]> Cypherpunks repositories - gostls13.git/log
gostls13.git
7 years agonet/http: deep copy Request.URL also in Request.WithContext's copy
Emmanuel Odeke [Fri, 21 Apr 2017 17:49:19 +0000 (11:49 -0600)]
net/http: deep copy Request.URL also in Request.WithContext's copy

Despite the previously known behavior of Request.WithContext
shallow copying a request, usage of the request inside server.ServeHTTP
mutates the request's URL. This CL implements deep copying of the URL.

Fixes #20068

Change-Id: I86857d7259e23ac624d196401bf12dde401c42af
Reviewed-on: https://go-review.googlesource.com/41308
Run-TryBot: Emmanuel Odeke <emm.odeke@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agonet/http: permit incoming CONNECT requests without Host headers
Brad Fitzpatrick [Tue, 23 May 2017 23:42:47 +0000 (23:42 +0000)]
net/http: permit incoming CONNECT requests without Host headers

Apparently they exist in the wild. See:
https://github.com/golang/go/issues/18215#issuecomment-301182496
(Facebook / iOS)

Fixes #18215

Change-Id: I9ddad3896b5d784cb3f5b3ee9c6819081a4a2702
Reviewed-on: https://go-review.googlesource.com/44004
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matt Layher <mdlayher@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agostrings: simplify indexFunc
Michael Darakananda [Tue, 8 Nov 2016 06:54:30 +0000 (17:54 +1100)]
strings: simplify indexFunc

A for-range loop is simpler and also generally faster nowadays:

TrimASCII/1:1-4      69.2ns ± 1%  72.3ns ± 4%  +4.55%  (p=0.001 n=8+8)
TrimASCII/1:2-4       114ns ± 4%   104ns ± 3%  -8.71%  (p=0.000 n=9+8)
TrimASCII/1:4-4       112ns ± 1%   109ns ± 2%  -2.57%  (p=0.000 n=8+9)
TrimASCII/1:8-4       120ns ± 2%   118ns ± 4%    ~     (p=0.097 n=9+9)
TrimASCII/1:16-4      137ns ± 3%   132ns ± 3%  -3.82%  (p=0.001 n=9+9)
TrimASCII/16:1-4      129ns ± 1%   125ns ± 2%  -3.38%  (p=0.000 n=8+9)
TrimASCII/16:2-4      167ns ± 3%   159ns ± 1%  -4.99%  (p=0.000 n=9+8)
TrimASCII/16:4-4      165ns ± 2%   162ns ± 1%  -1.91%  (p=0.005 n=8+9)
TrimASCII/16:8-4      173ns ± 2%   170ns ± 1%  -1.29%  (p=0.018 n=9+9)
TrimASCII/16:16-4     188ns ± 2%   186ns ± 2%  -1.13%  (p=0.022 n=8+9)
TrimASCII/256:1-4    1.06µs ± 1%  0.98µs ± 2%  -7.64%  (p=0.000 n=8+9)
TrimASCII/256:2-4    1.08µs ± 1%  1.06µs ± 2%  -1.95%  (p=0.006 n=9+9)
TrimASCII/256:4-4    1.09µs ± 1%  1.07µs ± 3%    ~     (p=0.059 n=9+9)
TrimASCII/256:8-4    1.10µs ± 1%  1.07µs ± 2%  -2.63%  (p=0.000 n=9+8)
TrimASCII/256:16-4   1.10µs ± 1%  1.08µs ± 1%  -1.90%  (p=0.000 n=8+9)
TrimASCII/4096:1-4   15.8µs ± 1%  14.5µs ± 1%  -8.59%  (p=0.000 n=9+9)
TrimASCII/4096:2-4   15.6µs ± 1%  15.4µs ± 2%  -1.27%  (p=0.021 n=8+8)
TrimASCII/4096:4-4   15.6µs ± 1%  15.4µs ± 2%    ~     (p=0.094 n=9+9)
TrimASCII/4096:8-4   15.7µs ± 1%  15.8µs ± 6%    ~     (p=0.555 n=8+8)
TrimASCII/4096:16-4  15.7µs ± 2%  15.3µs ± 1%  -2.64%  (p=0.000 n=8+9)

Change-Id: I9b06689b67c0cf2c7ff446fc63a8c44cc5d6a246
Reviewed-on: https://go-review.googlesource.com/32891
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/dist: ensure android-implies-linux for file names in dist
David Chase [Mon, 20 Mar 2017 18:24:00 +0000 (14:24 -0400)]
cmd/dist: ensure android-implies-linux for file names in dist

This is one hurdle to building Go on Android; the runtime does
not build properly because *_linux.go files are excluded from
the "Building go_bootstrap" step when GOOS=android.

There are other hurdles; this is the first one.

Change-Id: I766e4bbf6ffc0d273888913f2516cf3e995a1786
Reviewed-on: https://go-review.googlesource.com/38308
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agonet/http: polish gzip case insensitive test
Emmanuel Odeke [Wed, 24 May 2017 01:40:52 +0000 (19:40 -0600)]
net/http: polish gzip case insensitive test

Avoid directly using the binary of the
gzipped encoded string in the handler.

Follow up of CL 37431.

Change-Id: Idcd04acb7940e67b7a35b2d6cb163d75b0e22e04
Reviewed-on: https://go-review.googlesource.com/44008
Run-TryBot: Emmanuel Odeke <emm.odeke@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agonet/http: make Transport respect non lower case Content-Encoding
Jan Berktold [Thu, 23 Feb 2017 14:53:27 +0000 (15:53 +0100)]
net/http: make Transport respect non lower case Content-Encoding

The existing Transport implementation does not detect gzip encoding
when the Content-Encoding header is not lower-case. This is not
compliant with RFC2616 section 3.5 "All content-coding values are
case-insensitive." and caused issues in the wild.

Fixes #19248

Change-Id: I1b49992832dc3c8ef700058596a27dd9909640a3
Reviewed-on: https://go-review.googlesource.com/37431
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agovendor: update golang.org/x/net/http2/hpack
Brad Fitzpatrick [Wed, 24 May 2017 00:36:23 +0000 (00:36 +0000)]
vendor: update golang.org/x/net/http2/hpack

Updates golang.org/x/net/http2/hpack to x/net git rev 4dbf598 for:

   http2/hpack: remove unused pair function from package
   https://golang.org/cl/43851

   http2/hpack: remove pair function and initialize directly
   https://golang.org/cl/43190

   http2/hpack: move initialization to a static table
   https://golang.org/cl/43090

Updates #6853

Change-Id: I933a094623143c7333a0da7867d5d2e41da05234
Reviewed-on: https://go-review.googlesource.com/44007
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Tom Bergan <tombergan@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agonet/http: update bundled x/net/http2
Brad Fitzpatrick [Wed, 24 May 2017 00:05:04 +0000 (00:05 +0000)]
net/http: update bundled x/net/http2

This updates the bundled copy of x/net/http2 to x/net git rev a8e8f92cd6 for:

    http2: remove extra goroutine stack from awaitGracefulShutdown
    https://golang.org/cl/43230

    http2: Discard DATA frames from the server after the response body is closed
    https://golang.org/cl/43810

Fixes #20302
Fixes #18471
Fixes #20448

Change-Id: I00972836deb2fe6049f631ee44901732a641b171
Reviewed-on: https://go-review.googlesource.com/44006
Reviewed-by: Tom Bergan <tombergan@google.com>
7 years agonet/http: Add Server.RegisterOnShutdown
Tom Bergan [Tue, 23 May 2017 23:03:21 +0000 (16:03 -0700)]
net/http: Add Server.RegisterOnShutdown

This will be used to allow http2 servers to register a shutdown function
so that net/http.Server.Shutdown will work when the http2 server is
configured via a manual call to http2.ConfigureServer. Currently, Shutdown
only works when the http2 server is configured automatically by the
net/http package.

Updates #20302
Updates #18471

Change-Id: Ifc2b5f3126126a106b49ea4a7e999279852b9cc9
Reviewed-on: https://go-review.googlesource.com/44003
Run-TryBot: Tom Bergan <tombergan@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agonet/http/httptest: make ResponseRecorder.Result.Status match http.Transport
Brad Fitzpatrick [Tue, 23 May 2017 22:35:15 +0000 (22:35 +0000)]
net/http/httptest: make ResponseRecorder.Result.Status match http.Transport

Fixes #18438

Change-Id: I9599c1536d5e8bad7662b8ffa19e9b0746e27e60
Reviewed-on: https://go-review.googlesource.com/44000
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agomime/multipart: parse boundary with spaces properly
Michael Fraenkel [Tue, 24 Jan 2017 14:40:17 +0000 (07:40 -0700)]
mime/multipart: parse boundary with spaces properly

- spaces are allowed anywhere but the last character of a boundary

Fixes #18768

Change-Id: I36b054462533ff6dfc060e37e7a58777ae4b66fe
Reviewed-on: https://go-review.googlesource.com/35507
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agoruntime: use pselect6 for usleep on linux/386
Austin Clements [Tue, 23 May 2017 21:54:24 +0000 (17:54 -0400)]
runtime: use pselect6 for usleep on linux/386

Commit 4dcba023c6 replaced select with pselect6 on linux/amd64 and
linux/arm, but it turns out the Android emulator uses linux/386. This
makes the equivalent change there, too.

Fixes #20409 more.

Change-Id: If542d6ade06309aab8758d5f5f6edec201ca7670
Reviewed-on: https://go-review.googlesource.com/44011
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/cgo: support indirect macro expansion for string
Hiroshi Ioka [Fri, 21 Apr 2017 10:18:36 +0000 (19:18 +0900)]
cmd/cgo: support indirect macro expansion for string

Current code cannot handle string #define macros if those macros are
defined via other macros. This CL solve the issue.

Updates #18720

Change-Id: Ibed0773d10db3d545bb246b97e81c0d19e3af3d5
Reviewed-on: https://go-review.googlesource.com/41312
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/go: include GOARM and GO386 in computed build ID
Brad Fitzpatrick [Tue, 23 May 2017 05:22:25 +0000 (05:22 +0000)]
cmd/go: include GOARM and GO386 in computed build ID

Now:
$ GOARCH=arm GOARM=5 go install -x cmd/go
... followed by:
$ GOARCH=arm GOARM= go install -x cmd/go

... actually does work. Previously the second "go install" would reuse
the cached binaries from the GOARM=5 command and not rebuild.
(Or vice versa from GOARM= to GOARM=5)

And do the same for GO386.

