]> Cypherpunks repositories - gostls13.git/log
gostls13.git
9 years agoruntime: directly track GC assist balance
Austin Clements [Mon, 5 Oct 2015 03:16:57 +0000 (20:16 -0700)]
runtime: directly track GC assist balance

Currently we track the per-G GC assist balance as two monotonically
increasing values: the bytes allocated by the G this cycle (gcalloc)
and the scan work performed by the G this cycle (gcscanwork). The
assist balance is hence assistRatio*gcalloc - gcscanwork.

This works, but has two important downsides:

1) It requires floating-point math to figure out if a G is in debt or
   not. This makes it inappropriate to check for assist debt in the
   hot path of mallocgc, so we only do this when a G allocates a new
   span. As a result, Gs can operate "in the red", leading to
   under-assist and extended GC cycle length.

2) Revising the assist ratio during a GC cycle can lead to an "assist
   burst". If you think of plotting the scan work performed versus
   heaps size, the assist ratio controls the slope of this line.
   However, in the current system, the target line always passes
   through 0 at the heap size that triggered GC, so if the runtime
   increases the assist ratio, there has to be a potentially large
   assist to jump from the current amount of scan work up to the new
   target scan work for the current heap size.

This commit replaces this approach with directly tracking the GC
assist balance in terms of allocation credit bytes. Allocating N bytes
simply decreases this by N and assisting raises it by the amount of
scan work performed divided by the assist ratio (to get back to
bytes).

This will make it cheap to figure out if a G is in debt, which will
let us efficiently check if an assist is necessary *before* performing
an allocation and hence keep Gs "in the black".

This also fixes assist bursts because the assist ratio is now in terms
of *remaining* work, rather than work from the beginning of the GC
cycle. Hence, the plot of scan work versus heap size becomes
continuous: we can revise the slope, but this slope always starts from
where we are right now, rather than where we were at the beginning of
the cycle.

Change-Id: Ia821c5f07f8a433e8da7f195b52adfedd58bdf2c
Reviewed-on: https://go-review.googlesource.com/15408
Reviewed-by: Rick Hudson <rlh@golang.org>
9 years agoruntime: ensure minimum heap distance via heap goal
Austin Clements [Thu, 8 Oct 2015 05:37:15 +0000 (22:37 -0700)]
runtime: ensure minimum heap distance via heap goal

Currently we ensure a minimum heap distance of 1MB when computing the
assist ratio. Rather than enforcing this minimum on the heap distance,
it makes more sense to enforce that the heap goal itself is at least
1MB over the live heap size at the beginning of GC. Currently the two
approaches are semantically equivalent, but this will let us switch to
basing the assist ratio on current heap distance rather than the
initial heap distance, since we can't enforce this minimum on the
current heap distance (the GC may never finish because the goal posts
will always be 1MB away).

Change-Id: I0027b1c26a41a0152b01e5b67bdb1140d43ee903
Reviewed-on: https://go-review.googlesource.com/15604
Reviewed-by: Rick Hudson <rlh@golang.org>
9 years agoruntime: update gcController.scanWork regularly
Austin Clements [Mon, 5 Oct 2015 03:00:01 +0000 (23:00 -0400)]
runtime: update gcController.scanWork regularly

Currently, gcController.scanWork is updated as lazily as possible
since it is only read at the end of the GC cycle. We're about to read
it during the GC cycle to improve the assist ratio revisions, so
modify gcDrain* to regularly flush to gcController.scanWork in much
the same way as we regularly flush to gcController.bgScanCredit.

One consequence of this is that it's difficult to keep gcw.scanWork
monotonic, so we give up on that and simply return the amount of scan
work done by gcDrainN rather than calculating it in the caller.

Change-Id: I7b50acdc39602f843eed0b5c6d2dacd7e762b81d
Reviewed-on: https://go-review.googlesource.com/15407
Reviewed-by: Rick Hudson <rlh@golang.org>
9 years agoruntime: control background scan credit flushing with flag
Austin Clements [Mon, 5 Oct 2015 02:47:27 +0000 (22:47 -0400)]
runtime: control background scan credit flushing with flag

Currently callers of gcDrain control whether it flushes scan work
credit to gcController.bgScanCredit by passing a value other than -1
for the flush threshold. Shortly we're going to make this always flush
scan work to gcController.scanWork and optionally also flush scan work
to gcController.bgScanCredit. This will be much easier if the flush
threshold is simply a constant (which it is in practice) and callers
merely control whether or not the flush includes the background
credit. Hence, replace the flush threshold argument with a flag.

