]> Cypherpunks repositories - gostls13.git/log
gostls13.git
8 years agoruntime/cgo: fix callback on big-endian MIPS64
Cherry Zhang [Thu, 8 Sep 2016 12:33:19 +0000 (08:33 -0400)]
runtime/cgo: fix callback on big-endian MIPS64

Use MOVW, instead of MOVV, to pass an int32 arg. Also no need to
restore arg registers.

Fix big-endian MIPS64 build.

Change-Id: Ib43c71075c988153e5e5c5c6e7297b3fee28652a
Reviewed-on: https://go-review.googlesource.com/28830
Reviewed-by: Minux Ma <minux@golang.org>
8 years agodoc: avoid mentioning non-existence of u flag to fmt.Printf.
Jim Kingdon [Thu, 8 Sep 2016 18:27:04 +0000 (11:27 -0700)]
doc: avoid mentioning non-existence of u flag to fmt.Printf.

It is better to document what golang does, rather than how it differs
from languages which readers may or may not know.

That the output format is based on the type is basically self-evident
if you consider this just in go terms.

Change-Id: I0223e9b4cb67cc83a9ebe4d424e6c151d7ed600f
Reviewed-on: https://go-review.googlesource.com/28393
Reviewed-by: Rob Pike <r@golang.org>
8 years agonet/url: handle escaped paths in ResolveReference
Dave Day [Thu, 1 Sep 2016 03:13:37 +0000 (13:13 +1000)]
net/url: handle escaped paths in ResolveReference

Currently, path resolution is done using the un-escaped version of
paths. This means that path elements like one%2ftwo%2fthree are
handled incorrectly, and optional encodings (%2d vs. -) are dropped.

This function makes escaped handling consistent with Parse: provided
escapings are honoured, and RawPath is only set if necessary.

A helper method setPath is introduced to handle the correct setting of
Path and RawPath given the encoded path.

Fixes #16947

Change-Id: I40b1215e9066e88ec868b41635066eee220fde37
Reviewed-on: https://go-review.googlesource.com/28343
Run-TryBot: Dave Day <djd@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile: fix bug in oconv
Robert Griesemer [Thu, 8 Sep 2016 21:50:24 +0000 (14:50 -0700)]
cmd/compile: fix bug in oconv

Introduced by https://go-review.googlesource.com/#/c/28331/ .

Change-Id: Id75aed6410f06b302d5347f6ca6a2e19c61f6fb6
Reviewed-on: https://go-review.googlesource.com/28779
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: update and reenable Formats test
Robert Griesemer [Thu, 8 Sep 2016 21:39:38 +0000 (14:39 -0700)]
cmd/compile: update and reenable Formats test

Change-Id: I9c0da13d21551dbf766156472224370ab9badfe9
Reviewed-on: https://go-review.googlesource.com/28778
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: remove superfluous returns in fmt.go
Robert Griesemer [Thu, 1 Sep 2016 00:49:49 +0000 (17:49 -0700)]
cmd/compile: remove superfluous returns in fmt.go

Change-Id: Ie73fb460a3838c6d1b9348965a8b69c1bfa6a882
Reviewed-on: https://go-review.googlesource.com/28341
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Dave Cheney <dave@cheney.net>
8 years agocmd/compile: remove fmt.go printer again, now that it's unused
Robert Griesemer [Wed, 31 Aug 2016 23:28:18 +0000 (16:28 -0700)]
cmd/compile: remove fmt.go printer again, now that it's unused

Change-Id: I9a6e5b9cbcfc264c61fd39ed65330ca737707e1f
Reviewed-on: https://go-review.googlesource.com/28340
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: implement fmt.Formatter for Nodes formats %s, %v
Robert Griesemer [Wed, 31 Aug 2016 23:19:50 +0000 (16:19 -0700)]
cmd/compile: implement fmt.Formatter for Nodes formats %s, %v

Change-Id: Iac3a72cb6c5394f3c1a49f39125b0256d570e006
Reviewed-on: https://go-review.googlesource.com/28339
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: use fmt.State in nodedump
Robert Griesemer [Wed, 31 Aug 2016 23:07:49 +0000 (16:07 -0700)]
cmd/compile: use fmt.State in nodedump

Change-Id: Icd83e88fc879b30b34f8697d540619efeb25c25b
Reviewed-on: https://go-review.googlesource.com/28338
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: use fmt.State in exprfmt
Robert Griesemer [Wed, 31 Aug 2016 23:02:36 +0000 (16:02 -0700)]
cmd/compile: use fmt.State in exprfmt

Change-Id: If6c2d469c66a7aa8471bf54310354efdac3e0235
Reviewed-on: https://go-review.googlesource.com/28337
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: use fmt.State in nodefmt and stmtfmt
Robert Griesemer [Wed, 31 Aug 2016 22:32:35 +0000 (15:32 -0700)]
cmd/compile: use fmt.State in nodefmt and stmtfmt

Change-Id: Iac87007af4ee268b45f11ec05bf4757f2e7eedd8
Reviewed-on: https://go-review.googlesource.com/28336
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: implement fmt.Formatter for *Node formats %s, %v
Robert Griesemer [Wed, 31 Aug 2016 22:22:36 +0000 (15:22 -0700)]
cmd/compile: implement fmt.Formatter for *Node formats %s, %v

Change-Id: I80ed668cdeab0c4342b734d34b429927e0213e5a
Reviewed-on: https://go-review.googlesource.com/28335
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: implement fmt.Formatter for *Sym formats %s, %v
Robert Griesemer [Wed, 31 Aug 2016 21:12:35 +0000 (14:12 -0700)]
cmd/compile: implement fmt.Formatter for *Sym formats %s, %v

Change-Id: I0c362edba66c763e84990e3c5508013021f3e6fe
Reviewed-on: https://go-review.googlesource.com/28334
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: implement fmt.Formatter for *Type formats %s, %v
Robert Griesemer [Wed, 31 Aug 2016 17:32:40 +0000 (10:32 -0700)]
cmd/compile: implement fmt.Formatter for *Type formats %s, %v

Change-Id: I878ac549430abc7859c30d176d52d52ce02c5827
Reviewed-on: https://go-review.googlesource.com/28333
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: implement fmt.Formatter for Val formats %s, %v
Robert Griesemer [Tue, 30 Aug 2016 22:01:48 +0000 (15:01 -0700)]
cmd/compile: implement fmt.Formatter for Val formats %s, %v

Change-Id: Id56e886793161b48b445439e9a12109142064d3f
Reviewed-on: https://go-review.googlesource.com/28332
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: implement fmt.Formatter for Op formats %s, %v
Robert Griesemer [Tue, 30 Aug 2016 21:21:33 +0000 (14:21 -0700)]
cmd/compile: implement fmt.Formatter for Op formats %s, %v

Change-Id: I59e18fab37fd688fc1e578e2192e32e29fdf37f0
Reviewed-on: https://go-review.googlesource.com/28331
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: implement fmt.Formatter for *Node formats %s, %v, %j
Robert Griesemer [Tue, 30 Aug 2016 21:13:41 +0000 (14:13 -0700)]
cmd/compile: implement fmt.Formatter for *Node formats %s, %v, %j

Change-Id: I44ee5843bb9dfd65b9a18091f365355e84888f21
Reviewed-on: https://go-review.googlesource.com/28330
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agocmd/compile: temporarily disable Formats test
Robert Griesemer [Thu, 8 Sep 2016 21:12:28 +0000 (14:12 -0700)]
cmd/compile: temporarily disable Formats test

So we can submit a sequence of older changes that don't yet
update the formats in this file. We will then re-enable the
test with the updated formats.

Change-Id: I6ed559b83adc891bbf4b3d855a7dc1e428366f7f
Reviewed-on: https://go-review.googlesource.com/28776
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
8 years agoencoding/hex: implement examples using all exported functions
Emmanuel Odeke [Sun, 4 Sep 2016 10:49:02 +0000 (03:49 -0700)]
encoding/hex: implement examples using all exported functions

Fixes #11254.
Updates #16360.

Implements examples using all exported functions.

This CL also updates Decode documentation to
state that only hexadecimal characters are accepted
in the source slice src, but also that the length
of src must be even.

Change-Id: Id016a4ba814f940cd300f26581fb4b9d2aded306
Reviewed-on: https://go-review.googlesource.com/28482
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

8 years agocmd/compile: intrinsify Ctz, Bswap, and some atomics on ARM64
Cherry Zhang [Mon, 29 Aug 2016 20:26:57 +0000 (16:26 -0400)]
cmd/compile: intrinsify Ctz, Bswap, and some atomics on ARM64

Change-Id: Ia5bf72b70e6f6522d6fb8cd050e78f862d37b5ae
Reviewed-on: https://go-review.googlesource.com/27936
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agonet/http/httputil: remove custom hop-by-hop headers from response in ReverseProxy
Sina Siadat [Thu, 8 Sep 2016 07:09:12 +0000 (11:39 +0430)]
net/http/httputil: remove custom hop-by-hop headers from response in ReverseProxy

Hop-by-hop headers (explicitly mentioned in RFC 2616) were already
removed from the response. This removes the custom hop-by-hop
headers listed in the "Connection" header of the response.