Fixes #9737

Change-Id: I9630aab34d06465d5033e6743dfe6592c8247aa0
Reviewed-on: https://go-review.googlesource.com/43855
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocrypto/rand: only read necessary bytes for Int
Wade Simmons [Tue, 23 May 2017 00:31:17 +0000 (20:31 -0400)]
crypto/rand: only read necessary bytes for Int

We only need to read the number of bytes required to store the value
"max - 1" to generate a random number in the range [0, max).

Before, there was an off-by-one error where an extra byte was read from
the io.Reader for inputs like "256" (right at the boundary for a byte).
There was a similar off-by-one error in the logic for clearing bits and
thus for any input that was a power of 2, there was a 50% chance the
read would continue to be retried as the mask failed to remove a bit.

Fixes #18165.

Change-Id: I548c1368990e23e365591e77980e9086fafb6518
Reviewed-on: https://go-review.googlesource.com/43891
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoruntime: remove unused copies of special stack guards
Austin Clements [Thu, 18 May 2017 18:50:12 +0000 (14:50 -0400)]
runtime: remove unused copies of special stack guards

There are two copies each of the stackPreempt/_StackPreempt and
stackFork/_StackFork constants. Remove the ones left over from C that
are no longer used.

Change-Id: I849604c72c11e4a0cb08e45e9817eb3f5a6ce8ba
Reviewed-on: https://go-review.googlesource.com/43638
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agoruntime: fix stackCache=0 debug mode
Austin Clements [Thu, 18 May 2017 18:35:53 +0000 (14:35 -0400)]
runtime: fix stackCache=0 debug mode

Setting stackCache to 0 to disable stack caches for debugging hasn't
worked for a long time. It causes stackalloc to fall back to full span
allocation, round sub-page stacks down to 0 pages, and blow up.

Fix this debug mode so it disables the per-P caches, but continues to
use the global stack pools for small stacks, which correctly handle
sub-page stacks. While we're here, rename stackCache to stackNoCache
so it acts like the rest of the stack allocator debug modes where "0"
is the right default value.

Fixes #17291.

Change-Id: If401c41cee3448513cbd7bb2e9334a8efab257a7
Reviewed-on: https://go-review.googlesource.com/43637
Reviewed-by: Keith Randall <khr@golang.org>
7 years agoruntime: fix stackFromSystem returning memory
Austin Clements [Thu, 18 May 2017 17:59:00 +0000 (13:59 -0400)]
runtime: fix stackFromSystem returning memory

The stackFromSystem debug mode has two problems:

1) It rounds the stack allocation to _PageSize. If the physical page
size is >8K, this can cause unmapping the memory later to either
under-unmap or over-unmap.

2) It doesn't return the rounded-up allocation size to its caller, so
when we later unmap the memory, we may pass the wrong length.

Fix these problems by rounding the size up to the physical page size
and putting that rounded-up size in the returned stack bounds.

Fixes #17289.

Change-Id: I6b854af3b06bb16e3750798397bb5e2a722ec1cb
Reviewed-on: https://go-review.googlesource.com/43636
Reviewed-by: Keith Randall <khr@golang.org>
7 years agodoc: document go1.8.2 and go1.7.6
Chris Broadfoot [Tue, 23 May 2017 17:36:28 +0000 (10:36 -0700)]
doc: document go1.8.2 and go1.7.6

Change-Id: I2ed2e8c4890a65288cf3066ebe3c1d9a16fb4c05
Reviewed-on: https://go-review.googlesource.com/43990
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agoruntime: don't corrupt arena bounds on low mmap
Austin Clements [Mon, 22 May 2017 19:53:49 +0000 (15:53 -0400)]
runtime: don't corrupt arena bounds on low mmap

If mheap.sysAlloc doesn't have room in the heap arena for an
allocation, it will attempt to map more address space with sysReserve.
sysReserve is given a hint, but can return any unused address range.
Currently, mheap.sysAlloc incorrectly assumes the returned region will
never fall between arena_start and arena_used. If it does,
mheap.sysAlloc will blindly accept the new region as the new
arena_used and arena_end, causing these to decrease and make it so any
Go heap above the new arena_used is no longer considered part of the
Go heap. This assumption *used to be* safe because we had all memory
between arena_start and arena_used mapped, but when we switched to an
arena_start of 0 on 32-bit, it became no longer safe.

Most likely, we've only recently seen this bug occur because we
usually start arena_used just above the binary, which is low in the
address space. Hence, the kernel is very unlikely to give us a region
before arena_used.

Since mheap.sysAlloc is a linear allocator, there's not much we can do
to handle this well. Hence, we fix this problem by simply rejecting
the new region if it isn't after arena_end. In this case, we'll take
the fall-back path and mmap a small region at any address just for the
requested memory.

Fixes #20259.

Change-Id: Ib72e8cd621545002d595c7cade1e817cfe3e5b1e
Reviewed-on: https://go-review.googlesource.com/43870
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
7 years agocrypto/rand: use blocking getrandom call on Linux when supported
Joe Richey joerichey@google.com [Mon, 22 May 2017 21:36:43 +0000 (14:36 -0700)]
crypto/rand: use blocking getrandom call on Linux when supported

By changing getRandomLinux to immediately use the getrandom() syscall
without GRND_NONBLOCK, we now only fall back to reading from
/dev/urandom on Linux if the kernel does not support the getrandom()
syscall. This means reads for crypto/rand will now block if the kernel
has insufficient entropy on Linux kernels after v3.16.

Before, if the kernel had insufficient entropy, it would fall back to
reading from /dev/urandom. This would potentially return predictable
data.

Fixes #19274

Change-Id: I1cb081ce2f3096f18ad2820e52ecdbd993dc2afc
Reviewed-on: https://go-review.googlesource.com/43852
Reviewed-by: Filippo Valsorda <hi@filippo.io>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoos/exec: ignore some pipe write errors on windows
Alex Brainman [Mon, 22 May 2017 07:17:39 +0000 (17:17 +1000)]
os/exec: ignore some pipe write errors on windows

This change is windows version of CL 12152.
It also extends test to cover scenarios reported on issue #20445.
Some source files copied and renamed to make code clearer.

Fixes #20445

Change-Id: Idd2f636f27c6bd5cfe98017ba2df911358263382
Reviewed-on: https://go-review.googlesource.com/43910
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agoencoding/base32: add Encoding.WithPadding, StdPadding, NoPadding
Gustav Westling [Fri, 24 Mar 2017 23:35:40 +0000 (00:35 +0100)]
encoding/base32: add Encoding.WithPadding, StdPadding, NoPadding

Fixes #19478

Change-Id: I9fc186610d79fd003e7b5d88c0955286ebe7d3cf
Reviewed-on: https://go-review.googlesource.com/38634
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agolog: Prevent getting time if it's unnecessary
Albert Nigmatzianov [Mon, 22 May 2017 19:39:38 +0000 (21:39 +0200)]
log: Prevent getting time if it's unnecessary

Small performance gain:
name              old time/op    new time/op    delta
Itoa-4              95.4ns ± 4%    95.6ns ± 3%    ~     (p=0.256 n=45+46)
Println-4            480ns ± 4%     476ns ± 5%  -0.87%  (p=0.003 n=45+45)
PrintlnNoFlags-4     316ns ± 3%     299ns ± 4%  -5.38%  (p=0.000 n=42+44)

name              old alloc/op   new alloc/op   delta
Itoa-4               0.00B          0.00B         ~     (all equal)
Println-4            21.0B ± 0%     21.0B ± 0%    ~     (all equal)
PrintlnNoFlags-4     21.0B ± 0%     21.0B ± 0%    ~     (all equal)

name              old allocs/op  new allocs/op  delta
Itoa-4                0.00           0.00         ~     (all equal)
Println-4             2.00 ± 0%      2.00 ± 0%    ~     (all equal)
PrintlnNoFlags-4      2.00 ± 0%      2.00 ± 0%    ~     (all equal)

Change-Id: Idcd03609a5a437a69ffa7004a673bf0b8d22e7ad
Reviewed-on: https://go-review.googlesource.com/38056
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

7 years agocmd/go: warn on -race with -covermode=set.
Dhananjay Nakrani [Sat, 20 May 2017 02:52:59 +0000 (19:52 -0700)]
cmd/go: warn on -race with -covermode=set.

Fixes #20435.

Change-Id: I15576f36b26d01642c1187325baea82d3077e578
Reviewed-on: https://go-review.googlesource.com/43777
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/go: Document that -cover causes incorrect line numbers
Steven Hartland [Mon, 27 Mar 2017 13:28:19 +0000 (13:28 +0000)]
cmd/go: Document that -cover causes incorrect line numbers

Due to the fact that -cover injects additional code to the original
source, tests run with -cover will often have incorrect line numbers.

Also includes docs for -list regexp missed by ba8ff87

Updates #6329

Change-Id: I87f0618ac31e96071bca61055cc17c0cbdee208a
Reviewed-on: https://go-review.googlesource.com/38640
Reviewed-by: Rob Pike <r@golang.org>
7 years agonet/http/cookiejar: increase test coverage
Volker Dobler [Wed, 22 Mar 2017 08:14:41 +0000 (09:14 +0100)]
net/http/cookiejar: increase test coverage

The jarKey function handles broken PublicSuffixList implementations but
no test verified it.

Change-Id: Ifb76de9e8c3941f3b08d3e43970056e023013457
Reviewed-on: https://go-review.googlesource.com/38357
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: report correct position in redeclaration errors
Robert Griesemer [Mon, 22 May 2017 18:20:34 +0000 (11:20 -0700)]
cmd/compile: report correct position in redeclaration errors

When restoring syms, we must also restore the original Lastlineno.
Bug introduced with https://golang.org/cl/41390/.

Fixes #20415.

Change-Id: Ie81d36279d717e330951b52f42dcee4b0025b9f0
Reviewed-on: https://go-review.googlesource.com/43811
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agonet/http: send more cookie values in double quotes
Volker Dobler [Wed, 22 Feb 2017 13:40:17 +0000 (14:40 +0100)]
net/http: send more cookie values in double quotes

According to RFC 6255 a cookie value may contain neither spaces " "
nor commas ",". But browsers seem to handle these pretty well and such
values are not uncommon in the wild so we do allow spaces and commas
in cookie values too. Up to now we use the double-quoted wire format
only for cookie values with leading and/or trailing spaces and commas.
Values with internal spaces/commas are sent without the optional double
quotes. This seems to be a problem for some agents.