Change-Id: Ia27db17de8a3f1e462a5d7137d4b5dc72f99a04e
Reviewed-on: https://go-review.googlesource.com/15406
Reviewed-by: Rick Hudson <rlh@golang.org>
9 years agoruntime: consolidate gcDrain and gcDrainUntilPreempt
Austin Clements [Mon, 5 Oct 2015 02:42:43 +0000 (22:42 -0400)]
runtime: consolidate gcDrain and gcDrainUntilPreempt

These functions were nearly identical. Consolidate them by adding a
flags argument. In addition to cleaning up this code, this makes
further changes that affect both functions easier.

Change-Id: I6ec5c947603bbbd3ff4040113b2fbc240e99745f
Reviewed-on: https://go-review.googlesource.com/15405
Reviewed-by: Rick Hudson <rlh@golang.org>
9 years agoruntime: explain why continuous assist revising is necessary
Austin Clements [Mon, 5 Oct 2015 03:16:07 +0000 (20:16 -0700)]
runtime: explain why continuous assist revising is necessary

Change-Id: I950af8d80433b3ae8a1da0aa7a8d2d0b295dd313
Reviewed-on: https://go-review.googlesource.com/15404
Reviewed-by: Rick Hudson <rlh@golang.org>
9 years agoruntime: fix comment for gcAssistAlloc
Austin Clements [Mon, 5 Oct 2015 01:34:27 +0000 (21:34 -0400)]
runtime: fix comment for gcAssistAlloc

Change-Id: I312e56e95d8ef8ae036d16444ab1e2df1285845d
Reviewed-on: https://go-review.googlesource.com/15403
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoruntime: fix comment for assistRatio
Austin Clements [Sun, 4 Oct 2015 23:28:42 +0000 (19:28 -0400)]
runtime: fix comment for assistRatio

The comment for assistRatio claimed it to be the reciprocal of what it
actually is.

Change-Id: If7f9bb853d75d0097facff3aa6704b224d9108b8
Reviewed-on: https://go-review.googlesource.com/15402
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoruntime: remove redundant type cast
Nodir Turakulov [Fri, 9 Oct 2015 18:15:53 +0000 (11:15 -0700)]
runtime: remove redundant type cast

  (*T)(unsafe.Pointer(&t)) === &t
for t of type T

Change-Id: I43c1aa436747dfa0bf4cb0d615da1647633f9536
Reviewed-on: https://go-review.googlesource.com/15656
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agomath: Modf(-0) returns -0,-0
Charlie Dorian [Wed, 7 Oct 2015 22:23:28 +0000 (18:23 -0400)]
math: Modf(-0) returns -0,-0

Fixes #12867

Change-Id: I8ba81c622bce2a77a6142f941603198582eaf8a4
Reviewed-on: https://go-review.googlesource.com/15570
Reviewed-by: Robert Griesemer <gri@golang.org>
9 years agomime/multipart: fix peekBufferSeparatorIndex edge case
mpl [Sun, 4 Oct 2015 21:32:26 +0000 (23:32 +0200)]
mime/multipart: fix peekBufferSeparatorIndex edge case

The case fixed by this change happens when, in func (pr partReader)
Read, the Peek happens to read so that peek looks like:

  "somedata\r\n--Boundary\r"

peekBufferSeparatorIndex was returning (-1, false) because it didn't
find the trailing '\n'.

This was wrong because:

1) It didn't match the documentation: as "\r\n--Boundary" was found, it
should return the index of that pattern, not -1.

2) It lead to an nCopy cut such as:
  "somedata\r| |\n--Boundary\r" instead of "somedata| |\r\n--Boundary\r"
which made the subsequent Read miss the boundary, and eventually end
with a "return 0, io.ErrUnexpectedEOF" case, as reported in:

https://github.com/camlistore/camlistore/issues/642

Change-Id: I1ba78a741bc0c7719e160add9cca932d10f8a615
Reviewed-on: https://go-review.googlesource.com/15269
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

9 years agoC: Add Brian Gitonga Marete's Google email address (Google CLA).
Brian Gitonga Marete [Fri, 9 Oct 2015 12:54:35 +0000 (13:54 +0100)]
C: Add Brian Gitonga Marete's Google email address (Google CLA).