Updates #16875

Change-Id: I6b8f261d38b8d72040722f3ded29755ef0303427
Reviewed-on: https://go-review.googlesource.com/28810
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agoencoding/json: Use a lookup table for safe characters
Kevin Burke [Sun, 26 Jun 2016 16:47:43 +0000 (09:47 -0700)]
encoding/json: Use a lookup table for safe characters

The previous check for characters inside of a JSON string that needed
to be escaped performed seven different boolean comparisons before
determining that a ASCII character did not need to be escaped. Most
characters do not need to be escaped, so this check can be done in a
more performant way.

Use the same strategy as the unicode package for precomputing a range
of characters that need to be escaped, then do a single lookup into a
character array to determine whether the character needs escaping.

On an AWS c4.large node:

$ benchstat benchmarks/master-bench benchmarks/json-table-bench
name                   old time/op    new time/op     delta
CodeEncoder-2            19.0ms ± 0%     15.5ms ± 1%  -18.16%        (p=0.000 n=19+20)
CodeMarshal-2            20.1ms ± 1%     16.8ms ± 2%  -16.35%        (p=0.000 n=20+21)
CodeDecoder-2            49.3ms ± 1%     49.5ms ± 2%     ~           (p=0.498 n=16+20)
DecoderStream-2           416ns ± 0%      416ns ± 1%     ~           (p=0.978 n=19+19)
CodeUnmarshal-2          51.0ms ± 1%     50.9ms ± 1%     ~           (p=0.490 n=19+17)
CodeUnmarshalReuse-2     48.5ms ± 2%     48.5ms ± 2%     ~           (p=0.989 n=20+19)
UnmarshalString-2         541ns ± 1%      532ns ± 1%   -1.75%        (p=0.000 n=20+21)
UnmarshalFloat64-2        485ns ± 1%      481ns ± 1%   -0.92%        (p=0.000 n=20+21)
UnmarshalInt64-2          429ns ± 1%      427ns ± 1%   -0.49%        (p=0.000 n=19+20)
Issue10335-2              631ns ± 1%      619ns ± 1%   -1.84%        (p=0.000 n=20+20)
NumberIsValid-2          19.1ns ± 0%     19.1ns ± 0%     ~     (all samples are equal)
NumberIsValidRegexp-2     689ns ± 1%      690ns ± 0%     ~           (p=0.150 n=20+20)
SkipValue-2              14.0ms ± 0%     14.0ms ± 0%   -0.05%        (p=0.000 n=18+18)
EncoderEncode-2           525ns ± 2%      512ns ± 1%   -2.33%        (p=0.000 n=20+18)

name                   old speed      new speed       delta
CodeEncoder-2           102MB/s ± 0%    125MB/s ± 1%  +22.20%        (p=0.000 n=19+20)
CodeMarshal-2          96.6MB/s ± 1%  115.6MB/s ± 2%  +19.56%        (p=0.000 n=20+21)
CodeDecoder-2          39.3MB/s ± 1%   39.2MB/s ± 2%     ~           (p=0.464 n=16+20)
CodeUnmarshal-2        38.1MB/s ± 1%   38.1MB/s ± 1%     ~           (p=0.525 n=19+17)
SkipValue-2             143MB/s ± 0%    143MB/s ± 0%   +0.05%        (p=0.000 n=18+18)

I also took the data set reported in #5683 (browser
telemetry data from Mozilla), added named structs for
the data set, and turned it into a proper benchmark:
https://github.com/kevinburke/jsonbench/blob/master/go/bench_test.go

The results from that test are similarly encouraging. On a 64-bit
Mac:

$ benchstat benchmarks/master-benchmark benchmarks/json-table-benchmark
name              old time/op    new time/op    delta
CodeMarshal-4       1.19ms ± 2%    1.08ms ± 2%   -9.33%  (p=0.000 n=21+17)
Unmarshal-4         3.09ms ± 3%    3.06ms ± 1%   -0.83%  (p=0.027 n=22+17)
UnmarshalReuse-4    3.04ms ± 1%    3.04ms ± 1%     ~     (p=0.169 n=20+15)

name              old speed      new speed      delta
CodeMarshal-4     80.3MB/s ± 1%  88.5MB/s ± 1%  +10.29%  (p=0.000 n=21+17)
Unmarshal-4       31.0MB/s ± 2%  31.2MB/s ± 1%   +0.83%  (p=0.025 n=22+17)

On the c4.large:

$ benchstat benchmarks/master-bench benchmarks/json-table-bench
name              old time/op    new time/op    delta
CodeMarshal-2       1.10ms ± 1%    0.98ms ± 1%  -10.12%  (p=0.000 n=20+54)
Unmarshal-2         2.82ms ± 1%    2.79ms ± 0%   -1.09%  (p=0.000 n=20+51)
UnmarshalReuse-2    2.80ms ± 0%    2.77ms ± 0%   -1.03%  (p=0.000 n=20+52)

name              old speed      new speed      delta
CodeMarshal-2     87.3MB/s ± 1%  97.1MB/s ± 1%  +11.27%  (p=0.000 n=20+54)
Unmarshal-2       33.9MB/s ± 1%  34.2MB/s ± 0%   +1.10%  (p=0.000 n=20+51)

For what it's worth, I tried other heuristics - short circuiting the
conditional for common ASCII characters, for example:

if (b >= 63 && b != 92) || (b >= 39 && b <= 59) || (rest of the conditional)

This offered a speedup around 7-9%, not as large as the submitted
change.

Change-Id: Idcf88f7b93bfcd1164cdd6a585160b7e407a0d9b
Reviewed-on: https://go-review.googlesource.com/24466
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agobytes: improve WriteRune performance
Martin Möhrmann [Wed, 7 Sep 2016 06:59:00 +0000 (08:59 +0200)]
bytes: improve WriteRune performance

Remove the runeBytes buffer and write the utf8 encoding directly
to the internal buf byte slice.

name         old time/op   new time/op   delta
WriteRune-4   80.5µs ± 2%   57.1µs ± 2%  -29.06%  (p=0.000 n=20+20)

name         old speed     new speed     delta
WriteRune-4  153MB/s ± 2%  215MB/s ± 2%  +40.96%  (p=0.000 n=20+20)

Change-Id: Ic15f6e2d6e56a3d15c74f56159e2eae020ba73ba
Reviewed-on: https://go-review.googlesource.com/28816
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agoruntime: simplify getargp
Josh Bleecher Snyder [Thu, 8 Sep 2016 16:49:31 +0000 (09:49 -0700)]
runtime: simplify getargp

Change-Id: I9ed62e8a6d8b9204c18748efd7845adabf3460b9
Reviewed-on: https://go-review.googlesource.com/28775
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile/internal/ssa/gen: fix error message for wrong arg length
Cherry Zhang [Wed, 7 Sep 2016 16:06:43 +0000 (12:06 -0400)]
cmd/compile/internal/ssa/gen: fix error message for wrong arg length

When arg length is wrong, op is not set, so it always prints
"should have 0 args".

Change-Id: If7bcb41d993919d0038d2a09e16188c79dfbd858
Reviewed-on: https://go-review.googlesource.com/28831
Reviewed-by: Keith Randall <khr@golang.org>
8 years agodoc: fix typo in the release notes
Michal Bohuslávek [Thu, 8 Sep 2016 08:52:36 +0000 (09:52 +0100)]
doc: fix typo in the release notes

Change-Id: I003795d8dc2176532ee133740bf35e23a3aa3878
Reviewed-on: https://go-review.googlesource.com/28811
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agoruntime: remove maxstring
Martin Möhrmann [Tue, 6 Sep 2016 08:38:16 +0000 (10:38 +0200)]
runtime: remove maxstring

Before this CL the runtime prevented printing of overlong strings with the print
function when the length of the string was determined to be corrupted.
Corruption was checked by comparing the string size against the limit
which was stored in maxstring.

However maxstring was not updated everywhere were go strings were created
e.g. for string constants during compile time. Thereby the check for maximum
string length prevented the printing of some valid strings.

The protection maxstring provided did not warrant the bookkeeping
and global synchronization needed to keep maxstring updated to the
correct limit everywhere.

Fixes #16999

Change-Id: I62cc2f4362f333f75b77f199ce1a71aac0ff7aeb
Reviewed-on: https://go-review.googlesource.com/28813
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agoio/ioutil: return better error when TempDir called with non-extant dir
Brad Fitzpatrick [Thu, 8 Sep 2016 04:51:21 +0000 (04:51 +0000)]
io/ioutil: return better error when TempDir called with non-extant dir

Fixes #14196

Change-Id: Ife7950289ac6adbcfc4d0f2fce31f20bc2657858
Reviewed-on: https://go-review.googlesource.com/28772
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile: remove unnecessary FuncType cloning
Matthew Dempsky [Tue, 6 Sep 2016 21:26:56 +0000 (14:26 -0700)]
cmd/compile: remove unnecessary FuncType cloning

Since FuncTypes are represented as structs rather than linking the
parameter lists together, we no longer need to worry about duplicating
the parameter lists.