This CL changes the behaviour for cookie values with spaces or commas:
Such values are always sent in double quotes. This should not have
any impact on existing agents and the increases of data transmitted
is negligible.

Fixes #18627

Change-Id: I575a98d589e048aa39d976a3c984550daaca730a
Reviewed-on: https://go-review.googlesource.com/37328
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agonet/http/httputil: ReverseProxy should pass on unannounced Trailers
Tristan Colgate [Sat, 20 May 2017 13:50:06 +0000 (14:50 +0100)]
net/http/httputil: ReverseProxy should pass on unannounced Trailers

Trailers that are not announced in the Trailer must be passed on to
the downstream client.

Rather than iterate over each and find missing trailer values,
this re-adds all trailers to the headers if there is a disparity
between the number of announced trailers and the final number.

This fixes #20437

Change-Id: I867e85f45feff68616a9a9bd6f65f12d73825eb7
Reviewed-on: https://go-review.googlesource.com/43712
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agodoc: mention that go tools is needed to access some tool
Alberto Donizetti [Mon, 22 May 2017 12:00:09 +0000 (14:00 +0200)]
doc: mention that go tools is needed to access some tool

Change-Id: I020cd3d10a441ba4047800fdf4f93433c458398a
Reviewed-on: https://go-review.googlesource.com/43717
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agodatabase/sql: allow Stmt Query and Exec methods to open a new conn
Daniel Theophanes [Mon, 22 May 2017 02:51:46 +0000 (19:51 -0700)]
database/sql: allow Stmt Query and Exec methods to open a new conn

Query and Exec functions on DB first attempt to get a cached
connection before requesting the connection pool to ignore
the cache and get a new connection. This change aligns Stmt to
that behavior as well.

Fixes #20433

Change-Id: Idda5f61927289d7ad0882effa3a50ffc9efd88e6
Reviewed-on: https://go-review.googlesource.com/43790
Run-TryBot: Daniel Theophanes <kardianos@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agomime/multipart: Allow ReadForm to process large non-file parts
Steven Hartland [Thu, 16 Mar 2017 12:08:35 +0000 (12:08 +0000)]
mime/multipart: Allow ReadForm to process large non-file parts

Allow the memory limit passed into ReadForm to be used as the
memory limit for processing non-file form data as well as file
form data, rather than the existing behaviour of the memory limit
only applying to the file parts and the non-file parts being
arbitrarily limited to 10MB.

This ensures backwards compatibility while still providing the
user with control over the amount of non-file data that can be
processed instead of enforcing an arbitrary 10MB limit.

Change-Id: I53c09eae00147d3ff2d6bdfd4e50949267932c3d
Reviewed-on: https://go-review.googlesource.com/38195
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agonet/http: make ServeMux preserve query string during redirects
Emmanuel Odeke [Sun, 21 May 2017 00:19:54 +0000 (18:19 -0600)]
net/http: make ServeMux preserve query string during redirects

Ensure that the implicitly created redirect
for
  "/route"
after
  "/route/"
has been registered doesn't lose the query string information.

Fixes #17841.

Change-Id: Ib7df9242fab8c9368a18fc0da678003d6bec63b8
Reviewed-on: https://go-review.googlesource.com/43779
Run-TryBot: Emmanuel Odeke <emm.odeke@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/internal/obj/arm: report invalid .S/.P/.W suffix in ARM instructions
Ben Shi [Wed, 17 May 2017 07:00:27 +0000 (07:00 +0000)]
cmd/internal/obj/arm: report invalid .S/.P/.W suffix in ARM instructions

Many instructions can not have a .S suffix, such as MULS, SWI, CLZ,
CMP, STREX and others. And so do .P and .W suffixes. Even wrong
assembly code is generated for some instructions with invalid
suffixes.

This patch tries to simplify .S/.W/.P checks. And a wrong assembly
test for arm is added.

fixes #20377

Change-Id: Iba1c99d9e6b7b16a749b4d93ca2102e17c5822fe
Reviewed-on: https://go-review.googlesource.com/43561
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agonet: add examples for ParseIP, IP.DefaultMask & IP.Mask
Rob Phoenix [Fri, 19 May 2017 14:18:17 +0000 (15:18 +0100)]
net: add examples for ParseIP, IP.DefaultMask & IP.Mask

Further examples to support the net package.

Updates #5757

Change-Id: I9b65521d211f6c404b9103c1eaf22b0772eb242e
Reviewed-on: https://go-review.googlesource.com/43711
Reviewed-by: Mikio Hara <mikioh.mikioh@gmail.com>
7 years agonet: add test for RawConn.Control on Windows
Mikio Hara [Fri, 19 May 2017 10:03:35 +0000 (19:03 +0900)]
net: add test for RawConn.Control on Windows

This is a followup to https://go-review.googlesource.com/37039.

Updates #19435.

Change-Id: Ia795bd5158d26effa56e897698208ccf73f9e0d2
Reviewed-on: https://go-review.googlesource.com/43693
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agointernal/poll: implement RawControl of FD on Windows
Mikio Hara [Fri, 19 May 2017 09:49:28 +0000 (18:49 +0900)]
internal/poll: implement RawControl of FD on Windows

This is a followup to https://go-review.googlesource.com/37038.

Updates #19435.

Change-Id: If3d56bca0e8816d1a169f5cf97f27b20695a9955
Reviewed-on: https://go-review.googlesource.com/43692
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agodoc: remove mentions of yacc tool
Brad Fitzpatrick [Fri, 19 May 2017 19:39:20 +0000 (19:39 +0000)]
doc: remove mentions of yacc tool

It was removed in CL 27325.

Fixes #20431

Change-Id: I6842851444186e19029d040f61fdf4f87a3103a6
Reviewed-on: https://go-review.googlesource.com/43771
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agonet/url: add examples for URL.Hostname and URL.RequestURI
Kevin Burke [Mon, 8 May 2017 06:19:31 +0000 (23:19 -0700)]
net/url: add examples for URL.Hostname and URL.RequestURI

Change-Id: I72a10cd5dfb863f8219bb3b5b8280c017f523cf4
Reviewed-on: https://go-review.googlesource.com/42856
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

7 years agocmd/compile: make duplicate expr cases readable
Daniel Martí [Wed, 26 Apr 2017 11:04:08 +0000 (12:04 +0100)]
cmd/compile: make duplicate expr cases readable

Instead of just printing the value, print the original node to make the
error more human-friendly. Also print the value if its string form is
different than the original node, to make sure it's obvious what value
was duplicated.

This means that "case '@', '@':", which used to print:

duplicate case 64 in switch

Will now print:

duplicate case '@' (value 64) in switch

Factor this logic out into its own function to reuse it in range cases
and any other place where we might want to print a node and its value in
the future.

Also needed to split the errorcheck files because expression switch case
duplicates are now detected earlier, so they stop the compiler before it
gets to generating the AST and detecting the type switch case
duplicates.

Fixes #20112.

Change-Id: I9009b50dec0d0e705e5de9c9ccb08f1dce8a5a99
Reviewed-on: https://go-review.googlesource.com/41852
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agoruntime: use pselect6 for usleep on linux/amd64 and linux/arm
Austin Clements [Thu, 18 May 2017 20:56:48 +0000 (16:56 -0400)]
runtime: use pselect6 for usleep on linux/amd64 and linux/arm

Android O black-lists the select system call because its libc, Bionic,
does not use this system call. Replace our use of select with pselect6
(which is allowed) on the platforms that support targeting Android.
linux/arm64 already uses pselect6 because there is no select on arm64,
so only linux/amd64 and linux/arm need changing. pselect6 has been
available since Linux 2.6.16, which is before Go's minimum
requirement.

Fixes #20409.

Change-Id: Ic526b5b259a9e01d2f145a1f4d2e76e8c49ce809
Reviewed-on: https://go-review.googlesource.com/43641
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agonet: make {TCP,UDP,IP,Unix}Conn compliant of syscall.Conn interface
Mikio Hara [Mon, 13 Feb 2017 21:38:10 +0000 (06:38 +0900)]
net: make {TCP,UDP,IP,Unix}Conn compliant of syscall.Conn interface

This change makes {TCP,UDP,IP,Unix}Conn types compliant of
syscall.Conn interface and adds type rawConn as an implementation of
syscall.RawConn interface.

By this change, the long-standing issues regarding unsupported socket
options and system calls can be solved partly and the broken x/net
packages due to https://go-review.googlesource.com/36799 can be
repaired.

Fixes #3661.
Updates #9661.
Updates #19051.
Updates #19435.

Change-Id: Ic996b040418b54f6d043bc70591789d5a5b23270
Reviewed-on: https://go-review.googlesource.com/37039
Run-TryBot: Mikio Hara <mikioh.mikioh@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agointernal/poll: add RawControl, RawRead and RawWrite methods to FD
Mikio Hara [Mon, 13 Feb 2017 20:55:41 +0000 (05:55 +0900)]
internal/poll: add RawControl, RawRead and RawWrite methods to FD

This change adds RawControl, RawRead and RawWrite methods to type FD
to make the runtime-integrated network poller work together with a
user-defined function. The methods are used via the net package from
external packages and type FD is considered as an implementation of
syscall.Conn and syscall.RawConn interfaces.

Updates #19435.

Change-Id: I4ad04b10ffddb2b54fa8d70587440960d73c0a2d
Reviewed-on: https://go-review.googlesource.com/37038
Run-TryBot: Mikio Hara <mikioh.mikioh@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agosyscall: add Conn and RawConn interfaces
Mikio Hara [Tue, 7 Mar 2017 20:52:07 +0000 (05:52 +0900)]
syscall: add Conn and RawConn interfaces

This change adds Conn and RawConn interfaces which can be used to
manipulate raw network connection end points typically represented as
socket descriptors.

Fixes #19435.