Change-Id: I51edc77ad4621ad8f3908e69dcb7b4eab086b5fe
Reviewed-on: https://go-review.googlesource.com/15680
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agonet/http/httputil: clarify NewSingleHostReverseProxy's Host header behavior
Brad Fitzpatrick [Thu, 8 Oct 2015 20:37:39 +0000 (13:37 -0700)]
net/http/httputil: clarify NewSingleHostReverseProxy's Host header behavior

Fixes #10342

Change-Id: I69c69352016a8fd0b62541128c2e86b242ebbe26
Reviewed-on: https://go-review.googlesource.com/15630
Reviewed-by: Minux Ma <minux@golang.org>
9 years agocmd/internal/obj: improve obj.ProgInfo struct layout
Dave Cheney [Thu, 8 Oct 2015 09:14:03 +0000 (20:14 +1100)]
cmd/internal/obj: improve obj.ProgInfo struct layout

obj.ProgInfo is a field inside obj.Prog, which is currently 320 bytes
on 64bit platforms. By moving the Flags field below the other fields
the size of obj.Prog drops into the 288 byte size class, a saving of
32 bytes per value allocated on the heap.

Change-Id: If8bb12f45328996d7df1d0bac9d1c019d2af73bd
Reviewed-on: https://go-review.googlesource.com/15522
Run-TryBot: Dave Cheney <dave@cheney.net>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agosyscall: fix vet warning in UnixRights
Matthew Dempsky [Thu, 8 Oct 2015 19:43:57 +0000 (12:43 -0700)]
syscall: fix vet warning in UnixRights

Change-Id: Ic19c94fe0af55e17f6c2fcfd36085ccb1584da6f
Reviewed-on: https://go-review.googlesource.com/15608
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/link: pass -Wl,-z,nodelete when creating an ELF shared library
Ian Lance Taylor [Thu, 8 Oct 2015 17:22:50 +0000 (10:22 -0700)]
cmd/link: pass -Wl,-z,nodelete when creating an ELF shared library

Go shared libraries do not support dlclose, and there is no likelihood
that they will suppose dlclose in the future.  Set the DF_1_NODELETE
flag to tell the dynamic linker to not attempt to remove them from
memory.  This makes the shared library act as though every call to
dlopen passed the RTLD_NODELETE flag.

Fixes #12582.
Update #11100.
Update #12873.

Change-Id: Id4b6e90a1b54e2e6fc8355b5fb22c5978fc762b4
Reviewed-on: https://go-review.googlesource.com/15605
Reviewed-by: Michael Hudson-Doyle <michael.hudson@canonical.com>
9 years agoruntime: make aeshash more DOS-proof
Keith Randall [Mon, 31 Aug 2015 23:26:12 +0000 (16:26 -0700)]
runtime: make aeshash more DOS-proof

Improve the aeshash implementation to make it harder to engineer collisions.

1) Scramble the seed before xoring with the input string.  This
   makes it harder to cancel known portions of the seed (like the size)
   because it mixes the per-table seed into those other parts.

2) Use table-dependent seeds for all stripes when hashing >16 byte strings.

For small strings this change uses 4 aesenc ops instead of 3, so it
is somewhat slower.  The first two can run in parallel, though, so
it isn't 33% slower.

benchmark                            old ns/op     new ns/op     delta
BenchmarkHash64-12                   10.2          11.2          +9.80%
BenchmarkHash16-12                   5.71          6.13          +7.36%
BenchmarkHash5-12                    6.64          7.01          +5.57%
BenchmarkHashBytesSpeed-12           30.3          31.9          +5.28%
BenchmarkHash65536-12                2785          2882          +3.48%
BenchmarkHash1024-12                 53.6          55.4          +3.36%
BenchmarkHashStringArraySpeed-12     54.9          56.5          +2.91%
BenchmarkHashStringSpeed-12          18.7          19.2          +2.67%
BenchmarkHashInt32Speed-12           14.8          15.1          +2.03%
BenchmarkHashInt64Speed-12           14.5          14.5          +0.00%

Change-Id: I59ea124b5cb92b1c7e8584008257347f9049996c
Reviewed-on: https://go-review.googlesource.com/14124
Reviewed-by: jcd . <jcd@golang.org>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoruntime: adjust the arm64 memmove and memclr to operate by word as much as they can
Michael Hudson-Doyle [Tue, 22 Sep 2015 02:34:39 +0000 (14:34 +1200)]
runtime: adjust the arm64 memmove and memclr to operate by word as much as they can