Change-Id: I3767aa3cd1cbeddfb80a6eef6b42290dc2ac14ae
Reviewed-on: https://go-review.googlesource.com/28574
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agonet/http/httputil: copy header map if necessary in ReverseProxy
Sina Siadat [Sun, 4 Sep 2016 07:50:14 +0000 (12:20 +0430)]
net/http/httputil: copy header map if necessary in ReverseProxy

We were already making a copy of the map before removing
hop-by-hop headers. This commit does the same for proxied
headers mentioned in the "Connection" header.

A test is added to ensure request headers are not modified.

Updates #16875

Change-Id: I85329d212787958d5ad818915eb0538580a4653a
Reviewed-on: https://go-review.googlesource.com/28493
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

8 years agogo/format: add format.Node example
Alberto Donizetti [Wed, 10 Aug 2016 17:01:06 +0000 (19:01 +0200)]
go/format: add format.Node example

Updates #16360

Change-Id: I5927cffa961cd85539a3ba9606b116c5996d1096
Reviewed-on: https://go-review.googlesource.com/26696
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

8 years agoio: add test that MultiReader zeros exhausted Readers
Brad Fitzpatrick [Mon, 5 Sep 2016 21:13:24 +0000 (21:13 +0000)]
io: add test that MultiReader zeros exhausted Readers

Updates #16983
Updates #16996

Change-Id: I76390766385b2668632c95e172b2d243d7f66651
Reviewed-on: https://go-review.googlesource.com/28771
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agodoc: document go1.7.1
Chris Broadfoot [Wed, 7 Sep 2016 18:59:58 +0000 (11:59 -0700)]
doc: document go1.7.1

Change-Id: I4dc1ff7bfc67351a046f199dee8b7a9eadb1e524
Reviewed-on: https://go-review.googlesource.com/28693
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agonet: remove parsing of negative decimals in IPv4 literal
Joe Tsai [Fri, 2 Sep 2016 08:14:57 +0000 (01:14 -0700)]
net: remove parsing of negative decimals in IPv4 literal

https://golang.org/cl/27206 fixed the dtoi function such that
it now properly parses negative number. Ironically, this causes
several other functions that depended on dtoi to now (incorrectly)
parse negative numbers.

For example, ParseCIDR("-1.0.0.0/32") used to be rejected prior to the
above CL, but is now accepted even though it is an invalid CIDR notation.
This CL fixes that regression.

We fix this by removing the signed parsing logic entirely from dtoi.
It was introduced relatively recently in https://golang.org/cl/12447
to fix a bug where an invalid port was improperly being parsed as OK.
It seems to me that the fix in that CL to the port handling logic was
sufficient such that a change to dtoi was unnecessary.

Updates #16350

Change-Id: I414bb1aa27d0a226ebd4b05a09cb40d784691b43
Reviewed-on: https://go-review.googlesource.com/28414
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Mikio Hara <mikioh.mikioh@gmail.com>
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>

8 years agocmd/compile: add format verification test
Robert Griesemer [Fri, 2 Sep 2016 22:12:49 +0000 (15:12 -0700)]
cmd/compile: add format verification test

TestFormats finds potential (Printf, etc.) format strings.
If they are used in a call, the format verbs are verified
based on the matching argument type against a precomputed
table of valid formats (formatMapping, below). The table
can be used to automatically rewrite format strings with
the -u flag.

Run as: go test -run Formats [-u]

A formatMapping based on the existing formats is printed
when the test is run in verbose mode (-v flag). The table
needs to be updated whenever a new (type, format) combination
is found and the format verb is not 'v' (as in "%v").

Known bugs:
- indexed format strings ("%[2]s", etc.) are not suported
  (the test will fail)
- format strings that are not simple string literals cannot
  be updated automatically
  (the test will fail with respective warnings)

Change-Id: I1ca5bb6421d57ac78a00f1a80b9547a72837adc9
Reviewed-on: https://go-review.googlesource.com/28419
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agostrings: use AVX2 for Index if available
Ilya Tocar [Thu, 28 Apr 2016 14:39:55 +0000 (17:39 +0300)]
strings: use AVX2 for Index if available

IndexHard4-4      1.50ms ± 2%  0.71ms ± 0%  -52.36%  (p=0.000 n=20+19)

This also fixes a bug, that caused a string of length 16 to use
two 8-byte comparisons instead of one 16-byte. And adds a test for
cases when partial_match fails.

Change-Id: I1ee8fc4e068bb36c95c45de78f067c822c0d9df0
Reviewed-on: https://go-review.googlesource.com/22551
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agocmd/compile: ignore contentEscapes for marking nodes as escaping
Keith Randall [Wed, 7 Sep 2016 04:08:21 +0000 (21:08 -0700)]
cmd/compile: ignore contentEscapes for marking nodes as escaping

Redo of CL 28575 with fixed test.
We're in a pre-KeepAlive world for a bit yet, the old tests
were in a client which was in a post-KeepAlive world.

Change-Id: I114fd630339d761ab3306d1d99718d3cb973678d
Reviewed-on: https://go-review.googlesource.com/28582
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agosyscall: avoid convT2I allocs for common Windows error values
Brad Fitzpatrick [Sun, 4 Sep 2016 19:07:22 +0000 (19:07 +0000)]
syscall: avoid convT2I allocs for common Windows error values

This is was already done for Unix in https://golang.org/cl/6701 +
https://golang.org/cl/8192. Do it for Windows also.

Fixes #16988

Change-Id: Ia7832b0d0d48566b0cd205652b85130df529592e
Reviewed-on: https://go-review.googlesource.com/28484
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agogo/constant: document that Value.String and ExactString return quoted strings
Robert Griesemer [Wed, 7 Sep 2016 00:26:15 +0000 (17:26 -0700)]
go/constant: document that Value.String and ExactString return quoted strings

This has always been the case but it was not obvious from the documentation.
The reason for the quoting is that String() may return an abbreviated string,
starting with double-quote (") but ending in ... (w/o a quote). The missing
quote indicates the abbreviation (in contrast to a string ending in ...").

constant.StringVal can be used to obtain the unquoted string of a String Value.

Change-Id: Id0ba45b6ff62b3e024386ba8d907d6b3a4fcb6d7
Reviewed-on: https://go-review.googlesource.com/28576
Reviewed-by: Alan Donovan <adonovan@google.com>
8 years agoRevert of cmd/compile: ignore contentEscapes for marking nodes as escaping
Brad Fitzpatrick [Wed, 7 Sep 2016 03:23:20 +0000 (03:23 +0000)]
Revert of cmd/compile: ignore contentEscapes for marking nodes as escaping

Reason for revert: broke the build due to cherrypick;
relies on an unsubmitted parent CL.

Original issue's description:
> cmd/compile: ignore contentEscapes for marking nodes as escaping
>
> We can still stack allocate and VarKill nodes which don't
> escape but their content does.
>
> Fixes #16996
>
> Change-Id: If8aa0fcf2c327b4cb880a3d5af8d213289e6f6bf
> Reviewed-on: https://go-review.googlesource.com/28575
> Run-TryBot: Keith Randall <khr@golang.org>
> TryBot-Result: Gobot Gobot <gobot@golang.org>
> Reviewed-by: David Chase <drchase@google.com>
>

Change-Id: Ie1a325209de14d70af6acb2d78269b7a0450da7a
Reviewed-on: https://go-review.googlesource.com/28578
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile: ignore contentEscapes for marking nodes as escaping
Keith Randall [Tue, 6 Sep 2016 21:48:47 +0000 (14:48 -0700)]
cmd/compile: ignore contentEscapes for marking nodes as escaping

We can still stack allocate and VarKill nodes which don't
escape but their content does.

Fixes #16996

Change-Id: If8aa0fcf2c327b4cb880a3d5af8d213289e6f6bf
Reviewed-on: https://go-review.googlesource.com/28575
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
8 years agocmd/compile: add BVC/BVS to branch ops in ppc64/prog.go
David Chase [Tue, 6 Sep 2016 20:39:22 +0000 (13:39 -0700)]
cmd/compile: add BVC/BVS to branch ops in ppc64/prog.go

Includes test case shown to fail with unpatched compiler.

Fixes #17005.

Change-Id: I49b7b1a3f02736d85846a2588018b73f68d50320
Reviewed-on: https://go-review.googlesource.com/28573
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Minux Ma <minux@golang.org>
8 years agostrings: make IndexRune faster
Hiroshi Ioka [Tue, 6 Sep 2016 11:23:40 +0000 (20:23 +0900)]
strings: make IndexRune faster

re-implement IndexRune by Index which is well optimized to get
performance gain.

name                   old time/op  new time/op  delta
IndexRune-4            30.2ns ± 1%  28.3ns ± 1%   -6.22%  (p=0.000 n=20+19)
IndexRuneLongString-4   156ns ± 1%    49ns ± 1%  -68.72%  (p=0.000 n=19+19)
IndexRuneFastPath-4    10.6ns ± 2%  10.0ns ± 1%   -6.30%  (p=0.000 n=18+18)

Change-Id: Ie663b8f7860ca51892dd4be182fca3caa5f8ae61
Reviewed-on: https://go-review.googlesource.com/28546
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agosyscall: use MNT_NOWAIT in TestGetfsstat
Brad Fitzpatrick [Tue, 6 Sep 2016 17:02:19 +0000 (17:02 +0000)]
syscall: use MNT_NOWAIT in TestGetfsstat

Fixes test failure when VMWare's shared folder filesystem is present.

