzip                  144ms × (0.99,1.03)   144ms × (0.99,1.01)    ~    (p=0.867)
HTTPClientServer       92.1µs × (0.97,1.05)  90.3µs × (0.99,1.01)  -1.89% (p=0.005)
JSONEncode             31.0ms × (0.96,1.07)  30.2ms × (0.98,1.03)  -2.66% (p=0.001)
JSONDecode              110ms × (0.97,1.04)   107ms × (0.99,1.01)  -2.59% (p=0.000)
Mandelbrot200          6.15ms × (0.98,1.04)  6.07ms × (0.99,1.02)  -1.32% (p=0.045)
GoParse                6.79ms × (0.97,1.04)  6.74ms × (0.97,1.04)    ~    (p=0.242)
RegexpMatchEasy0_32     158ns × (0.98,1.05)   155ns × (0.99,1.01)  -1.64% (p=0.010)
RegexpMatchEasy0_1K     548ns × (0.97,1.04)   540ns × (0.99,1.01)  -1.34% (p=0.042)
RegexpMatchEasy1_32     133ns × (0.97,1.04)   132ns × (0.97,1.05)    ~    (p=0.466)
RegexpMatchEasy1_1K     899ns × (0.96,1.05)   878ns × (0.99,1.01)  -2.32% (p=0.002)
RegexpMatchMedium_32    250ns × (0.96,1.03)   243ns × (0.99,1.01)  -2.90% (p=0.000)
RegexpMatchMedium_1K   73.4µs × (0.98,1.04)  73.0µs × (0.98,1.04)    ~    (p=0.411)
RegexpMatchHard_32     3.87µs × (0.97,1.07)  3.84µs × (0.98,1.04)    ~    (p=0.273)
RegexpMatchHard_1K      120µs × (0.97,1.08)   117µs × (0.99,1.01)  -2.06% (p=0.010)
Revcomp                 940ms × (0.96,1.07)   924ms × (0.97,1.07)    ~    (p=0.071)
Template                128ms × (0.96,1.05)   128ms × (0.99,1.01)    ~    (p=0.502)
TimeParse               632ns × (0.96,1.07)   616ns × (0.99,1.01)  -2.58% (p=0.001)
TimeFormat              671ns × (0.97,1.06)   657ns × (0.99,1.02)  -2.10% (p=0.002)

In contrast to the one in test/bench/go1 (above), the binarytree program on the
shootout site uses more goroutines, batches allocations, and sets GOMAXPROCS
to runtime.NumCPU()*2.

Using that version, before vs after:

name          old mean             new mean             delta
BinaryTree20  18.6s × (0.96,1.05)  11.3s × (0.98,1.02)  -39.46% (p=0.000)

And Go 1.4 vs after:

name          old mean             new mean             delta
BinaryTree20  13.0s × (0.97,1.02)  11.3s × (0.98,1.02)  -13.21% (p=0.000)

There is still a scheduling problem - the raw run times are hiding the fact that
this chews up 2x the CPU - but we'll take care of that separately.

Change-Id: I3f5da879b24ae73a0d06745381ffb88c3744948b
Reviewed-on: https://go-review.googlesource.com/10220
Reviewed-by: Austin Clements <austin@google.com>
9 years agocmd/internal/gc: add missing write barrier in append(x, BigStructWithPointers)
Russ Cox [Mon, 18 May 2015 20:54:59 +0000 (16:54 -0400)]
cmd/internal/gc: add missing write barrier in append(x, BigStructWithPointers)

Fixes #10897.

Change-Id: I5c2d1f9d26333e2b2a0613ebf496daa465e07c24
Reviewed-on: https://go-review.googlesource.com/10221
Reviewed-by: Austin Clements <austin@google.com>
9 years agotime: document that not all Unix time can be represented
Shenghou Ma [Tue, 19 May 2015 04:25:47 +0000 (00:25 -0400)]
time: document that not all Unix time can be represented

Fixes #10906.

Change-Id: I7ae25a500df493c1e78183d69d89b3e2a64a0d1a
Reviewed-on: https://go-review.googlesource.com/10223
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agoflag: Fix up a package comment a bit.
Aaron Jacobs [Tue, 19 May 2015 00:47:24 +0000 (10:47 +1000)]
flag: Fix up a package comment a bit.

I think "the flag" was a typo, and the word "after" was repetitive.

Change-Id: I81c034ca11a3a778ff1eb4b3af5b96bc525ab985
Reviewed-on: https://go-review.googlesource.com/10195
Reviewed-by: Rob Pike <r@golang.org>
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agocmd/internal/gc: rearrange Node fields
Josh Bleecher Snyder [Mon, 18 May 2015 22:49:02 +0000 (15:49 -0700)]
cmd/internal/gc: rearrange Node fields

Rearrange Node fields to enable better struct packing.
This reduces readability in favor of shrinking
the size of Nodes.

This reduces the size of Node from 328 to 312.
This reduces the memory usage to compile the
rotate tests by about 4.4%.

No functional changes. Passes toolstash -cmp.

Updates #9933.

Change-Id: I2764c5847fb1635ddc898e2ee385d007d67f03c5
Reviewed-on: https://go-review.googlesource.com/10141
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/internal/gc: separate Node param fields
Josh Bleecher Snyder [Mon, 18 May 2015 17:27:59 +0000 (10:27 -0700)]
cmd/internal/gc: separate Node param fields

Param will be converted from an anonymous to a
named field in a subsequent, automated CL.

Reduces Node size from 368 to 328.
Reduces inuse_space on the rotate tests by about 3%.

No functional changes. Passes toolstash -cmp.

Updates #9933.

Change-Id: I5867b00328abf17ee24aea6ca58876bae9d8bfed
Reviewed-on: https://go-review.googlesource.com/10210
Reviewed-by: Russ Cox <rsc@golang.org>