Not only is this an obvious optimization:

benchmark                           old MB/s     new MB/s     speedup
BenchmarkMemmove1-4                 35.35        29.65        0.84x
BenchmarkMemmove2-4                 63.78        52.53        0.82x
BenchmarkMemmove3-4                 89.72        73.96        0.82x
BenchmarkMemmove4-4                 109.94       95.73        0.87x
BenchmarkMemmove5-4                 127.60       112.80       0.88x
BenchmarkMemmove6-4                 143.59       126.67       0.88x
BenchmarkMemmove7-4                 157.90       138.92       0.88x
BenchmarkMemmove8-4                 167.18       231.81       1.39x
BenchmarkMemmove9-4                 175.23       252.07       1.44x
BenchmarkMemmove10-4                165.68       261.10       1.58x
BenchmarkMemmove11-4                174.43       263.31       1.51x
BenchmarkMemmove12-4                180.76       267.56       1.48x
BenchmarkMemmove13-4                189.06       284.93       1.51x
BenchmarkMemmove14-4                186.31       284.72       1.53x
BenchmarkMemmove15-4                195.75       281.62       1.44x
BenchmarkMemmove16-4                202.96       439.23       2.16x
BenchmarkMemmove32-4                264.77       775.77       2.93x
BenchmarkMemmove64-4                306.81       1209.64      3.94x
BenchmarkMemmove128-4               357.03       1515.41      4.24x
BenchmarkMemmove256-4               380.77       2066.01      5.43x
BenchmarkMemmove512-4               385.05       2556.45      6.64x
BenchmarkMemmove1024-4              381.23       2804.10      7.36x
BenchmarkMemmove2048-4              379.06       2814.83      7.43x
BenchmarkMemmove4096-4              387.43       3064.96      7.91x
BenchmarkMemmoveUnaligned1-4        28.91        25.40        0.88x
BenchmarkMemmoveUnaligned2-4        56.13        47.56        0.85x
BenchmarkMemmoveUnaligned3-4        74.32        69.31        0.93x
BenchmarkMemmoveUnaligned4-4        97.02        83.58        0.86x
BenchmarkMemmoveUnaligned5-4        110.17       103.62       0.94x
BenchmarkMemmoveUnaligned6-4        124.95       113.26       0.91x
BenchmarkMemmoveUnaligned7-4        142.37       130.82       0.92x
BenchmarkMemmoveUnaligned8-4        151.20       205.64       1.36x
BenchmarkMemmoveUnaligned9-4        166.97       215.42       1.29x
BenchmarkMemmoveUnaligned10-4       148.49       221.22       1.49x
BenchmarkMemmoveUnaligned11-4       159.47       239.57       1.50x
BenchmarkMemmoveUnaligned12-4       163.52       247.32       1.51x
BenchmarkMemmoveUnaligned13-4       167.55       256.54       1.53x
BenchmarkMemmoveUnaligned14-4       175.12       251.03       1.43x
BenchmarkMemmoveUnaligned15-4       192.10       267.13       1.39x
BenchmarkMemmoveUnaligned16-4       190.76       378.87       1.99x
BenchmarkMemmoveUnaligned32-4       259.02       562.98       2.17x
BenchmarkMemmoveUnaligned64-4       317.72       842.44       2.65x
BenchmarkMemmoveUnaligned128-4      355.43       1274.49      3.59x
BenchmarkMemmoveUnaligned256-4      378.17       1815.74      4.80x
BenchmarkMemmoveUnaligned512-4      362.15       2180.81      6.02x
BenchmarkMemmoveUnaligned1024-4     376.07       2453.58      6.52x
BenchmarkMemmoveUnaligned2048-4     381.66       2568.32      6.73x
BenchmarkMemmoveUnaligned4096-4     398.51       2669.36      6.70x
BenchmarkMemclr5-4                  113.83       107.93       0.95x
BenchmarkMemclr16-4                 223.84       389.63       1.74x
BenchmarkMemclr64-4                 421.99       1209.58      2.87x
BenchmarkMemclr256-4                525.94       2411.58      4.59x
BenchmarkMemclr4096-4               581.66       4372.20      7.52x
BenchmarkMemclr65536-4              565.84       4747.48      8.39x
BenchmarkGoMemclr5-4                194.63       160.31       0.82x
BenchmarkGoMemclr16-4               295.30       630.07       2.13x
BenchmarkGoMemclr64-4               480.24       1884.03      3.92x
BenchmarkGoMemclr256-4              540.23       2926.49      5.42x

but it turns out that it's necessary to avoid the GC seeing partially written
pointers.

It's of course possible to be more sophisticated (using ldp/stp to move 16
bytes at a time in the core loop and unrolling the tail copying loops being
the obvious ideas) but I wanted something simple and (reasonably) obviously
correct.