MNT_NOWAIT is what the mount(8) command does.

Fixes #16937

Change-Id: Id436185f544b7069db46c8716d6a0bf580b31da0
Reviewed-on: https://go-review.googlesource.com/28550
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agocmd/compile: omit some temp panicdivide calls
Josh Bleecher Snyder [Sun, 4 Sep 2016 23:59:46 +0000 (16:59 -0700)]
cmd/compile: omit some temp panicdivide calls

When the divisor is known to be a constant
non-zero, don't insert panicdivide calls
that will just be eliminated later.

The main benefit here is readability of the SSA
form for compiler developers.

Change-Id: Icb7d07fc996941fbaff84524ac3e4b53d8e75fda
Reviewed-on: https://go-review.googlesource.com/28530
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agobytes: make IndexRune faster
Hiroshi Ioka [Mon, 5 Sep 2016 23:09:27 +0000 (08:09 +0900)]
bytes: make IndexRune faster

re-implement IndexRune by IndexByte and Index which are well optimized
to get performance gain.

name                  old time/op   new time/op     delta
IndexRune/10-4         53.2ns ± 1%     29.1ns ± 1%    -45.32%  (p=0.008 n=5+5)
IndexRune/32-4          191ns ± 1%       27ns ± 1%    -85.75%  (p=0.008 n=5+5)
IndexRune/4K-4         23.5µs ± 1%      1.0µs ± 1%    -95.77%  (p=0.008 n=5+5)
IndexRune/4M-4         23.8ms ± 0%      1.0ms ± 2%    -95.90%  (p=0.008 n=5+5)
IndexRune/64M-4         384ms ± 1%       15ms ± 1%    -95.98%  (p=0.008 n=5+5)
IndexRuneASCII/10-4    61.5ns ± 0%     10.3ns ± 4%    -83.17%  (p=0.008 n=5+5)
IndexRuneASCII/32-4     203ns ± 0%       11ns ± 5%    -94.68%  (p=0.008 n=5+5)
IndexRuneASCII/4K-4    23.4µs ± 0%      0.3µs ± 2%    -98.60%  (p=0.008 n=5+5)
IndexRuneASCII/4M-4    24.0ms ± 1%      0.3ms ± 1%    -98.60%  (p=0.008 n=5+5)
IndexRuneASCII/64M-4    386ms ± 2%        6ms ± 1%    -98.57%  (p=0.008 n=5+5)

name                  old speed     new speed       delta
IndexRune/10-4        188MB/s ± 1%    344MB/s ± 1%    +82.91%  (p=0.008 n=5+5)
IndexRune/32-4        167MB/s ± 0%   1175MB/s ± 1%   +603.52%  (p=0.008 n=5+5)
IndexRune/4K-4        174MB/s ± 1%   4117MB/s ± 1%  +2262.71%  (p=0.008 n=5+5)
IndexRune/4M-4        176MB/s ± 0%   4299MB/s ± 2%  +2340.46%  (p=0.008 n=5+5)
IndexRune/64M-4       175MB/s ± 1%   4354MB/s ± 1%  +2388.57%  (p=0.008 n=5+5)
IndexRuneASCII/10-4   163MB/s ± 0%    968MB/s ± 4%   +494.66%  (p=0.008 n=5+5)
IndexRuneASCII/32-4   157MB/s ± 0%   2974MB/s ± 4%  +1788.59%  (p=0.008 n=5+5)
IndexRuneASCII/4K-4   175MB/s ± 0%  12481MB/s ± 2%  +7027.71%  (p=0.008 n=5+5)
IndexRuneASCII/4M-4   175MB/s ± 1%  12510MB/s ± 1%  +7061.15%  (p=0.008 n=5+5)
IndexRuneASCII/64M-4  174MB/s ± 2%  12143MB/s ± 1%  +6881.70%  (p=0.008 n=5+5)

Change-Id: I0632eadb83937c2a9daa7f0ce79df1dee64f992e
Reviewed-on: https://go-review.googlesource.com/28537
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agoruntime/debug: enable TestFreeOSMemory on all arches
Austin Clements [Fri, 19 Aug 2016 19:51:00 +0000 (15:51 -0400)]
runtime/debug: enable TestFreeOSMemory on all arches

TestFreeOSMemory was disabled on many arches because of issue #9993.
Since that's been fixed, enable the test everywhere.

Change-Id: I298c38c3e04128d9c8a1f558980939d5699bea03
Reviewed-on: https://go-review.googlesource.com/27403
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Minux Ma <minux@golang.org>
8 years agosyscall: make Getpagesize return page size from runtime
Austin Clements [Tue, 19 Jul 2016 02:00:45 +0000 (22:00 -0400)]
syscall: make Getpagesize return page size from runtime

syscall.Getpagesize currently returns hard-coded page sizes on all
architectures (some of which are probably always wrong, and some of
which are definitely not always right). The runtime now has this
information, queried from the OS during runtime init, so make
syscall.Getpagesize return the page size that the runtime knows.

Updates #10180.

Change-Id: I4daa6fbc61a2193eb8fa9e7878960971205ac346
Reviewed-on: https://go-review.googlesource.com/25051
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agoruntime: don't hard-code physical page size
Austin Clements [Mon, 18 Jul 2016 16:24:02 +0000 (12:24 -0400)]
runtime: don't hard-code physical page size

Now that the runtime fetches the true physical page size from the OS,
make the physical page size used by heap growth a variable instead of
a constant. This isn't used in any performance-critical paths, so it
shouldn't be an issue.

sys.PhysPageSize is also renamed to sys.DefaultPhysPageSize to make it
clear that it's not necessarily the true page size. There are no uses
of this constant any more, but we'll keep it around for now.

Updates #12480 and #10180.

Change-Id: I6c23b9df860db309c38c8287a703c53817754f03
Reviewed-on: https://go-review.googlesource.com/25022
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
8 years agoruntime: fetch physical page size from the OS
Austin Clements [Tue, 19 Jul 2016 01:40:02 +0000 (21:40 -0400)]
runtime: fetch physical page size from the OS

Currently the physical page size assumed by the runtime is hard-coded.
On Linux the runtime at least fetches the OS page size during init and
sanity checks against the hard-coded value, but they may still differ.
On other OSes we wouldn't even notice.

Add support on all OSes to fetch the actual OS physical page size
during runtime init and lift the sanity check of PhysPageSize from the
Linux init code to general malloc init. Currently this is the only use
of the retrieved page size, but we'll add more shortly.

Updates #12480 and #10180.

Change-Id: I065f2834bc97c71d3208edc17fd990ec9058b6da
Reviewed-on: https://go-review.googlesource.com/25050
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
8 years agoruntime: assume 64kB physical pages on ARM
Austin Clements [Mon, 18 Jul 2016 16:32:29 +0000 (12:32 -0400)]
runtime: assume 64kB physical pages on ARM

Currently we assume the physical page size on ARM is 4kB. While this
is usually true, the architecture also supports 16kB and 64kB physical
pages, and Linux (and possibly other OSes) can be configured to use
these larger page sizes.

With Go 1.6, such a configuration could potentially run, but generally
resulted in memory corruption or random panics. With current master,
this configuration will cause the runtime to panic during init on
Linux when it checks the true physical page size (and will still cause
corruption or panics on other OSes).

However, the assumed physical page size only has to be a multiple of
the true physical page size, the scavenger can now deal with large
physical page sizes, and the rest of the runtime can deal with a
larger assumed physical page size than the true size. Hence, there's
little disadvantage to conservatively setting the assumed physical
page size to 64kB on ARM.

This may result in some extra memory use, since we can only return
memory at multiples of the assumed physical page size. However, it is
a simple change that should make Go run on systems configured for
larger page sizes. The following commits will make the runtime query
the actual physical page size from the OS, but this is a simple step
there.

Updates #12480.

Change-Id: I851829595bc9e0c76235c847a7b5f62ad82b5302
Reviewed-on: https://go-review.googlesource.com/25021
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Minux Ma <minux@golang.org>
8 years agoruntime: bound scanobject to ~100 µs
Austin Clements [Sat, 28 May 2016 01:04:40 +0000 (21:04 -0400)]
runtime: bound scanobject to ~100 µs

Currently the time spent in scanobject is proportional to the size of
the object being scanned. Since scanobject is non-preemptible, large
objects can cause significant goroutine (and even whole application)
delays through several means:

1. If a GC assist picks up a large object, the allocating goroutine is
   blocked for the whole scan, even if that scan well exceeds that
   goroutine's debt.

2. Since the scheduler does not run on the P performing a large object
   scan, goroutines in that P's run queue do not run unless they are
   stolen by another P (which can take some time). If there are a few
   large objects, all of the Ps may get tied up so the scheduler
   doesn't run anywhere.