Change-Id: Ide2d28eeab91bfd27473ab47a87bec69950b64c9
Reviewed-on: https://go-review.googlesource.com/37913
Run-TryBot: Mikio Hara <mikioh.mikioh@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
7 years agonet: make syscall name consistent with others
Alex Brainman [Fri, 19 May 2017 05:12:49 +0000 (15:12 +1000)]
net: make syscall name consistent with others

Change-Id: Ic6d2de92e1f533a9f9a0cd6d7dab463bdafb0e11
Reviewed-on: https://go-review.googlesource.com/43691
Reviewed-by: Mikio Hara <mikioh.mikioh@gmail.com>
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>

7 years agointernal/poll: remove unused ioSrv.ExecIO parameter
Alex Brainman [Fri, 19 May 2017 05:08:29 +0000 (15:08 +1000)]
internal/poll: remove unused ioSrv.ExecIO parameter

Change-Id: If5cb80c3c086684ce6c2e8ed9bb23b2a20c8aacd
Reviewed-on: https://go-review.googlesource.com/43690
Reviewed-by: Mikio Hara <mikioh.mikioh@gmail.com>
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoos: make windows Stat as fast as Lstat for files and directories
Alex Brainman [Tue, 9 May 2017 06:50:41 +0000 (16:50 +1000)]
os: make windows Stat as fast as Lstat for files and directories

Recent CL 41834 made windows Stat work for all symlinks.
But CL 41834 also made Stat slow.

John Starks sugested
(see https://github.com/golang/go/issues/19922#issuecomment-300031421)
to use GetFileAttributesEx for files and directories instead.
This makes Stat as fast as at go1.9.

I see these improvements on my Windows 7

name       old time/op  new time/op  delta
StatDot    26.5µs ± 1%  20.6µs ± 2%  -22.37%  (p=0.000 n=9+10)
StatFile   22.8µs ± 2%   6.2µs ± 1%  -72.69%  (p=0.000 n=10+10)
StatDir    21.0µs ± 2%   6.1µs ± 3%  -71.12%  (p=0.000 n=10+9)
LstatDot   20.1µs ± 1%  20.7µs ± 6%   +3.37%  (p=0.000 n=9+10)
LstatFile  6.23µs ± 1%  6.36µs ± 8%     ~     (p=0.587 n=9+10)
LstatDir   6.10µs ± 0%  6.14µs ± 4%     ~     (p=0.590 n=9+10)

and on my Windows XP

name         old time/op  new time/op  delta
StatDot-2    20.6µs ± 0%  10.8µs ± 0%  -47.44%  (p=0.000 n=10+10)
StatFile-2   20.2µs ± 0%   7.9µs ± 0%  -60.91%  (p=0.000 n=8+10)
StatDir-2    19.3µs ± 0%   7.6µs ± 0%  -60.51%  (p=0.000 n=10+9)
LstatDot-2   10.8µs ± 0%  10.8µs ± 0%   -0.48%  (p=0.000 n=10+8)
LstatFile-2  7.83µs ± 0%  7.83µs ± 0%     ~     (p=0.844 n=10+8)
LstatDir-2   7.59µs ± 0%  7.56µs ± 0%   -0.46%  (p=0.000 n=10+10)

Updates #19922

Change-Id: Ice1fb5825defb05c79bab4dec0692e0fd1bcfcd5
Reviewed-on: https://go-review.googlesource.com/43071
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile/internal/ssa: fix spelling mistake
Kevin Burke [Wed, 10 May 2017 05:09:08 +0000 (22:09 -0700)]
cmd/compile/internal/ssa: fix spelling mistake

Change-Id: I4b8f1b61c10f60ddb3687759af0be1641c1f78ce
Reviewed-on: https://go-review.googlesource.com/43111
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
7 years agocmd/compile: skip TestScopeRanges on Plan 9
David du Colombier [Fri, 19 May 2017 00:13:50 +0000 (02:13 +0200)]
cmd/compile: skip TestScopeRanges on Plan 9

TestScopeRanges has been added in CL 40095. This
test is failing on Plan 9 because executables don't
have a DWARF symbol table.

Fixes #20418.

Change-Id: I6dd3baa636998134ccd042203c8b5c3199a4d6e1
Reviewed-on: https://go-review.googlesource.com/43670
Run-TryBot: David du Colombier <0intro@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
7 years agoos: add some comments and remove an unused variable in rename func
Brad Fitzpatrick [Thu, 18 May 2017 22:50:35 +0000 (22:50 +0000)]
os: add some comments and remove an unused variable in rename func

This slightly clarifies the just-submitted CL 40577.

Updates #19647

Change-Id: I5584ad0e1abbc31796e3e5752351857f2a13d6d7
Reviewed-on: https://go-review.googlesource.com/43625
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agocmd/compile: output DWARF lexical blocks for local variables
Alessandro Arzilli [Tue, 2 May 2017 14:46:01 +0000 (16:46 +0200)]
cmd/compile: output DWARF lexical blocks for local variables

Change compiler and linker to emit DWARF lexical blocks in .debug_info
section when compiling with -N -l.

Version of debug_info is updated from DWARF v2 to DWARF v3 since
version 2 does not allow lexical blocks with discontinuous PC ranges.

Remaining open problems:
- scope information is removed from inlined functions
- variables records do not have DW_AT_start_scope attributes so a
variable will shadow other variables with the same name as soon as its
containing scope begins, even before its declaration.

Updates #6913.
Updates #12899.

Change-Id: Idc6808788512ea20e7e45bcf782453acb416fb49
Reviewed-on: https://go-review.googlesource.com/40095
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
7 years agoos: lstat oldname before renaming
Alexander Menzhinsky [Thu, 13 Apr 2017 16:23:35 +0000 (11:23 -0500)]
os: lstat oldname before renaming

Fixes #19647

Change-Id: Ife4f98cf2c55ee9490843797213dae2f2647b0a3
Reviewed-on: https://go-review.googlesource.com/40577
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoencoding/base32: ensure base32 decoder propagates errors correctly
Mark Ryan [Sat, 29 Apr 2017 05:19:18 +0000 (06:19 +0100)]
encoding/base32: ensure base32 decoder propagates errors correctly

A number of issues in decoder.Read and newlineFilteringReader.Read were
preventing errors from the reader supplying the encoded data from being
propagated to the caller.  Fixing these issues revealed some additional
problems in which valid decoded data was not always returned to the user
when errors were actually propagated.

This commit fixes both the error propagation and the lost decoded data
problems.  It also adds some new unit tests to ensure errors are handled
correctly by decoder.Read.  The new unit tests increase the test coverage
of this package from 96.2% to 97.9%.

Fixes #20044

Change-Id: I1a8632da20135906e2d191c2a8825b10e7ecc4c5
Reviewed-on: https://go-review.googlesource.com/42094
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agotest: gofmt chan/*.go
Josh Bleecher Snyder [Thu, 18 May 2017 21:53:12 +0000 (14:53 -0700)]
test: gofmt chan/*.go

These are functional tests, so it is safe to gofmt them.

Change-Id: I3067279c1d49809ac6a62054448ab8a6c3de9bda
Reviewed-on: https://go-review.googlesource.com/43623
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agodatabase/sql: allow drivers to support custom arg types
Daniel Theophanes [Thu, 23 Mar 2017 20:17:59 +0000 (13:17 -0700)]
database/sql: allow drivers to support custom arg types

Previously all arguments were passed through driver.IsValid.
This checked arguments against a few fundamental go types and
prevented others from being passed in as arguments.

The new interface driver.NamedValueChecker may be implemented
by both driver.Stmt and driver.Conn. This allows
this new interface to completely supersede the
driver.ColumnConverter interface as it can be used for
checking arguments known to a prepared statement and
arbitrary query arguments. The NamedValueChecker may be
skipped with driver.ErrSkip after all special cases are
exhausted to use the default argument converter.

In addition if driver.ErrRemoveArgument is returned
the argument will not be passed to the query at all,
useful for passing in driver specific per-query options.

Add a canonical Out argument wrapper to be passed
to OUTPUT parameters. This will unify checks that need to
be written in the NameValueChecker.

The statement number check is also moved to the argument
converter so the NamedValueChecker may remove arguments
passed to the query.

Fixes #13567
Fixes #18079
Updates #18417
Updates #17834
Updates #16235
Updates #13067
Updates #19797

Change-Id: I89088bd9cca4596a48bba37bfd20d987453ef237
Reviewed-on: https://go-review.googlesource.com/38533
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agotest: restore fixedbugs/bug398.go to original version
griesemer [Thu, 18 May 2017 21:11:17 +0000 (14:11 -0700)]
test: restore fixedbugs/bug398.go to original version

Cannot reproduce original problem. Compiler internals
have changed enough such that this appears to work now.

Restore original test (exported interfaces), but also
keep version of the test using non-exported interfaces.

Fixes #15596.

Change-Id: Idb32da80239963242bd5d1609343c80f19773b0c
Reviewed-on: https://go-review.googlesource.com/43622
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
7 years agotest: add test for issue 18636
Josh Bleecher Snyder [Thu, 18 May 2017 20:53:43 +0000 (13:53 -0700)]
test: add test for issue 18636

Updates #18636

Change-Id: I143c670c3940231e29f1814e0a03165682f53243
Reviewed-on: https://go-review.googlesource.com/43621
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
7 years agocmd/compile: search for remaining WB ops from end to beginning
Josh Bleecher Snyder [Wed, 17 May 2017 13:28:03 +0000 (06:28 -0700)]
cmd/compile: search for remaining WB ops from end to beginning

The writebarrier pass processes WB ops from beginning to end,
replacing them by other values.
But it also checks whether there are more ops to process
by walking from beginning to end.
This is quadratic, so walk from end to beginning instead.

This speeds up compiling the code in issue 13554:

name  old time/op       new time/op       delta
Pkg         11.9s ± 2%         8.3s ± 3%  -29.88%  (p=0.000 n=18+17)

Updates #13554

Passes toolstash-check.

Change-Id: I5f8a872ddc4b783540220d89ea2ee188a6d2b2ff
Reviewed-on: https://go-review.googlesource.com/43571
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agocmd/compile: don't break up contiguous blocks in looprotate
Josh Bleecher Snyder [Sun, 14 May 2017 21:52:09 +0000 (14:52 -0700)]
cmd/compile: don't break up contiguous blocks in looprotate

looprotate finds loop headers and arranges for them to be placed
after the body of the loop. This eliminates a jump from the body.

However, if the loop header is a series of contiguously laid out blocks,
the rotation introduces a new jump in that series.
This CL expands the "loop header" to move to be the entire
run of contiguously laid out blocks in the same loop.