Fixes #12552

Change-Id: Iaeaf8a812cd06f4747ba2f792de1ded738890735
Reviewed-on: https://go-review.googlesource.com/14813
Reviewed-by: Austin Clements <austin@google.com>
9 years agotext/template: fix formatting calls in tests
Alexander Morozov [Wed, 7 Oct 2015 16:10:15 +0000 (09:10 -0700)]
text/template: fix formatting calls in tests

Change-Id: I8e94fa57482149f6ea8f13d02ddcc82d6764ddb8
Reviewed-on: https://go-review.googlesource.com/15496
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agocmd/compile: fix some C to Go translation leftovers
Didier Spezia [Wed, 7 Oct 2015 22:25:25 +0000 (22:25 +0000)]
cmd/compile: fix some C to Go translation leftovers

Following the C to Go translation, some useless variables
were left in the code. In fmt.go, this was harmless.
In lex.go, it broke the error message related to
non-canonical import paths.

Fix it, and remove the useless variables.

The added test case is ignored in the go/types tests, since
the behavior of the non-canonical import path check seems
to be different.

Fixes #11362

Change-Id: Ic9129139ede90357dc79ebf167af638cf44536fa
Reviewed-on: https://go-review.googlesource.com/15580
Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
Reviewed-by: Minux Ma <minux@golang.org>
Run-TryBot: Minux Ma <minux@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/internal/obj, cmd/link, runtime: lots of TLS cleanup
Michael Hudson-Doyle [Tue, 1 Sep 2015 23:23:15 +0000 (11:23 +1200)]
cmd/internal/obj, cmd/link, runtime: lots of TLS cleanup

It's particularly nice to get rid of the android special cases in the linker.

Change-Id: I516363af7ce8a6b2f196fe49cb8887ac787a6dad
Reviewed-on: https://go-review.googlesource.com/14197
Run-TryBot: Michael Hudson-Doyle <michael.hudson@canonical.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agomath/cmplx: make error tolerance test function of expected value
Charlie Dorian [Wed, 7 Oct 2015 03:16:47 +0000 (23:16 -0400)]
math/cmplx: make error tolerance test function of expected value

Copy math package CL 12230 to cmplx package.

Change-Id: I3345b782b84b5b98e2b6a60d8774c7e7cede2891
Reviewed-on: https://go-review.googlesource.com/15500
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agodebug/gosym: clean up and modernize pclntab_test
Ian Lance Taylor [Wed, 7 Oct 2015 14:24:20 +0000 (07:24 -0700)]
debug/gosym: clean up and modernize pclntab_test

The self tests do not need to build the binary; they won't read it.  The
self tests should work on any ELF system.

Use t.Skip instead of panic.  Use internal/testenv.  Don't worry about a
space in the temporary directory name.

Change-Id: I66ef0af90520d330820afa7b6c6b3a132ab27454
Reviewed-on: https://go-review.googlesource.com/15495
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agotest: adjust gccgo expected import errors
Ian Lance Taylor [Wed, 7 Oct 2015 00:57:46 +0000 (17:57 -0700)]
test: adjust gccgo expected import errors

Update two tests for the recently submitted gccgo change
https://golang.org/cl/14259.

Change-Id: Ib18bc87ea512074aa91fd4096d0874b72e2243e5
Reviewed-on: https://go-review.googlesource.com/15493
Reviewed-by: Chris Manghane <cmang@golang.org>
9 years agocmd/go: fix `go test -n`
Nodir Turakulov [Mon, 5 Oct 2015 23:47:31 +0000 (16:47 -0700)]
cmd/go: fix `go test -n`

The <importPath>/_test directory is not actually created in -n mode, so
`go test` fails to write _testmain.go.

Do not write _testmain.go if -n is passed.

Change-Id: I825d5040cacbc9d9a8c89443e5a3f83e6f210ce4
Reviewed-on: https://go-review.googlesource.com/15433
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agocmd/compile/internal/big: fix unused result from testing/quick.Check
Shenghou Ma [Sun, 4 Oct 2015 01:54:15 +0000 (21:54 -0400)]
cmd/compile/internal/big: fix unused result from testing/quick.Check

Update #12834.

Change-Id: If7bbcc249517f2f2d8a7dcbba6411ede92331abe
Reviewed-on: https://go-review.googlesource.com/15381
Reviewed-by: Damian Gryski <dgryski@gmail.com>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agodatabase/sql: fix conversions to and from user-defined types
Gordon Klaus [Mon, 5 Oct 2015 20:12:46 +0000 (22:12 +0200)]
database/sql: fix conversions to and from user-defined types

In particular, don't assume that one reflect.Value can be assigned to another just because they have the same reflect.Kind.

Fixes #12401

Change-Id: Ia4605a5c46557ff8f8f1d44f26d492850666c6d1
Reviewed-on: https://go-review.googlesource.com/15420
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoruntime: don't use duffcopy on Plan 9
David du Colombier [Mon, 5 Oct 2015 21:16:51 +0000 (23:16 +0200)]
runtime: don't use duffcopy on Plan 9

In CL 14836, the implementation of duffcopy on amd64
was changed to replace the use of the MOVQ instructions
by MOVUPS.