3. Even if a large object is scanned by a background worker and other
   Ps are still running the scheduler, the large object scan doesn't
   flush background credit until the whole scan is done. This can
   easily cause all allocations to block in assists, waiting for
   credit, causing an effective STW.

Fix this by splitting large objects into 128 KB "oblets" and scanning
at most one oblet at a time. Since we can scan 1–2 MB/ms, this equates
to bounding scanobject at roughly 100 µs. This improves assist
behavior both because assists can no longer get "unlucky" and be stuck
scanning a large object, and because it causes the background worker
to flush credit and unblock assists more frequently when scanning
large objects. This also improves GC parallelism if the heap consists
primarily of a small number of very large objects by letting multiple
workers scan a large objects in parallel.

Fixes #10345. Fixes #16293.

This substantially improves goroutine latency in the benchmark from
issue #16293, which exercises several forms of very large objects:

name                 old max-latency    new max-latency    delta
SliceNoPointer-12           154µs ± 1%        155µs ±  2%     ~     (p=0.087 n=13+12)
SlicePointer-12             314ms ± 1%       5.94ms ±138%  -98.11%  (p=0.000 n=19+20)
SliceLivePointer-12        1148ms ± 0%       4.72ms ±167%  -99.59%  (p=0.000 n=19+20)
MapNoPointer-12           72509µs ± 1%        408µs ±325%  -99.44%  (p=0.000 n=19+18)
ChanPointer-12              313ms ± 0%       4.74ms ±140%  -98.49%  (p=0.000 n=18+20)
ChanLivePointer-12         1147ms ± 0%       3.30ms ±149%  -99.71%  (p=0.000 n=19+20)

name                 old P99.9-latency  new P99.9-latency  delta
SliceNoPointer-12           113µs ±25%         107µs ±12%     ~     (p=0.153 n=20+18)
SlicePointer-12          309450µs ± 0%         133µs ±23%  -99.96%  (p=0.000 n=20+20)
SliceLivePointer-12         961ms ± 0%        1.35ms ±27%  -99.86%  (p=0.000 n=20+20)
MapNoPointer-12            448µs ±288%         119µs ±18%  -73.34%  (p=0.000 n=18+20)
ChanPointer-12           309450µs ± 0%         134µs ±23%  -99.96%  (p=0.000 n=20+19)
ChanLivePointer-12          961ms ± 0%        1.35ms ±27%  -99.86%  (p=0.000 n=20+20)

This has negligible effect on all metrics from the garbage, JSON, and
HTTP x/benchmarks.

It shows slight improvement on some of the go1 benchmarks,
particularly Revcomp, which uses some multi-megabyte buffers:

name                      old time/op    new time/op    delta
BinaryTree17-12              2.46s ± 1%     2.47s ± 1%  +0.32%  (p=0.012 n=20+20)
Fannkuch11-12                2.82s ± 0%     2.81s ± 0%  -0.61%  (p=0.000 n=17+20)
FmtFprintfEmpty-12          50.8ns ± 5%    50.5ns ± 2%    ~     (p=0.197 n=17+19)
FmtFprintfString-12          131ns ± 1%     132ns ± 0%  +0.57%  (p=0.000 n=20+16)
FmtFprintfInt-12             117ns ± 0%     116ns ± 0%  -0.47%  (p=0.000 n=15+20)
FmtFprintfIntInt-12          180ns ± 0%     179ns ± 1%  -0.78%  (p=0.000 n=16+20)
FmtFprintfPrefixedInt-12     186ns ± 1%     185ns ± 1%  -0.55%  (p=0.000 n=19+20)
FmtFprintfFloat-12           263ns ± 1%     271ns ± 0%  +2.84%  (p=0.000 n=18+20)
FmtManyArgs-12               741ns ± 1%     742ns ± 1%    ~     (p=0.190 n=19+19)
GobDecode-12                7.44ms ± 0%    7.35ms ± 1%  -1.21%  (p=0.000 n=20+20)
GobEncode-12                6.22ms ± 1%    6.21ms ± 1%    ~     (p=0.336 n=20+19)
Gzip-12                      220ms ± 1%     219ms ± 1%    ~     (p=0.130 n=19+19)
Gunzip-12                   37.9ms ± 0%    37.9ms ± 1%    ~     (p=1.000 n=20+19)
HTTPClientServer-12         82.5µs ± 3%    82.6µs ± 3%    ~     (p=0.776 n=20+19)
JSONEncode-12               16.4ms ± 1%    16.5ms ± 2%  +0.49%  (p=0.003 n=18+19)
JSONDecode-12               53.7ms ± 1%    54.1ms ± 1%  +0.71%  (p=0.000 n=19+18)
Mandelbrot200-12            4.19ms ± 1%    4.20ms ± 1%    ~     (p=0.452 n=19+19)
GoParse-12                  3.38ms ± 1%    3.37ms ± 1%    ~     (p=0.123 n=19+19)
RegexpMatchEasy0_32-12      72.1ns ± 1%    71.8ns ± 1%    ~     (p=0.397 n=19+17)
RegexpMatchEasy0_1K-12       242ns ± 0%     242ns ± 0%    ~     (p=0.168 n=17+20)
RegexpMatchEasy1_32-12      72.1ns ± 1%    72.1ns ± 1%    ~     (p=0.538 n=18+19)
RegexpMatchEasy1_1K-12       385ns ± 1%     384ns ± 1%    ~     (p=0.388 n=20+20)
RegexpMatchMedium_32-12      112ns ± 1%     112ns ± 3%    ~     (p=0.539 n=20+20)
RegexpMatchMedium_1K-12     34.4µs ± 2%    34.4µs ± 2%    ~     (p=0.628 n=18+18)
RegexpMatchHard_32-12       1.80µs ± 1%    1.80µs ± 1%    ~     (p=0.522 n=18+19)
RegexpMatchHard_1K-12       54.0µs ± 1%    54.1µs ± 1%    ~     (p=0.647 n=20+19)
Revcomp-12                   387ms ± 1%     369ms ± 5%  -4.89%  (p=0.000 n=17+19)
Template-12                 62.3ms ± 1%    62.0ms ± 0%  -0.48%  (p=0.002 n=20+17)
TimeParse-12                 314ns ± 1%     314ns ± 0%    ~     (p=1.011 n=20+13)
TimeFormat-12                358ns ± 0%     354ns ± 0%  -1.12%  (p=0.000 n=17+20)
[Geo mean]                  53.5µs         53.3µs       -0.23%

Change-Id: I2a0a179d1d6bf7875dd054b7693dd12d2a340132
Reviewed-on: https://go-review.googlesource.com/23540
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
8 years agoruntime: clean up more traces of the old mark bit
Austin Clements [Fri, 2 Sep 2016 17:56:52 +0000 (13:56 -0400)]
runtime: clean up more traces of the old mark bit

Commit 59877bf renamed bitMarked to bitScan, since the bitmap is no
longer used for marking. However, there were several other references
to this strewn about comments and in some other constant names. Fix
these up, too.

Change-Id: I4183d28c6b01977f1d75a99ad55b150f2211772d
Reviewed-on: https://go-review.googlesource.com/28450
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
8 years agocmd/compile: remove nil check if followed by storezero on ARM64, MIPS64
Cherry Zhang [Tue, 6 Sep 2016 17:56:48 +0000 (13:56 -0400)]
cmd/compile: remove nil check if followed by storezero on ARM64, MIPS64

Change-Id: Ib90c92056fa70b27feb734837794ef53e842c41a
Reviewed-on: https://go-review.googlesource.com/28513
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
8 years agocmd/compile: remove ld/st-followed nil checks for PPC64
David Chase [Tue, 6 Sep 2016 16:05:02 +0000 (09:05 -0700)]
cmd/compile: remove ld/st-followed nil checks for PPC64

Enabled checks (except for DUFF-ops which aren't implemented yet).
Added ppc64le to relevant test.

Also updated register list to reflect no-longer-reserved-
for-constants status (file was missed in that change).

Updates #16010.

Change-Id: I31b1aac19e14994f760f2ecd02edbeb1f78362e7
Reviewed-on: https://go-review.googlesource.com/28548
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
8 years agocmd/link: remove outdated cast and comment
David Crawshaw [Tue, 6 Sep 2016 16:33:36 +0000 (12:33 -0400)]
cmd/link: remove outdated cast and comment

This program is written in Go now.

Change-Id: Ieec21a1bcac7c7a59e88cd1e1359977659de1757
Reviewed-on: https://go-review.googlesource.com/28549
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agoregexp: reduce mallocs in Regexp.Find* and Regexp.ReplaceAll*.
Aliaksandr Valialkin [Wed, 11 May 2016 11:57:24 +0000 (14:57 +0300)]
regexp: reduce mallocs in Regexp.Find* and Regexp.ReplaceAll*.