This shrinks object files a little, and actually speeds up
the compiler noticeably. Numbers below.

Fannkuch performance seems to vary a lot by machine. On my laptop:

name          old time/op  new time/op  delta
Fannkuch11-8   2.89s ± 2%   2.85s ± 3%  -1.22%  (p=0.000 n=50+50)

This has a significant affect on the append benchmarks in #14758:

name   old time/op    new time/op    delta
Foo-8     312ns ± 3%     276ns ± 2%  -11.37%  (p=0.000 n=30+29)
Bar-8     565ns ± 2%     456ns ± 2%  -19.27%  (p=0.000 n=27+28)

Updates #18977
Fixes #20355

name        old time/op       new time/op       delta
Template          205ms ± 5%        204ms ± 8%    ~     (p=0.903 n=92+99)
Unicode          85.3ms ± 4%       85.1ms ± 3%    ~     (p=0.191 n=92+94)
GoTypes           512ms ± 4%        507ms ± 4%  -0.93%  (p=0.000 n=95+97)
Compiler          2.38s ± 3%        2.35s ± 3%  -1.27%  (p=0.000 n=98+95)
SSA               4.67s ± 3%        4.64s ± 3%  -0.62%  (p=0.000 n=95+96)
Flate             117ms ± 3%        117ms ± 3%    ~     (p=0.099 n=84+86)
GoParser          139ms ± 4%        137ms ± 4%  -0.90%  (p=0.000 n=97+98)
Reflect           329ms ± 5%        326ms ± 6%  -0.97%  (p=0.002 n=99+98)
Tar               102ms ± 6%        101ms ± 5%  -0.97%  (p=0.006 n=97+97)
XML               198ms ±10%        196ms ±13%    ~     (p=0.087 n=100+100)
[Geo mean]        318ms             316ms       -0.72%

name        old user-time/op  new user-time/op  delta
Template          250ms ± 7%        250ms ± 7%    ~     (p=0.850 n=94+92)
Unicode           107ms ± 8%        106ms ± 5%  -0.76%  (p=0.005 n=98+91)
GoTypes           665ms ± 5%        659ms ± 5%  -0.85%  (p=0.003 n=93+98)
Compiler          3.15s ± 3%        3.10s ± 3%  -1.60%  (p=0.000 n=99+98)
SSA               6.82s ± 3%        6.72s ± 4%  -1.55%  (p=0.000 n=94+98)
Flate             138ms ± 8%        138ms ± 6%    ~     (p=0.369 n=94+92)
GoParser          170ms ± 5%        168ms ± 6%  -1.13%  (p=0.002 n=96+98)
Reflect           412ms ± 8%        416ms ± 8%    ~     (p=0.169 n=100+100)
Tar               123ms ±18%        123ms ±14%    ~     (p=0.896 n=100+100)
XML               236ms ± 9%        234ms ±11%    ~     (p=0.124 n=100+100)
[Geo mean]        401ms             398ms       -0.63%

name        old alloc/op      new alloc/op      delta
Template         38.8MB ± 0%       38.8MB ± 0%    ~     (p=0.222 n=5+5)
Unicode          28.7MB ± 0%       28.7MB ± 0%    ~     (p=0.421 n=5+5)
GoTypes           109MB ± 0%        109MB ± 0%    ~     (p=0.056 n=5+5)
Compiler          457MB ± 0%        457MB ± 0%  +0.07%  (p=0.008 n=5+5)
SSA              1.10GB ± 0%       1.10GB ± 0%  +0.05%  (p=0.008 n=5+5)
Flate            24.5MB ± 0%       24.5MB ± 0%    ~     (p=0.222 n=5+5)
GoParser         30.9MB ± 0%       31.0MB ± 0%  +0.21%  (p=0.016 n=5+5)
Reflect          73.4MB ± 0%       73.4MB ± 0%    ~     (p=0.421 n=5+5)
Tar              25.5MB ± 0%       25.5MB ± 0%    ~     (p=0.548 n=5+5)
XML              40.9MB ± 0%       40.9MB ± 0%    ~     (p=0.151 n=5+5)
[Geo mean]       71.6MB            71.6MB       +0.07%

name        old allocs/op     new allocs/op     delta
Template           394k ± 0%         394k ± 0%    ~     (p=1.000 n=5+5)
Unicode            344k ± 0%         343k ± 0%    ~     (p=0.310 n=5+5)
GoTypes           1.16M ± 0%        1.16M ± 0%    ~     (p=1.000 n=5+5)
Compiler          4.42M ± 0%        4.42M ± 0%    ~     (p=1.000 n=5+5)
SSA               9.80M ± 0%        9.80M ± 0%    ~     (p=0.095 n=5+5)
Flate              237k ± 1%         238k ± 1%    ~     (p=0.310 n=5+5)
GoParser           320k ± 0%         322k ± 1%  +0.50%  (p=0.032 n=5+5)
Reflect            958k ± 0%         957k ± 0%    ~     (p=0.548 n=5+5)
Tar                252k ± 1%         252k ± 0%    ~     (p=1.000 n=5+5)
XML                400k ± 0%         400k ± 0%    ~     (p=0.841 n=5+5)
[Geo mean]         741k              742k       +0.06%

name        old object-bytes  new object-bytes  delta
Template           386k ± 0%         386k ± 0%  -0.05%  (p=0.008 n=5+5)
Unicode            202k ± 0%         202k ± 0%  -0.01%  (p=0.008 n=5+5)
GoTypes           1.16M ± 0%        1.16M ± 0%  -0.06%  (p=0.008 n=5+5)
Compiler          3.91M ± 0%        3.91M ± 0%  -0.06%  (p=0.008 n=5+5)
SSA               7.91M ± 0%        7.92M ± 0%  +0.01%  (p=0.008 n=5+5)
Flate              228k ± 0%         227k ± 0%  -0.04%  (p=0.008 n=5+5)
GoParser           283k ± 0%         283k ± 0%  -0.06%  (p=0.008 n=5+5)
Reflect            952k ± 0%         951k ± 0%  -0.02%  (p=0.008 n=5+5)
Tar                187k ± 0%         187k ± 0%  -0.04%  (p=0.008 n=5+5)
XML                406k ± 0%         406k ± 0%  -0.05%  (p=0.008 n=5+5)
[Geo mean]         648k              648k       -0.04%

Change-Id: I8630c4291a0eb2f7e7927bc04d7cc0efef181094
Reviewed-on: https://go-review.googlesource.com/43491
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoos: document that NewFile can return nil
Alberto Donizetti [Thu, 20 Apr 2017 09:13:55 +0000 (11:13 +0200)]
os: document that NewFile can return nil

Fixes #20023

Change-Id: I1bec3e69031ffcfd7ad71716be9597ec322528ff
Reviewed-on: https://go-review.googlesource.com/41211
Reviewed-by: Russ Cox <rsc@golang.org>
7 years agocmd/compile: don't panic in complex division
Alberto Donizetti [Thu, 4 May 2017 09:25:01 +0000 (11:25 +0200)]
cmd/compile: don't panic in complex division

Updates #20227

Change-Id: Idadc3137490181a5d380367660cee21e8313cf3b
Reviewed-on: https://go-review.googlesource.com/42650
Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agocmd/internal/obj/arm: remove illegal form of the SWI instruction
Ben Shi [Tue, 16 May 2017 11:36:54 +0000 (11:36 +0000)]
cmd/internal/obj/arm: remove illegal form of the SWI instruction

SWI only support "SWI $imm", but currently "SWI (Reg)" is also
accepted. This patch fixes it.

And more instruction tests are added to cmd/asm/internal/asm/testdata/arm.s

fixes #20375

Change-Id: Id437d853924a403e41da9b6cbddd20d994b624ff
Reviewed-on: https://go-review.googlesource.com/43552
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agonet: adjust TestInterfaceHardwareAddrWithGetmac
Alex Brainman [Fri, 31 Mar 2017 00:32:51 +0000 (11:32 +1100)]
net: adjust TestInterfaceHardwareAddrWithGetmac

Ignore adapters with "Transport Name:   N/A" line in getmac
command output. This allows us to skip duplicate MAC addresses.

Fixes #19537.

Change-Id: I6b7be9d31322f963e02023c8f1037f6e9042b479
Reviewed-on: https://go-review.googlesource.com/39071
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Avelino <t@avelino.xxx>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/link: make sure that runtime.epclntab lives in .text section
Alex Brainman [Tue, 2 May 2017 06:03:45 +0000 (16:03 +1000)]
cmd/link: make sure that runtime.epclntab lives in .text section

Second attempt to fix #14710.

CL 35272 already tried to fix this issue. But CL 35272 assumed
that runtime.epclntab type is STEXT, while it is actually SRODATA.

This CL uses Symbol.Sect.Seg to determine if symbol is part
of Segtext or Segdata.

Fixes #14710

Change-Id: Ic6b6f657555c87a64d2bc36cc4c07ab0591d00c4
Reviewed-on: https://go-review.googlesource.com/42390
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agotest: add missing copyright header to checkbce.go
Ian Lance Taylor [Wed, 17 May 2017 22:28:14 +0000 (15:28 -0700)]
test: add missing copyright header to checkbce.go

Change-Id: Iafeb8e033c876f482caa17cca414fe13b0fadb12
Reviewed-on: https://go-review.googlesource.com/43613
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Dave Cheney <dave@cheney.net>
7 years agoruntime/pprof: don't produce 0 location in count profiles
Austin Clements [Wed, 17 May 2017 18:27:28 +0000 (14:27 -0400)]
runtime/pprof: don't produce 0 location in count profiles

profileBuilder.locForPC returns 0 to mean "no location" because 0 is
an invalid location index. However, the code to build count profiles
doesn't check the result of locForPC, so this 0 location index ends up
in the profile's location list. This, in turn, causes problems later
when we decode the profile because it puts a nil *Location in the
sample's location slice, which can later lead to a nil pointer panic.

Fix this by making printCountProfile correctly discard the result of
locForPC if it returns 0. This makes this call match the other two
calls of locForPC.

Updates #15156.