However, it broke the build on plan9/amd64, since
Plan 9 doesn't allow floating point in note handler.

This change disables the use of duffcopy on Plan 9.

Fixes #12829.

Change-Id: Ifd5b17b17977a1b631b16c3dfe2dc7ab4ad00507
Reviewed-on: https://go-review.googlesource.com/15421
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
9 years agoarchive/tar: fix numeric overflow issues in readGNUSparseMap0x1
Joe Tsai [Thu, 1 Oct 2015 08:04:24 +0000 (01:04 -0700)]
archive/tar: fix numeric overflow issues in readGNUSparseMap0x1

Motivation:
* The logic to verify the numEntries can overflow and incorrectly
pass, allowing a malicious file to allocate arbitrary memory.
* The use of strconv.ParseInt does not set the integer precision
to 64bit, causing this code to work incorrectly on 32bit machines.

Change-Id: I1b1571a750a84f2dde97cc329ed04fe2342aaa60
Reviewed-on: https://go-review.googlesource.com/15173
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoarchive/tar: add missing error checks to Reader.Next
Joe Tsai [Tue, 6 Oct 2015 08:04:18 +0000 (01:04 -0700)]
archive/tar: add missing error checks to Reader.Next

A recursive call to Reader.Next did not check the error before
trying to use the result, leading to a nil pointer panic.
This specific CL addresses the immediate issue, which is the panic,
but does not solve the root issue, which is due to an integer
overflow in the base-256 parser.

Updates #12435

Change-Id: Ia908671f0f411a409a35e24f2ebf740d46734072
Reviewed-on: https://go-review.googlesource.com/15437
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoencoding/gob: fix typo in documentation
Tormod Erevik Lea [Tue, 6 Oct 2015 14:57:06 +0000 (16:57 +0200)]
encoding/gob: fix typo in documentation

Change-Id: I6a882d9f0bc20b7a8bf73765e055d9344f3f401f
Reviewed-on: https://go-review.googlesource.com/15422
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/link: set the ELF headers of ARM executables that use cgo correctly
Michael Hudson-Doyle [Wed, 23 Sep 2015 03:46:00 +0000 (15:46 +1200)]
cmd/link: set the ELF headers of ARM executables that use cgo correctly

It is generally expected that the ELF flags of a dynamically linked executable
and the libraries it links against match. Go's linker currently always produces
executables with flags that do not declare a float abi (hard, soft) at all, but
when cgo is involved it is unlikely that this matches the system libraries
being linked against -- really the decision about ABI is made by the C compiler
during the invocation of cgo.

This change is basically a port of the code from binutils that parses the
".ARM.attributes" section to check for the tag that declares that the code is
built for the hard-float ABI.

Fixes #7094

Change-Id: I737c8f3b5ed4af545cfc3e86722d03eb83083402
Reviewed-on: https://go-review.googlesource.com/14860
Run-TryBot: Michael Hudson-Doyle <michael.hudson@canonical.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Minux Ma <minux@golang.org>
9 years agocmd/go: given better error when -race is used without cgo
Shenghou Ma [Tue, 6 Oct 2015 02:32:19 +0000 (22:32 -0400)]
cmd/go: given better error when -race is used without cgo

Fixes #12844.

Change-Id: Id51b24aae239fd2e1fb1cd0bc9fe443186301044
Reviewed-on: https://go-review.googlesource.com/15440
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agocmd/cgo: update docs refering to 6g and 6l
Shenghou Ma [Mon, 14 Sep 2015 04:48:02 +0000 (00:48 -0400)]
cmd/cgo: update docs refering to 6g and 6l

Change-Id: Id6cb5e3d40e8a2ded6359aa7fcdc012861cc3994
Reviewed-on: https://go-review.googlesource.com/14545
Reviewed-by: Andrew Gerrand <adg@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agotest: gofmt fixedbugs/issue10975.go
Shenghou Ma [Sun, 4 Oct 2015 01:23:23 +0000 (21:23 -0400)]
test: gofmt fixedbugs/issue10975.go

Change-Id: I772d1bc3e394cdd707f210f2aaff77100d299e24
Reviewed-on: https://go-review.googlesource.com/15380
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agoarchive/tar: expand abilities of TestReader
Joe Tsai [Thu, 1 Oct 2015 09:59:49 +0000 (02:59 -0700)]
archive/tar: expand abilities of TestReader

Motivation:
* There are an increasing number of "one-off" corrupt files added
to make sure that package does not succeed or crash on them.
Instead, allow for the test to specify the error that is expected
to occur (if any).
* Also, fold in the logic to check the MD5 checksum into this
function.