This improves Regexp.Find* and Regexp.ReplaceAll* speed:

name                  old time/op    new time/op    delta
Find-4                   345ns ± 1%     314ns ± 1%    -8.94%    (p=0.000 n=9+8)
FindString-4             341ns ± 1%     308ns ± 0%    -9.85%   (p=0.000 n=10+9)
FindSubmatch-4           440ns ± 1%     404ns ± 0%    -8.27%   (p=0.000 n=10+8)
FindStringSubmatch-4     426ns ± 0%     387ns ± 0%    -9.07%   (p=0.000 n=10+9)
ReplaceAll-4            1.75µs ± 1%    1.67µs ± 0%    -4.45%   (p=0.000 n=9+10)

name                  old alloc/op   new alloc/op   delta
Find-4                   16.0B ± 0%     0.0B ±NaN%  -100.00%  (p=0.000 n=10+10)
FindString-4             16.0B ± 0%     0.0B ±NaN%  -100.00%  (p=0.000 n=10+10)
FindSubmatch-4           80.0B ± 0%     48.0B ± 0%   -40.00%  (p=0.000 n=10+10)
FindStringSubmatch-4     64.0B ± 0%     32.0B ± 0%   -50.00%  (p=0.000 n=10+10)
ReplaceAll-4              152B ± 0%      104B ± 0%   -31.58%  (p=0.000 n=10+10)

name                  old allocs/op  new allocs/op  delta
Find-4                    1.00 ± 0%     0.00 ±NaN%  -100.00%  (p=0.000 n=10+10)
FindString-4              1.00 ± 0%     0.00 ±NaN%  -100.00%  (p=0.000 n=10+10)
FindSubmatch-4            2.00 ± 0%      1.00 ± 0%   -50.00%  (p=0.000 n=10+10)
FindStringSubmatch-4      2.00 ± 0%      1.00 ± 0%   -50.00%  (p=0.000 n=10+10)
ReplaceAll-4              8.00 ± 0%      5.00 ± 0%   -37.50%  (p=0.000 n=10+10)

Fixes #15643

Change-Id: I594fe51172373e2adb98d1d25c76ca2cde54ff48
Reviewed-on: https://go-review.googlesource.com/23030
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile: generate table of main symbol types
David Crawshaw [Fri, 26 Aug 2016 00:29:15 +0000 (20:29 -0400)]
cmd/compile: generate table of main symbol types

For each exported symbol in package main, add its name and type to
go.plugin.tabs symbol. This is used by the runtime when loading a
plugin to return a typed interface{} value.

Change-Id: I23c39583e57180acb8f7a74d218dae4368614f46
Reviewed-on: https://go-review.googlesource.com/27818
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
8 years agomath: fix sqrt regression on AMD64
Ilya Tocar [Thu, 1 Sep 2016 18:04:08 +0000 (21:04 +0300)]
math: fix sqrt regression on AMD64

1.7 introduced a significant regression compared to 1.6:

SqrtIndirect-4  2.32ns ± 0%  7.86ns ± 0%  +238.79%        (p=0.000 n=20+18)

This is caused by sqrtsd preserving upper part of destination register.
Which introduces dependency on previous  value of X0.
In 1.6 benchmark loop didn't use X0 immediately after call:

callq  *%rbx
movsd  0x8(%rsp),%xmm2
movsd  0x20(%rsp),%xmm1
addsd  %xmm2,%xmm1
mov    0x18(%rsp),%rax
inc    %rax
jmp    loop

In 1.7 however xmm0 is used just after call:

callq  *%rbx
mov    0x10(%rsp),%rcx
lea    0x1(%rcx),%rax
movsd  0x8(%rsp),%xmm0
movsd  0x18(%rsp),%xmm1

I've  verified that this is caused by dependency, by inserting
XORPS X0,X0 in the beginning of math.Sqrt, which puts performance back on 1.6 level.

Splitting SQRTSD mem,reg into:
MOVSD mem,reg
SQRTSD reg,reg

Removes dependency, because MOVSD (load version)
doesn't need to preserve upper part of a register.
And reg,reg operation is solved by renamer in CPU.

As a result of this change regression is gone:
SqrtIndirect-4  7.86ns ± 0%  2.33ns ± 0%  -70.36%  (p=0.000 n=18+17)

This also removes old Sqrt benchmarks, in favor of benchmarks measuring latency.
Only SqrtIndirect is kept, to show impact of this patch.

Change-Id: Ic7eebe8866445adff5bc38192fa8d64c9a6b8872
Reviewed-on: https://go-review.googlesource.com/28392
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agocmd/go: run mkalldocs.sh
Josh Bleecher Snyder [Tue, 6 Sep 2016 14:43:15 +0000 (07:43 -0700)]
cmd/go: run mkalldocs.sh

This should have happened as part of CL 28485.

Change-Id: I63cd31303e542ceaec3f4002c5573f186a1e9a52
Reviewed-on: https://go-review.googlesource.com/28547
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile: fix intrinsifying sync/atomic.Swap* on AMD64
Cherry Zhang [Tue, 6 Sep 2016 12:48:14 +0000 (08:48 -0400)]
cmd/compile: fix intrinsifying sync/atomic.Swap* on AMD64

It should alias to Xchg instead of Swap. Found when testing #16985.

Change-Id: If9fd734a1f89b8b2656f421eb31b9d1b0d95a49f
Reviewed-on: https://go-review.googlesource.com/28512
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agocmd/compile: mark some AMD64 atomic ops as clobberFlags
Cherry Zhang [Mon, 5 Sep 2016 03:51:22 +0000 (23:51 -0400)]
cmd/compile: mark some AMD64 atomic ops as clobberFlags

Fixes #16985.

Change-Id: I5954db28f7b70dd3ac7768e471d5df871a5b20f9
Reviewed-on: https://go-review.googlesource.com/28510
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
8 years agosyscall: add yet more TestGetfsstat debugging
Brad Fitzpatrick [Tue, 6 Sep 2016 04:11:59 +0000 (04:11 +0000)]
syscall: add yet more TestGetfsstat debugging

Updates #16937

Change-Id: I98aa203176f8f2ca2fcca6e334a65bc60d6f824d
Reviewed-on: https://go-review.googlesource.com/28535
Reviewed-by: Ian Lance Taylor <iant@golang.org>
8 years agoruntime: remove redundant expression from SetFinalizer
Erik Staab [Sat, 3 Sep 2016 22:57:48 +0000 (15:57 -0700)]
runtime: remove redundant expression from SetFinalizer

The previous if condition already checks the same expression and doesn't
have side effects.

Change-Id: Ieaf30a786572b608d0a883052b45fd3f04bc6147
Reviewed-on: https://go-review.googlesource.com/28475
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agoos: deduplicate File definition
Shenghou Ma [Mon, 5 Sep 2016 09:19:30 +0000 (05:19 -0400)]
os: deduplicate File definition

Fixes #16993.

Change-Id: Ibe406f97d2a49acae94531d969c56dbac8ce53b2
Reviewed-on: https://go-review.googlesource.com/28511
Run-TryBot: Minux Ma <minux@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agoio: make MultiReader nil exhausted Readers for earlier GC
Brad Fitzpatrick [Mon, 5 Sep 2016 20:16:16 +0000 (20:16 +0000)]
io: make MultiReader nil exhausted Readers for earlier GC

No test because the language spec makes no promises in this area.

Fixes #16983

Change-Id: I1a6aa7ff87dd14aa27e8400040a6f6fc908aa1fd
Reviewed-on: https://go-review.googlesource.com/28533
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
8 years agocmd/go: use httpGET helper in bug command
Caio Marcelo de Oliveira Filho [Mon, 5 Sep 2016 18:14:55 +0000 (15:14 -0300)]
cmd/go: use httpGET helper in bug command

Use existing helper function instead of importing "net/http". This
allows the go_bootstrap build to not depend on "net/http" package.
See cmd/go/http.go for details.

Fixes build bootstrap build with all.bash.

Change-Id: I2fd0fb01af7774f1690a353af22137680ec78170
Reviewed-on: https://go-review.googlesource.com/28531
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/go: add bug command
Josh Bleecher Snyder [Sun, 28 Aug 2016 17:38:41 +0000 (10:38 -0700)]
cmd/go: add bug command

This is a slightly rough, skeletal implementation.
We will polish and add to it through use.

.github/ISSUE_TEMPLATE will be updated in a
separate CL.

Fixes #16635

Change-Id: Icf284170d87e61b5b643366c85cffc48f149f730
Reviewed-on: https://go-review.googlesource.com/28485
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agosyscall: add more TestGetfsstat debugging
Brad Fitzpatrick [Mon, 5 Sep 2016 01:20:50 +0000 (01:20 +0000)]
syscall: add more TestGetfsstat debugging

Updates #16937

Change-Id: I6d1b210c741269b58040bd68bf3b51644f891737
Reviewed-on: https://go-review.googlesource.com/28487
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agocmd/compile: clean up ctxt params in sinit
Josh Bleecher Snyder [Sun, 4 Sep 2016 16:34:03 +0000 (09:34 -0700)]
cmd/compile: clean up ctxt params in sinit

The ctxt parameter is always set to 0 on entry into anylit so make this
parameter a literal constant, and where possibly remove ctxt as a parameter
where it is known to be a constant zero.

Passes toolstash -cmp.