Change-Id: I4492b3652b513448bc56f4cfece4e37da5e42f94
Reviewed-on: https://go-review.googlesource.com/43630
Reviewed-by: Michael Matloob <matloob@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoruntime/pprof: deflake TestGoroutineCounts
Carl Henrik Lunde [Wed, 17 May 2017 17:37:33 +0000 (19:37 +0200)]
runtime/pprof: deflake TestGoroutineCounts

TestGoroutineCounts was flaky when running on a system under load.
This happened on three builds the last couple of days.

Fix this by running this test with a single operating system thread, so
we do not depend on the operating system scheduler.  50 000 tests ran
without failure with the new version, the old version failed 0.5% of the
time.

Fixes #15156.

Change-Id: I1e5a18d0fef4f72cc9a56e376822b2849cdb0f8b
Reviewed-on: https://go-review.googlesource.com/43590
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
7 years agocmd/compile: fuse from end to beginning
Josh Bleecher Snyder [Wed, 17 May 2017 04:57:18 +0000 (21:57 -0700)]
cmd/compile: fuse from end to beginning

fuseBlockPlain was accidentally quadratic.
If you had plain blocks b1 -> b2 -> b3 -> b4,
each containing single values v1, v2, v3, and v4 respectively,
fuseBlockPlain would move v1 from b1 to b2 to b3 to b4,
then v2 from b2 to b3 to b4, etc.

There are two obvious fixes.

* Look for runs of blocks in fuseBlockPlain
  and handle them in a single go.
* Fuse from end to beginning; any given value in a run
  of blocks to fuse then moves only once.

The latter is much simpler, so that's what this CL does.

Somewhat surprisingly, this change does not pass toolstash-check.

The resulting set of blocks is the same,
and the values in them are the same,
but the order of values in them differ,
and that order of values (while arbitrary)
is enough to change the compiler's output.
This may be due to #20178; deadstore is the next pass after fuse.

Adding basic sorting to the beginning of deadstore
is enough to make this CL pass toolstash-check:

for _, b := range f.Blocks {
obj.SortSlice(b.Values, func(i, j int) bool { return b.Values[i].ID < b.Values[j].ID })
}

Happily, this CL appears to result in better code on average,
if only by accident. It cuts 4k off of cmd/go; go1 benchmarks
are noisy as always but don't regress (numbers below).

No impact on the standard compilebench benchmarks.
For the code in #13554, this speeds up compilation dramatically:

name  old time/op       new time/op       delta
Pkg         53.1s ± 2%        12.8s ± 3%  -75.92%  (p=0.008 n=5+5)

name  old user-time/op  new user-time/op  delta
Pkg         55.0s ± 2%        14.9s ± 3%  -73.00%  (p=0.008 n=5+5)

name  old alloc/op      new alloc/op      delta
Pkg        2.04GB ± 0%       2.04GB ± 0%   +0.18%  (p=0.008 n=5+5)

name  old allocs/op     new allocs/op     delta
Pkg         6.21M ± 0%        6.21M ± 0%     ~     (p=0.222 n=5+5)

name  old object-bytes  new object-bytes  delta
Pkg         28.4M ± 0%        28.4M ± 0%   +0.00%  (p=0.008 n=5+5)

name  old export-bytes  new export-bytes  delta
Pkg           208 ± 0%          208 ± 0%     ~     (all equal)

Updates #13554

go1 benchmarks:

name                     old time/op    new time/op    delta
BinaryTree17-8              2.29s ± 2%     2.26s ± 2%  -1.43%  (p=0.000 n=48+50)
Fannkuch11-8                2.74s ± 2%     2.79s ± 2%  +1.63%  (p=0.000 n=50+49)
FmtFprintfEmpty-8          36.6ns ± 3%    34.6ns ± 4%  -5.29%  (p=0.000 n=49+50)
FmtFprintfString-8         58.3ns ± 3%    59.1ns ± 3%  +1.35%  (p=0.000 n=50+49)
FmtFprintfInt-8            62.4ns ± 2%    63.2ns ± 3%  +1.19%  (p=0.000 n=49+49)
FmtFprintfIntInt-8         95.1ns ± 2%    96.7ns ± 3%  +1.61%  (p=0.000 n=49+50)
FmtFprintfPrefixedInt-8     118ns ± 3%     113ns ± 2%  -4.00%  (p=0.000 n=50+49)
FmtFprintfFloat-8           191ns ± 2%     192ns ± 2%  +0.40%  (p=0.034 n=50+50)
FmtManyArgs-8               419ns ± 2%     420ns ± 2%    ~     (p=0.228 n=49+49)
GobDecode-8                5.26ms ± 3%    5.19ms ± 2%  -1.33%  (p=0.000 n=50+49)
GobEncode-8                4.12ms ± 2%    4.15ms ± 3%  +0.68%  (p=0.007 n=49+50)
Gzip-8                      198ms ± 2%     197ms ± 2%  -0.50%  (p=0.018 n=48+48)
Gunzip-8                   31.9ms ± 3%    31.8ms ± 3%  -0.47%  (p=0.024 n=50+50)
HTTPClientServer-8         64.4µs ± 0%    64.0µs ± 0%  -0.55%  (p=0.000 n=43+46)
JSONEncode-8               10.6ms ± 2%    10.6ms ± 3%    ~     (p=0.543 n=49+49)
JSONDecode-8               43.3ms ± 3%    43.1ms ± 2%    ~     (p=0.079 n=50+50)
Mandelbrot200-8            3.70ms ± 2%    3.70ms ± 2%    ~     (p=0.553 n=47+50)
GoParse-8                  2.70ms ± 2%    2.71ms ± 3%    ~     (p=0.843 n=49+50)
RegexpMatchEasy0_32-8      70.5ns ± 4%    70.4ns ± 4%    ~     (p=0.867 n=48+50)
RegexpMatchEasy0_1K-8       162ns ± 3%     162ns ± 2%    ~     (p=0.739 n=48+48)
RegexpMatchEasy1_32-8      66.1ns ± 5%    66.2ns ± 4%    ~     (p=0.970 n=50+50)
RegexpMatchEasy1_1K-8       297ns ± 7%     296ns ± 7%    ~     (p=0.406 n=50+50)
RegexpMatchMedium_32-8      105ns ± 5%     105ns ± 5%    ~     (p=0.702 n=50+50)
RegexpMatchMedium_1K-8     32.3µs ± 4%    32.2µs ± 3%    ~     (p=0.614 n=49+49)
RegexpMatchHard_32-8       1.75µs ±18%    1.74µs ±12%    ~     (p=0.738 n=50+48)
RegexpMatchHard_1K-8       52.2µs ±14%    51.3µs ±13%    ~     (p=0.230 n=50+50)
Revcomp-8                   366ms ± 3%     367ms ± 3%    ~     (p=0.745 n=49+49)
Template-8                 48.5ms ± 4%    48.5ms ± 4%    ~     (p=0.824 n=50+48)
TimeParse-8                 263ns ± 2%     256ns ± 2%  -2.98%  (p=0.000 n=48+49)
TimeFormat-8                265ns ± 3%     262ns ± 3%  -1.35%  (p=0.000 n=48+49)
[Geo mean]                 41.1µs         40.9µs       -0.48%

Change-Id: Ib35fa15b54282abb39c077d150beee27f610891a
Reviewed-on: https://go-review.googlesource.com/43570
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/compile: seed rand with time when race enabled
Josh Bleecher Snyder [Wed, 17 May 2017 15:29:18 +0000 (08:29 -0700)]
cmd/compile: seed rand with time when race enabled

When the race detector is enabled,
the compiler randomizes the order in which functions are compiled,
in an attempt to shake out bugs.
But we never re-seed the rand source, so every execution is identical.
Fix that to get more coverage.

Change-Id: If5cdde03ef4f1bab5f45e07f03fb6614945481d7
Reviewed-on: https://go-review.googlesource.com/43572
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/cgo: support large unsigned macros
Hiroshi Ioka [Tue, 16 May 2017 12:52:41 +0000 (21:52 +0900)]
cmd/cgo: support large unsigned macros

Currently, cgo converts integer macros into int64 if it's possible.
As a result, some macros which satisfy

math.MaxInt64 < x <= math.MaxUint64

will lose their original values.

This CL introduces the new probe to check signs,
so we can handle signed ints and unsigned ints separately.

Fixes #20369

Change-Id: I002ba452a82514b3a87440960473676f842cc9ee
Reviewed-on: https://go-review.googlesource.com/43476
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/go: don't fail on missing runtime/internal/sys/zversion.go
Ian Lance Taylor [Wed, 17 May 2017 01:02:56 +0000 (18:02 -0700)]
cmd/go: don't fail on missing runtime/internal/sys/zversion.go

The generated file runtime/internal/sys/zversion.go is deleted by
`go tool cmd dist clean` as part of running clean.bash. Don't treat
a missing file as a reason to stop running the go tool; just treat
is as meaning that runtime/internal/sys is stale.

No test because I don't particularly want to clobber $GOROOT.

Fixes #20385.

Change-Id: I5251a99542cc93c33f627f133d7118df56e18af1
Reviewed-on: https://go-review.googlesource.com/43559
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/go: fix TestExecutableGOROOT if GOROOT is a symlink
Ian Lance Taylor [Mon, 15 May 2017 23:54:38 +0000 (16:54 -0700)]
cmd/go: fix TestExecutableGOROOT if GOROOT is a symlink

Fixes #20365.

Change-Id: If1a4866193cff3bc836d8bbf18b6a1f5deb9808d
Reviewed-on: https://go-review.googlesource.com/43550
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Alessandro Arzilli <alessandro.arzilli@gmail.com>
7 years agoos: fix handling of ErrShortWrite in (*File).Write
Ian Lance Taylor [Wed, 17 May 2017 00:41:42 +0000 (17:41 -0700)]
os: fix handling of ErrShortWrite in (*File).Write

Restore the handling of io.ErrShortWrite in (*File).Write:
if we write less than the requested amount, and there is no error from
the syscall, then return io.ErrShortWrite.

I can't figure out how to write a test for this. It would require a
non-pollable file (not a pipe) on a device that is almost but not
quite entirely full. The original code (https://golang.org/cl/36800043,
committed as part of https://golang.org/cl/36930044) does not have a test.

Fixes #20386.

Change-Id: Ied7b411e621e1eaf49f864f8db90069f276256f5
Reviewed-on: https://go-review.googlesource.com/43558
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoruntime: mmap(2) on Solaris & Illumos can return EAGAIN.
Sean Chittenden [Sun, 14 May 2017 16:42:35 +0000 (09:42 -0700)]
runtime: mmap(2) on Solaris & Illumos can return EAGAIN.