The following tests are being removed:
* TestIncrementalRead: Done by TestReader by using io.CopyBuffer
with a buffer of 8. This achieves the same behavior as this test.
* TestSparseEndToEnd: Since TestReader checks the MD5 checksums
if the input corpus provides them, then this is redundant.
* TestSparseIncrementalRead: Redundant for the same reasons that
TestIncrementalRead is now redundant
* TestNegativeHdrSize: Added to TestReader corpus
* TestIssue10968: Added to TestReader corpus
* TestIssue11169: Added to TestReader corpus

With this change, code coverage did not change: 85.3%

Change-Id: I8550d48657d4dbb8f47dfc3dc280758ef73b47ec
Reviewed-on: https://go-review.googlesource.com/15176
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agocmd/compile/internal/ppc64: fix the epilogue for non-leaf generated methods
Michael Hudson-Doyle [Thu, 24 Sep 2015 23:26:40 +0000 (11:26 +1200)]
cmd/compile/internal/ppc64: fix the epilogue for non-leaf generated methods

This lets us re-enable duffzero.

Fixes #12108

Change-Id: Iefd24d26eaa56067caa2c29ff99cd20a42d8714a
Reviewed-on: https://go-review.googlesource.com/14937
Reviewed-by: Keith Randall <khr@golang.org>
9 years agodoc: fix typo in contributing guide
Katrina Owen [Tue, 6 Oct 2015 03:15:23 +0000 (21:15 -0600)]
doc: fix typo in contributing guide

Change-Id: I6d9a8886cccf1c396ea2dbc659c5bf7548179751
Reviewed-on: https://go-review.googlesource.com/15435
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agoio: simplify documentation for ByteReader
Joe Tsai [Thu, 1 Oct 2015 22:21:40 +0000 (15:21 -0700)]
io: simplify documentation for ByteReader

Let C be whether c valid
Let E be whether err is non-nil

The old comment explicitly says that (~C → E). However, that does call
into question whether (E → ~C), which causes doubts for users.
Without a comment at all, it is obvious that only (E ↔ ~C) makes sense.

Fixes #11308

Change-Id: I5a7d51ceb509057eccca91f57a7e48c9d1c6d112
Reviewed-on: https://go-review.googlesource.com/15256
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agoio: clarify documentation for Seeker
Joe Tsai [Thu, 1 Oct 2015 22:09:11 +0000 (15:09 -0700)]
io: clarify documentation for Seeker

The later part of the docstring simply talks about "offset" but does
not disambiguate what it is relative to. For both the return value
and valid offsets to seek to, it only makes sense in the context of
"offset relative to origin of file".

Fixes #11877

Change-Id: Ic238a407cf8e8fdd64991d98a6584cdc8a51cd6b
Reviewed-on: https://go-review.googlesource.com/15257
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agocmd/go: print `go test -help` to stderr
Nodir Turakulov [Tue, 6 Oct 2015 00:17:58 +0000 (17:17 -0700)]
cmd/go: print `go test -help` to stderr

Usage of all commands is printed to stderr, except go test, which is printed to
stdout. This is inconsistent.

Print `go test -help` to stderr instead.

R=rsc@golang.org

Change-Id: I079f4788134bf9aedcccc26838879eedad1c925e
Reviewed-on: https://go-review.googlesource.com/15434
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agonet: remove imports of strconv
Ian Lance Taylor [Mon, 5 Oct 2015 18:05:10 +0000 (11:05 -0700)]
net: remove imports of strconv

The net package already has support for limited uses of the strconv
package.  Despite this, a few uses of strconv have crept in over time.
Remove them and use the existing net support instead.

Change-Id: Icdb4bdaa8e1197f1119a96cddcf548ed4a551b74
Reviewed-on: https://go-review.googlesource.com/15400
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/cgo: fix panic on references to non-existing C types
Didier Spezia [Sat, 3 Oct 2015 15:50:45 +0000 (15:50 +0000)]
cmd/cgo: fix panic on references to non-existing C types

cgo panics in Package.rewriteRef for:

var a = C.enum_test(1)
or
p := new(C.enum_test)

when the corresponding enum type is not defined.

Check nil values for Type fields and issue a proper
error instead.

Fixes #11097
Updates #12160

Change-Id: I5821d29097ef0a36076ec5273125b09846c7d832
Reviewed-on: https://go-review.googlesource.com/15264
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agonet/http: add response body to 413 and 400 errors
Kenny Grant [Sun, 27 Sep 2015 13:53:18 +0000 (14:53 +0100)]
net/http: add response body to 413 and 400 errors

The existing serve() method returns a zero-length response body when
it encounters an error, which results in a blank page and no visible
error in browsers.