This is a re-creation of CL 28221 by Dave Cheney.
That CL was graciously reverted in CL 28480
to make merging other CLs easier.

Change-Id: If7a57bf0e27774d9890adbc30af9fabb4aff1058
Reviewed-on: https://go-review.googlesource.com/28483
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Dave Cheney <dave@cheney.net>
8 years agopath/filepath: handle "C:." correctly in EvalSymlinks on Windows
Hiroshi Ioka [Fri, 26 Aug 2016 04:50:34 +0000 (13:50 +0900)]
path/filepath: handle "C:." correctly in EvalSymlinks on Windows

Fixes #16886

Change-Id: Idfacb0cf44d9994559c8e09032b4595887e76433
Reviewed-on: https://go-review.googlesource.com/28214
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agonet/http/httputil: t.Error -> t.Errorf
Josh Bleecher Snyder [Sun, 4 Sep 2016 21:52:59 +0000 (14:52 -0700)]
net/http/httputil: t.Error -> t.Errorf

Found by vet.

Change-Id: I09b79d68c7a5fc97e0edda4700a82bfbb00a4f45
Reviewed-on: https://go-review.googlesource.com/28486
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Hudson-Doyle <michael.hudson@canonical.com>
8 years agocmd/compile: simplify staticname
Josh Bleecher Snyder [Mon, 20 Jun 2016 15:18:22 +0000 (08:18 -0700)]
cmd/compile: simplify staticname

Add docs.
Give it a more natural signature.

Passes toolstash -cmp.

Change-Id: Iab368dd10e8f16e41b725c2980020bbead2cdefb
Reviewed-on: https://go-review.googlesource.com/26756
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile: document sinit ctxt and pass/kind arguments
Josh Bleecher Snyder [Sun, 19 Jun 2016 21:12:59 +0000 (14:12 -0700)]
cmd/compile: document sinit ctxt and pass/kind arguments

No functional changes. Passes toolstash -cmp.

Change-Id: I1ad467e574fd2ea80ab1459c0c943d9ff66c23ec
Reviewed-on: https://go-review.googlesource.com/26755
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
8 years agocmd/compile: unify arraylit and structlit
Josh Bleecher Snyder [Sun, 19 Jun 2016 19:11:47 +0000 (12:11 -0700)]
cmd/compile: unify arraylit and structlit

They were almost identical.
Merge them and some of their calling code.

Passes toolstash -cmp.

Change-Id: I9e92a864a6c09c9e18ed52dc247a678467e344ba
Reviewed-on: https://go-review.googlesource.com/26754
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
8 years agocmd/compile: add OSLICELIT
Josh Bleecher Snyder [Sun, 19 Jun 2016 14:20:28 +0000 (07:20 -0700)]
cmd/compile: add OSLICELIT

Does not pass toolstash -cmp due to changed export data,
but the cmd/go binary (which doesn't contain export data)
is bit-for-bit identical.

Change-Id: I6b12f9de18cf7da528e9207dccbf8f08c969f142
Reviewed-on: https://go-review.googlesource.com/26753
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
8 years agocmd/compile: refactor out KeepAlive
Michael Pratt [Sun, 4 Sep 2016 01:48:30 +0000 (18:48 -0700)]
cmd/compile: refactor out KeepAlive

Reduce the duplication in every arch by moving the code into package gc.

Change-Id: Ia111add8316492571825431ecd4f0154c8792ae1
Reviewed-on: https://go-review.googlesource.com/28481
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agocompress/flate: clarify the behavior of Writer.Flush
Joe Tsai [Sun, 4 Sep 2016 01:29:43 +0000 (18:29 -0700)]
compress/flate: clarify the behavior of Writer.Flush

Fixes #16068

Change-Id: I04e80a181c0b7356996f7a1158ea4895ff9e1e39
Reviewed-on: https://go-review.googlesource.com/28477
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agoRevert "cmd/compile/internal/gc: clean up sinit.go"
Dave Cheney [Sun, 4 Sep 2016 01:39:16 +0000 (01:39 +0000)]
Revert "cmd/compile/internal/gc: clean up sinit.go"

Revert to make josharians branch land cleanly

This reverts commit 38abd43b6a4d215375901d137a3eac9d0d3393a5.

Change-Id: Idde1df953baf6e5742c87c4edd4bee0b6b418aca
Reviewed-on: https://go-review.googlesource.com/28480
Run-TryBot: Dave Cheney <dave@cheney.net>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile: use CheckLoweredPhi on PPC64
Michael Pratt [Sun, 4 Sep 2016 01:32:51 +0000 (18:32 -0700)]
cmd/compile: use CheckLoweredPhi on PPC64

This custom version is identical to CheckLoweredPhi. The addition of
CheckLoweredPhi likely raced with adding PPC64.

Change-Id: I294dcb758d312e93fb8842f4d1e12bf0f63a1e06
Reviewed-on: https://go-review.googlesource.com/28479
Run-TryBot: Michael Pratt <mpratt@google.com>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agostrconv: fix function name in errors for Atoi
Joe Tsai [Sat, 3 Sep 2016 19:01:27 +0000 (12:01 -0700)]
strconv: fix function name in errors for Atoi

Fixes #16980

Change-Id: I902a02b157c2c7d1772f5122b850dc48b1d7a224
Reviewed-on: https://go-review.googlesource.com/28474
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agotest: add test for issue 15895
Josh Bleecher Snyder [Sat, 3 Sep 2016 23:31:56 +0000 (16:31 -0700)]
test: add test for issue 15895

It was fixed earlier in the Go 1.8 cycle.
Add a test.

Fixes #15895

Change-Id: I5834831235d99b9fcf21b435932cdd7ac6dc2c6e
Reviewed-on: https://go-review.googlesource.com/28476
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agounicode/utf8: reduce bounds checks in EncodeRune
Martin Möhrmann [Sat, 3 Sep 2016 16:39:25 +0000 (18:39 +0200)]
unicode/utf8: reduce bounds checks in EncodeRune

Provide bounds elim hints in EncodeRune.

name                  old time/op  new time/op  delta
EncodeASCIIRune-4     2.69ns ± 2%  2.69ns ± 2%    ~     (p=0.193 n=47+46)
EncodeJapaneseRune-4  5.97ns ± 2%  5.38ns ± 2%  -9.93%  (p=0.000 n=49+50)

Change-Id: I1a6dcffff3bdd64ab93c2130021e3b00981de4c8
Reviewed-on: https://go-review.googlesource.com/28492
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile/internal/gc: clean up sinit.go
Dave Cheney [Wed, 31 Aug 2016 04:39:41 +0000 (14:39 +1000)]
cmd/compile/internal/gc: clean up sinit.go

The ctxt parameter is always set to 0 on entry into anylit so make this
parameter a literal constant, and where possibly remove ctxt as a parameter
where it is known to be a constant zero.

Change-Id: I3e76e06456d7b1a1ea875ffeb2efefa4a1ff5a7e
Reviewed-on: https://go-review.googlesource.com/28221
Run-TryBot: Dave Cheney <dave@cheney.net>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agonet/http/httputil: make ReverseProxy send nil Body requests when possible
Brad Fitzpatrick [Fri, 2 Sep 2016 05:00:05 +0000 (05:00 +0000)]
net/http/httputil: make ReverseProxy send nil Body requests when possible

The http.Transport's retry can't retry requests with non-nil
bodies. When cloning an incoming server request into an outgoing
client request, nil out the Body field if the ContentLength is 0. (For
server requests, Body is always non-nil, even for GET, HEAD, etc.)

Also, don't use the deprecated CancelRequest and use Context instead.

And don't set Proto, ProtoMajor, ProtoMinor. Those are ignored in
client requests, which was probably a later documentation
clarification.

Fixes #16036
Updates #16696 (remove useless Proto lines)

Change-Id: I70a869e9bd4bf240c5838e82fb5aa695a539b343
Reviewed-on: https://go-review.googlesource.com/28412
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
8 years agoarchive/tar: reapply Header.Size to regFileReader after merging
Joe Tsai [Fri, 2 Sep 2016 21:37:35 +0000 (14:37 -0700)]
archive/tar: reapply Header.Size to regFileReader after merging

The use of PAX headers can modify the overall file size, thus the
formerly created regFileReader may be stale.

The relevant PAX specification for this behavior is:
<<<
Any fields in the preceding optional extended header shall override
the associated fields in this header block for this file.
>>>
Where "optional extended header" refers to the preceding PAX header.
Where "this header block" refers to the subsequent USTAR header.

Fixes #15573
Fixes #15564

Change-Id: I83b1c3f05a9ca2d3be38647425ad21a9fe450ee2
Reviewed-on: https://go-review.googlesource.com/28418
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/compile: fix argument for given format verb
Robert Griesemer [Fri, 2 Sep 2016 21:21:57 +0000 (14:21 -0700)]
cmd/compile: fix argument for given format verb

Follow-up to https://go-review.googlesource.com/28394.