In low memory situations mmap(2) on Illumos[2] can return EAGAIN when it
is unable to reserve the necessary space for the requested mapping.  Go
was not previously handling this correctly for Illumos and would fail to
recognize it was in a low-memory situation, the result being the program
would terminate with a panic instead of running the GC.

Fixes: #14930
[1]: https://www.illumos.org/man/2/mmap

Change-Id: I889cc0547e23f9d6c56e4fdd7bcbd0e15403873a
Reviewed-on: https://go-review.googlesource.com/43461
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agoruntime, syscall: in Plan 9 ARM asm, replace "SWI 0" with "SWI $0"
Cherry Zhang [Tue, 16 May 2017 13:30:27 +0000 (09:30 -0400)]
runtime, syscall: in Plan 9 ARM asm, replace "SWI 0" with "SWI $0"

On other systems we use "SWI $n". Change Plan 9 files to be
consistent. Generated binary is unchanged.

Fixes #20378.

Change-Id: Ia2a722061da2450c7b30cb707ed4f172fafecf74
Reviewed-on: https://go-review.googlesource.com/43533
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocrypto/tls: pass argument to serverInit rather than using a field in Config.
Adam Langley [Fri, 28 Apr 2017 20:37:52 +0000 (13:37 -0700)]
crypto/tls: pass argument to serverInit rather than using a field in Config.

Updates #20164.

Change-Id: Ib900095e7885f25cd779750674a712c770603ca8
Reviewed-on: https://go-review.googlesource.com/42137
Reviewed-by: Russ Cox <rsc@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/internal/obj/mips: add support of LLV, SCV, NOOP instructions
Cherry Zhang [Tue, 11 Apr 2017 14:39:55 +0000 (10:39 -0400)]
cmd/internal/obj/mips: add support of LLV, SCV, NOOP instructions

LLV and SCV are 64-bit load-linked and store-conditional. They
were used in runtime as #define WORD. Change them to normal
instruction form.

NOOP is hardware no-op. It was written as WORD $0. Make a name
for it for better disassembly output.

Fixes #12561.
Fixes #18238.

Change-Id: I82c667ce756fa83ef37b034b641e8c4366335e83
Reviewed-on: https://go-review.googlesource.com/40297
Reviewed-by: Minux Ma <minux@golang.org>
Run-TryBot: Minux Ma <minux@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agopath: add examples
Mark Harrison [Tue, 9 May 2017 05:48:08 +0000 (22:48 -0700)]
path: add examples

This change adds several examples, with emphasis on special or edge
cases such as a directory parameter consisting of an empty string.

Change-Id: Ib4ac3d0f6d503493eeed0c4fda7c12acf782e9e2
Reviewed-on: https://go-review.googlesource.com/43010
Reviewed-by: Steve Francia <spf@golang.org>
Run-TryBot: Jaana Burcu Dogan <jbd@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agocmd/compile: eliminate some bounds checks from generated rewrite rules
Josh Bleecher Snyder [Mon, 15 May 2017 16:40:19 +0000 (09:40 -0700)]
cmd/compile: eliminate some bounds checks from generated rewrite rules

Noticed while looking at #20356.

Cuts 160k (1%) off of the cmd/compile binary.

Change-Id: If2397bc6971d6be9be6975048adecb0b5efa6d66
Reviewed-on: https://go-review.googlesource.com/43501
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/compile: lay out exit blocks last
Josh Bleecher Snyder [Fri, 12 May 2017 13:10:25 +0000 (06:10 -0700)]
cmd/compile: lay out exit blocks last

In Go 1.8.x, panics are generally scheduled at the very end of functions.
That property was lost in Go 1.9; this CL restores it.

This helps with the Fannkuch benchmark:

name          old time/op  new time/op  delta
Fannkuch11-8   2.74s ± 2%   2.55s ± 2%  -7.03%  (p=0.000 n=20+20)

This increases the fannkuch function size from 801 bytes to 831 bytes,
but that is still smaller than Go 1.8.1 at 844 bytes.

It generally increases binary size a tiny amount.
Negligible compiler performance impact.

For the code in #14758:

name   old time/op    new time/op    delta
Foo-8     326ns ± 3%     312ns ± 3%  -4.32%  (p=0.000 n=28+30)
Bar-8     560ns ± 2%     565ns ± 2%  +0.96%  (p=0.002 n=30+27)

Updates #18977

name        old alloc/op      new alloc/op      delta
Template         38.8MB ± 0%       38.8MB ± 0%    ~     (p=0.690 n=5+5)
Unicode          28.7MB ± 0%       28.7MB ± 0%    ~     (p=0.841 n=5+5)
GoTypes           109MB ± 0%        109MB ± 0%    ~     (p=0.690 n=5+5)
Compiler          457MB ± 0%        457MB ± 0%    ~     (p=0.841 n=5+5)
SSA              1.10GB ± 0%       1.10GB ± 0%  +0.03%  (p=0.032 n=5+5)
Flate            24.4MB ± 0%       24.5MB ± 0%    ~     (p=0.690 n=5+5)
GoParser         30.9MB ± 0%       30.9MB ± 0%    ~     (p=0.421 n=5+5)
Reflect          73.3MB ± 0%       73.3MB ± 0%    ~     (p=1.000 n=5+5)
Tar              25.5MB ± 0%       25.5MB ± 0%    ~     (p=0.095 n=5+5)
XML              40.8MB ± 0%       40.9MB ± 0%    ~     (p=0.056 n=5+5)
[Geo mean]       71.6MB            71.6MB       +0.01%

name        old allocs/op     new allocs/op     delta
Template           395k ± 0%         394k ± 1%    ~     (p=1.000 n=5+5)
Unicode            344k ± 0%         344k ± 0%    ~     (p=0.690 n=5+5)
GoTypes           1.16M ± 0%        1.16M ± 0%    ~     (p=0.421 n=5+5)
Compiler          4.41M ± 0%        4.41M ± 0%    ~     (p=0.841 n=5+5)
SSA               9.79M ± 0%        9.79M ± 0%    ~     (p=0.310 n=5+5)
Flate              237k ± 0%         237k ± 0%    ~     (p=0.841 n=5+5)
GoParser           321k ± 0%         321k ± 1%    ~     (p=0.421 n=5+5)
Reflect            956k ± 0%         956k ± 0%    ~     (p=1.000 n=5+5)
Tar                251k ± 1%         252k ± 0%    ~     (p=0.095 n=5+5)
XML                399k ± 0%         400k ± 0%    ~     (p=0.222 n=5+5)
[Geo mean]         741k              741k       +0.03%

name        old object-bytes  new object-bytes  delta
Template           386k ± 0%         386k ± 0%  +0.05%  (p=0.008 n=5+5)
Unicode            202k ± 0%         202k ± 0%  +0.02%  (p=0.008 n=5+5)
GoTypes           1.16M ± 0%        1.16M ± 0%  +0.07%  (p=0.008 n=5+5)
Compiler          3.91M ± 0%        3.91M ± 0%  +0.05%  (p=0.008 n=5+5)
SSA               7.86M ± 0%        7.87M ± 0%  +0.07%  (p=0.008 n=5+5)
Flate              227k ± 0%         227k ± 0%  +0.10%  (p=0.008 n=5+5)
GoParser           283k ± 0%         283k ± 0%  +0.04%  (p=0.008 n=5+5)
Reflect            950k ± 0%         951k ± 0%  +0.04%  (p=0.008 n=5+5)
Tar                187k ± 0%         187k ± 0%  -0.03%  (p=0.008 n=5+5)
XML                406k ± 0%         406k ± 0%  +0.04%  (p=0.008 n=5+5)
[Geo mean]         647k              647k       +0.04%

Change-Id: I2015aa26338b90cf41e47f89564e336dc02608df
Reviewed-on: https://go-review.googlesource.com/43293
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/internal/objabi: fix the bug of shrinking SymType down to a uint8
Wei Xiao [Tue, 9 May 2017 08:25:13 +0000 (16:25 +0800)]
cmd/internal/objabi: fix the bug of shrinking SymType down to a uint8

Previous CL (cmd/internal/objabi: shrink SymType down to a uint8) shrinks
SymType down to a uint8 but forgot making according change in goobj.

Fixes #20296
Also add a test to catch such Goobj format inconsistency bug

Change-Id: Ib43dd7122cfcacf611a643814e95f8c5a924941f
Reviewed-on: https://go-review.googlesource.com/42971
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Hudson-Doyle <michael.hudson@canonical.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agogo/types: fewer spurious "declared but not used" follow-on errors
Robert Griesemer [Mon, 15 May 2017 18:46:05 +0000 (11:46 -0700)]
go/types: fewer spurious "declared but not used" follow-on errors

Mark variables as used even when they appear within an expression
context which we can't type-check; e.g., because the expression is
erroneous, or comes from an import "C" declaration.

Fixes #20358.

Change-Id: Ib28cc78d3867c597c7a1ace54de09ada02f5b33a
Reviewed-on: https://go-review.googlesource.com/43500
Reviewed-by: Alan Donovan <adonovan@google.com>
7 years agocmd/compile: don't attach lines to SB, SP, similar constants
David Chase [Mon, 15 May 2017 17:49:30 +0000 (13:49 -0400)]
cmd/compile: don't attach lines to SB, SP, similar constants

Attaching positions to SB, SP, initial mem can result in
less-good line-numbering when compiled for debugging.
This "fix" also removes source position from a zero-valued
struct (but not from its fields) and from a zero-length
array constant.

This may be a general problem for constants in entry blocks.

Fixes #20367.

Change-Id: I7e9df3341be2e2f60f127d35bb31e43cdcfce9a1
Reviewed-on: https://go-review.googlesource.com/43531
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agocmd/compile: better check for single live memory
Keith Randall [Mon, 15 May 2017 16:00:55 +0000 (09:00 -0700)]
cmd/compile: better check for single live memory

Enhance the one-live-memory-at-a-time check to run during many
more phases of the SSA backend. Also make it work in an interblock
fashion.

Change types.IsMemory to return true for tuples containing a memory type.

Fix trim pass to build the merged phi correctly. Doesn't affect
code but allows the check to pass after trim runs.

Switch the AddTuple* ops to take the memory-containing tuple argument second.