Change-Id: Ic4147e9ae786a4de0a3454131fac03e940ae2e76
Reviewed-on: https://go-review.googlesource.com/28417
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agopath/filepath: use new style deprecation message
Jaana Burcu Dogan [Fri, 2 Sep 2016 06:28:26 +0000 (23:28 -0700)]
path/filepath: use new style deprecation message

Change-Id: I242a8960583e333f372929aad4adb8efbe441cd4
Reviewed-on: https://go-review.googlesource.com/28413
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Jaana Burcu Dogan <jbd@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agoruntime: remove a load and shift from scanobject
Josh Bleecher Snyder [Mon, 2 May 2016 01:02:41 +0000 (18:02 -0700)]
runtime: remove a load and shift from scanobject

hbits.morePointers and hbits.isPointer both
do a load and a shift. Do it only once.

Benchmarks using compilebench (because it is
the benchmark I have the most tooling around),
on a quiet machine.

name       old time/op      new time/op      delta
Template        291ms ±14%       290ms ±15%    ~          (p=0.702 n=100+99)
Unicode         143ms ± 9%       142ms ± 9%    ~           (p=0.126 n=99+98)
GoTypes         934ms ± 4%       933ms ± 4%    ~         (p=0.937 n=100+100)
Compiler        4.92s ± 2%       4.90s ± 1%  -0.28%        (p=0.003 n=98+98)

name       old user-ns/op   new user-ns/op   delta
Template   360user-ms ± 5%  355user-ms ± 4%  -1.37%        (p=0.000 n=97+96)
Unicode    178user-ms ± 6%  176user-ms ± 6%  -1.24%        (p=0.001 n=96+99)
GoTypes    1.22user-s ± 5%  1.21user-s ± 5%  -0.94%      (p=0.000 n=100+100)
Compiler   6.50user-s ± 2%  6.44user-s ± 3%  -0.94%        (p=0.000 n=96+98)

On amd64, before:

"".scanobject t=1 size=581 args=0x10 locals=0x78

After:

"".scanobject t=1 size=540 args=0x10 locals=0x78

Change-Id: I420ac3704549d484a5d85e19fea82c85da389514
Reviewed-on: https://go-review.googlesource.com/22712
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agoruntime: fix global buffer reset in StopTrace
Dmitry Vyukov [Mon, 8 Aug 2016 14:17:14 +0000 (16:17 +0200)]
runtime: fix global buffer reset in StopTrace

We reset global buffer only if its pos != 0.
We ought to do it always, but queue it only if pos != 0.
This is a latent bug. Currently it does not fire because
whenever we create a global buffer, we increment pos.

Change-Id: I01e28ae88ce9a5412497c524391b8b7cb443ffd9
Reviewed-on: https://go-review.googlesource.com/25574
Run-TryBot: Dmitry Vyukov <dvyukov@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
8 years agoruntime: rename variable
Gleb Stepanov [Mon, 25 Jul 2016 13:25:44 +0000 (16:25 +0300)]
runtime: rename variable

Rename variable to bitScan according to
TODO comment.

Change-Id: I81dd8cc1ca28c0dc9308a654ad65cdf5b2fd2ce3
Reviewed-on: https://go-review.googlesource.com/25175
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
8 years agoruntime: improve message when a bad pointer is found on the stack
Austin Clements [Fri, 2 Sep 2016 16:14:38 +0000 (12:14 -0400)]
runtime: improve message when a bad pointer is found on the stack

Currently this message says "invalid stack pointer", which could be
interpreted as the value of SP being invalid. Change it to "invalid
pointer found on stack" to emphasize that it's a pointer on the stack
that's invalid.

Updates #16948.

Change-Id: I753624f8cc7e08cf13d3ea5d9c790cc4af9fa372
Reviewed-on: https://go-review.googlesource.com/28430
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agocrypto/cipher: enforce message size limits for GCM.
Adam Langley [Thu, 1 Sep 2016 23:00:25 +0000 (16:00 -0700)]
crypto/cipher: enforce message size limits for GCM.

The maximum input plaintext for GCM is 64GiB - 64. Since the GCM
interface is one-shot, it's very hard to hit this in Go (one would need
a 64GiB buffer in memory), but we should still enforce this limit.

Thanks to Quan Nguyen for pointing it out.

Change-Id: Icced47bf8d4d5dfbefa165cf13e893205c9577b8
Reviewed-on: https://go-review.googlesource.com/28410
Run-TryBot: Adam Langley <agl@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Andrew Gerrand <adg@golang.org>
8 years agonet/http/httputil: remove proxied headers mentioned in connection-tokens
Sina Siadat [Sat, 27 Aug 2016 16:16:25 +0000 (20:46 +0430)]
net/http/httputil: remove proxied headers mentioned in connection-tokens

RFC 2616, section 14.10 says:

>>>
HTTP/1.1 proxies MUST parse the Connection header field before a message
is forwarded and, for each connection-token in this field, remove any
header field(s) from the message with the same name as the
connection-token. Connection options are signaled by the presence of a
connection-token in the Connection header field, not by any
corresponding additional header field(s), since the additional header
field may not be sent if there are no parameters associated with that
connection option.
<<<

The same requirement was included in RFC 7230, section 6.1.

Fixes #16875

Change-Id: I57ad4a4a17775537c8810d0edd7de1604317b5fa
Reviewed-on: https://go-review.googlesource.com/27970
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agomath/rand: document that NewSource sources race
David Glasser [Thu, 4 Aug 2016 23:49:05 +0000 (16:49 -0700)]
math/rand: document that NewSource sources race

While it was previously explicitly documented that "the default Source"
is safe for concurrent use, a careless reader can interpret that as
meaning "the implementation of the Source interface created by functions
in this package" rather than "the default shared Source used by
top-level functions". Be explicit that the Source returned by NewSource
is not safe for use by multiple goroutines.

Fixes #3611.

Change-Id: Iae4bc04c3887ad6e2491e36e38feda40324022c5
Reviewed-on: https://go-review.googlesource.com/25501
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
8 years agocmd/dist: make test names consistent
Josh Bleecher Snyder [Wed, 31 Aug 2016 01:11:10 +0000 (18:11 -0700)]
cmd/dist: make test names consistent

Current banners are:

##### Building Go bootstrap tool.
##### Building Go toolchain using /Users/josh/go/1.4.
##### Building go_bootstrap for host, darwin/amd64.
##### Building packages and commands for darwin/amd64.
##### Testing packages.
##### GOMAXPROCS=2 runtime -cpu=1,2,4
##### Testing without libgcc.
##### sync -cpu=10
##### ../misc/cgo/stdio
##### ../misc/cgo/life
##### ../misc/cgo/fortran
##### ../misc/cgo/test
##### Testing race detector
##### ../misc/cgo/testso
##### ../misc/cgo/testsovar
##### misc/cgo/testcarchive
##### ../misc/cgo/testcshared
##### ../misc/cgo/errors
##### ../test/bench/go1
##### ../test
##### API check

One of these things is not like the others.
Fix that.

Change-Id: If0bd8ea9293d73b5d1b70d6bf676bd9192991505
Reviewed-on: https://go-review.googlesource.com/26759
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agogo/doc: allow ToHTML to properly handle URLs containing semicolons
Matt Layher [Mon, 1 Aug 2016 16:50:11 +0000 (12:50 -0400)]
go/doc: allow ToHTML to properly handle URLs containing semicolons

Fixes #16565

Change-Id: I3edfd2576a7ca5270644a4e7f126854f821f2c9a
Reviewed-on: https://go-review.googlesource.com/25385
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agonet/http: omit Content-Length in Response.Write for 1xx or 204 status
Matt Layher [Thu, 1 Sep 2016 20:30:15 +0000 (16:30 -0400)]
net/http: omit Content-Length in Response.Write for 1xx or 204 status

Per RFC 7230, Section 3.3.2: "A server MUST NOT send a Content-Length
header field in any response with a status code of 1xx (Informational)
or 204 (No Content).".

Fixes #16942

Change-Id: I8006c76c126304e13618966e6eafb08a3885d3cd
Reviewed-on: https://go-review.googlesource.com/28351
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

8 years agocmd/compile: disallow typed non-integer constant len and cap make arguments
Martin Möhrmann [Thu, 1 Sep 2016 06:31:37 +0000 (08:31 +0200)]
cmd/compile: disallow typed non-integer constant len and cap make arguments

make(T, n, m) returns a slice of type T with length n and capacity m
where "The size arguments n and m must be of integer type or untyped."
https://tip.golang.org/ref/spec#Making_slices_maps_and_channels

The failure to reject typed non-integer size arguments in make
during compile time was uncovered after https://golang.org/cl/27851
changed the generation of makeslice calls.

Fixes   #16940
Updates #16949

Change-Id: Ib1e3576f0e6ad199c9b16b7a50c2db81290c63b4
Reviewed-on: https://go-review.googlesource.com/28301
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
8 years agocmd/compile: fix missing format verb
Josh Bleecher Snyder [Thu, 1 Sep 2016 19:23:43 +0000 (12:23 -0700)]
cmd/compile: fix missing format verb

Found by vet.

Change-Id: I50420771678b1a3695348ce1a81f410479ed09a1
Reviewed-on: https://go-review.googlesource.com/28394
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>