Update #20335

Change-Id: I5b03ef3606b75a9e4f765276bb8b183cdc172b43
Reviewed-on: https://go-review.googlesource.com/43495
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
7 years agotest: add test of deferred delete
Josh Bleecher Snyder [Mon, 15 May 2017 17:41:25 +0000 (10:41 -0700)]
test: add test of deferred delete

Updates #19710

Change-Id: I37d19a4a02b9010cb5f9062b3d141d5d65e12e01
Reviewed-on: https://go-review.googlesource.com/43497
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agogo/build: return partial information on Import error, for local import paths
Dmitri Shuralyov [Mon, 1 May 2017 21:28:33 +0000 (17:28 -0400)]
go/build: return partial information on Import error, for local import paths

Documentation of build.Import says:

// If the path is a local import path naming a package that can be imported
// using a standard import path, the returned package will set p.ImportPath
// to that path.
// ...
// If an error occurs, Import returns a non-nil error and a non-nil
// *Package containing partial information.

That behavior was previously untested, and broken by change in CL 33158.

Fix that by avoiding returning early on error for local import paths.
First, gather partial information, and only then check that the p.Dir
directory exists.

Add tests for this behavior.

Fixes #19769.
Fixes #20175 (duplicate of #19769).
Updates #17863.

Change-Id: I169cb35291099d05e02aaa3cb23a7403d1cc3657
Reviewed-on: https://go-review.googlesource.com/42350
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agoruntime/pprof: expand inlined frames in symbolized proto profiles
Austin Clements [Thu, 20 Apr 2017 19:54:12 +0000 (15:54 -0400)]
runtime/pprof: expand inlined frames in symbolized proto profiles

Currently proto symbolization uses runtime.FuncForPC and assumes each
PC maps to a single frame. This isn't true in the presence of inlining
(even with leaf-only inlining this can get incorrect results).

Change PC symbolization to use runtime.CallersFrames to expand each PC
to all of the frames at that PC.

Change-Id: I8d20dff7495a5de495ae07f569122c225d433ced
Reviewed-on: https://go-review.googlesource.com/41256
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Matloob <matloob@golang.org>
7 years agoruntime/pprof: clean up call/return PCs in memory profiles
Austin Clements [Tue, 9 May 2017 02:31:41 +0000 (22:31 -0400)]
runtime/pprof: clean up call/return PCs in memory profiles

Proto profile conversion is inconsistent about call vs return PCs in
profile locations. The proto defines locations to be call PCs. This is
what we do when proto-izing CPU profiles, but we fail to convert the
return PCs in memory and count profile stacks to call PCs when
converting them to proto locations.

Fix this in the heap and count profile conversion functions.
TestConvertMemProfile also hard-codes this failure to convert from
return PCs to call PCs, so fix up the addresses in the synthesized
profile to be return PCs while checking that we get call PCs out of
the conversion.

Change-Id: If1fc028b86fceac6d71a2d9fa6c41ff442c89296
Reviewed-on: https://go-review.googlesource.com/42951
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Matloob <matloob@golang.org>
7 years agocmd/link: actually generate .debug_gdb_scripts section on windows
Alex Brainman [Thu, 11 May 2017 01:55:59 +0000 (11:55 +1000)]
cmd/link: actually generate .debug_gdb_scripts section on windows

Adjust finddebugruntimepath to look for runtime/debug.go file
instead of runtime/runtime.go. This actually finds runtime.GOMAXPROCS
in every Go executable (including windows).

I also included "-Wl,-T,fix_debug_gdb_scripts.ld" parameter to gcc
invocation on windows to work around gcc bug (see #20183 for details).

This CL only fixes windows -buildmode=exe, buildmode=c-archive
is still broken.

Thanks to Egon Elbre and Nick Clifton for investigation.

Fixes #20183
Fixes #20218

Change-Id: I5369a4db3913226aef3d9bd6317446856b0a1c34
Reviewed-on: https://go-review.googlesource.com/43331
Reviewed-by: Ian Lance Taylor <iant@golang.org>
7 years agotext/scanner: clarify documentation on Pos and Position
Robert Griesemer [Fri, 12 May 2017 19:49:47 +0000 (12:49 -0700)]
text/scanner: clarify documentation on Pos and Position

For #20292. (See discussion in that issue.)

Change-Id: I44cd69394fc47a01776905ec34305ba524c89883
Reviewed-on: https://go-review.googlesource.com/43452
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/compile: tweak ssa.html colors
Josh Bleecher Snyder [Sat, 13 May 2017 15:29:47 +0000 (08:29 -0700)]
cmd/compile: tweak ssa.html colors

Make yellow the last highlight color rather than the first.
Yellow is also the color that Chrome uses to highlight
search results, which can be confusing.
Also, when Night Shift is on on macOS,
yellow highlighting is completely invisible.
I suppose should be sleeping instead.

Also, remove a completed TODO.

Change-Id: I0eb4439272fad9ccb5fe8e2cf409fdd5dc15b26e
Reviewed-on: https://go-review.googlesource.com/43463
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agocmd/compile: don't update outer variables after capturevars is complete
Josh Bleecher Snyder [Fri, 5 May 2017 22:22:59 +0000 (15:22 -0700)]
cmd/compile: don't update outer variables after capturevars is complete

When compiling concurrently, we walk all functions before compiling
any of them. Walking functions can cause variables to switch from
being non-addrtaken to addrtaken, e.g. to prepare for a runtime call.
Typechecking propagates addrtaken-ness of closure variables to
their outer variables, so that capturevars can decide whether to
pass the variable's value or a pointer to it.

When all functions are compiled immediately, as long as the containing
function is compiled prior to the closure, this propagation has no effect.
When compilation is deferred, though, in rare cases, this results in
a change in the addrtaken-ness of a variable in the outer function,
which in turn changes the compiler's output.
(This is rare because in a great many cases, a temporary has been
introduced, insulating the outer variable from modification.)
But concurrent compilation must generate identical results.

To fix this, track whether capturevars has run.
If it has, there is no need to update outer variables
when closure variables change.
Capturevars always runs before any functions are walked or compiled.

The remainder of the changes in this CL are to support the test.
In particular, -d=compilelater forces the compiler to walk all
functions before compiling any of them, despite being non-concurrent.
This is useful because -live is fundamentally incompatible with
concurrent compilation, but we want -c=1 to have no behavior changes.

Fixes #20250

Change-Id: I89bcb54268a41e8588af1ac8cc37fbef856a90c2
Reviewed-on: https://go-review.googlesource.com/42853
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agocontainer/heap: avoid up() invoke if down() success at heap.Remove()
wuyunzhou [Sat, 13 May 2017 07:41:15 +0000 (15:41 +0800)]
container/heap: avoid up() invoke if down() success at heap.Remove()

Change-Id: I6c210e0c23ca533e1f303f88ef9dcb629a294a2a
Reviewed-on: https://go-review.googlesource.com/43472
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>

7 years agobuiltin: fix signature of the builtin function make
Brad Fitzpatrick [Sat, 13 May 2017 15:49:15 +0000 (15:49 +0000)]
builtin: fix signature of the builtin function make

Fixes #20325

Change-Id: Ie363fef73d0deae40af41ee3f4403ad18546eba6
Reviewed-on: https://go-review.googlesource.com/43460
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
7 years agoarchive/tar: remove file type bits from mode field
Lars Jeppesen [Sat, 29 Apr 2017 21:25:34 +0000 (23:25 +0200)]
archive/tar: remove file type bits from mode field

When writing tar files by using the FileInfoHeader
the type bits was set in the mode field of the header
This is not correct according to the standard (GNU/Posix) and
other implementations.

Fixed #20150

Change-Id: I3be7d946a1923ad5827cf45c696546a5e287ebba
Reviewed-on: https://go-review.googlesource.com/42093
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agotext/scanner: use correct token position in example
Robert Griesemer [Fri, 12 May 2017 18:58:09 +0000 (11:58 -0700)]
text/scanner: use correct token position in example

While at it, unindent source text so column values are easier
to read, remove unnecessary text in output, and simplify the
loop.

Fixes #20346.

Change-Id: I0fde02b9e4242383da427f4cf4c6c13dd0ab3b47
Reviewed-on: https://go-review.googlesource.com/43450
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

7 years agonet: allow Resolver to use a custom dialer
Matt Harden [Mon, 20 Feb 2017 13:58:55 +0000 (05:58 -0800)]
net: allow Resolver to use a custom dialer

In some cases it is desirable to customize the way the DNS server is
contacted, for instance to use a specific LocalAddr. While most
operating-system level resolvers do not allow this, we have the
opportunity to do so with the Go resolver. Most of the code was
already in place to allow tests to override the dialer. This exposes
that functionality, and as a side effect eliminates the need for a
testing hook.

Fixes #17404

Change-Id: I1c5e570f8edbcf630090f8ec6feb52e379e3e5c0
Reviewed-on: https://go-review.googlesource.com/37260
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
7 years agoruntime: doubly fix "double wakeup" panic
Austin Clements [Thu, 11 May 2017 19:28:39 +0000 (15:28 -0400)]
runtime: doubly fix "double wakeup" panic

runtime.gchelper depends on the non-atomic load of work.ndone
happening strictly before the atomic add of work.nwait. Until very
recently (commit 978af9c2db, fixing #20334), the compiler reordered
these operations. This created a race since work.ndone can change as
soon as work.nwait is equal to work.ndone. If that happened, more than
one gchelper could attempt to wake up the work.alldone note, causing a
"double wakeup" panic.

This was fixed in the compiler, but to make this code less subtle,
make the load of work.ndone atomic. This clearly forces the order of
these operations, ensuring the race doesn't happen.

Fixes #19305 (though really 978af9c2db fixed it).

Change-Id: Ieb1a84e1e5044c33ac612c8a5ab6297e7db4c57d
Reviewed-on: https://go-review.googlesource.com/43311
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
7 years agoruntime: print debug info on "base out of range"
Austin Clements [Thu, 11 May 2017 19:22:10 +0000 (15:22 -0400)]
runtime: print debug info on "base out of range"

This adds debugging information when we panic with "heapBitsForSpan:
base out of range".

Updates #20259.

Change-Id: I0dc1a106aa9e9531051c7d08867ace5ef230eb3f
Reviewed-on: https://go-review.googlesource.com/43310
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>