This change sends a response body explaining the error for display in browsers.

Fixes #12745

Change-Id: I9dc3b95ad88cb92c18ced51f6b52bd3b2c1b974c
Reviewed-on: https://go-review.googlesource.com/15018
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/internal/obj/x86: Add initial VEX support.
Ilya Tocar [Tue, 1 Sep 2015 11:56:37 +0000 (14:56 +0300)]
cmd/internal/obj/x86: Add initial VEX support.

Support VZEROUPPER, VMOVNTDQ, VMOVDQU, VMOVDQA.
Use MOVHD* for names, where HD stands for HexaDeca (16).

Change-Id: I9b1ea52e7ef0714a3d2aeb31ec1823fe509a047e
Reviewed-on: https://go-review.googlesource.com/14127
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agopprof: improve sampling for heap profiling
Raul Silvera [Mon, 14 Sep 2015 21:03:45 +0000 (14:03 -0700)]
pprof: improve sampling for heap profiling

The current heap sampling introduces some bias that interferes
with unsampling, producing unexpected heap profiles.
The solution is to use a Poisson process to generate the
sampling points, using the formulas described at
https://en.wikipedia.org/wiki/Poisson_process

This fixes #12620

Change-Id: If2400809ed3c41de504dd6cff06be14e476ff96c
Reviewed-on: https://go-review.googlesource.com/14590
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Minux Ma <minux@golang.org>
Run-TryBot: Minux Ma <minux@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/compile: do not let regopt use REGTMP on ppc64
Michael Hudson-Doyle [Mon, 5 Oct 2015 03:07:59 +0000 (16:07 +1300)]
cmd/compile: do not let regopt use REGTMP on ppc64

ppc64 codegen assumes that it is OK to stomp on r31 at any time, but it is not
excluded from the set of registers that regopt is allowed to use.

Fixes #12597

Change-Id: I29c7655e32abd22f3c21d88427b73e4fca055233
Reviewed-on: https://go-review.googlesource.com/15245
Reviewed-by: Minux Ma <minux@golang.org>
9 years agocmd/compile: escape analysis, don't always escape variadic args
David Chase [Wed, 30 Sep 2015 18:41:00 +0000 (14:41 -0400)]
cmd/compile: escape analysis, don't always escape variadic args

Turns out the summary information for the ... args was
already correctly computed, all that lacked was to make
use of it and correct tests that documented our prior
deficiencies.

Fixes #12006

Change-Id: Ie8adfab7547f179391d470679598f0904aabf9f7
Reviewed-on: https://go-review.googlesource.com/15200
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
9 years agocmd/go: remove the bootstrap with make.bash error message
Shenghou Ma [Thu, 1 Oct 2015 01:27:30 +0000 (21:27 -0400)]
cmd/go: remove the bootstrap with make.bash error message

The runtime/zgoos_$GOOS.go and runtime/zgoarch_$GOARCH.go files
are in the repository now, so the message is actually incorrect
(running make.bash won't generate those). The reason is probably
wrong $GOROOT.

Change-Id: I8dc125594c52d666eca91fd5af48b60d12d599b8
Reviewed-on: https://go-review.googlesource.com/15221
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/yacc: fix copying action code when line comment is disabled
Shenghou Ma [Mon, 14 Sep 2015 05:28:57 +0000 (01:28 -0400)]
cmd/yacc: fix copying action code when line comment is disabled

Fixes #12601.

Change-Id: I0be69ffe9ba19934aaef1651845c725708db77de
Reviewed-on: https://go-review.googlesource.com/14546
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/cgo: remove support for CGOPKGPATH
Didier Spezia [Sat, 3 Oct 2015 17:42:52 +0000 (17:42 +0000)]
cmd/cgo: remove support for CGOPKGPATH

CGOPKGPATH variable was undocumented, but it is not needed anymore.

It was used before the existence of the go tool to tell cgo the full
path of the package that it was building, which in turn set the name
of the shared library that cgo expected to load back when cgo used
shared libraries. CGOPKGPATH no longer does anything useful;
it just affects the comments in the generated header file.

Remove it to avoid any future confusion.

Fixes #11852

Change-Id: Ieb452e5bbcfd05b87a4a3618b5b8f44423341858
Reviewed-on: https://go-review.googlesource.com/15266
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/compile/internal/gc: fix the build
Brad Fitzpatrick [Sat, 3 Oct 2015 16:35:40 +0000 (09:35 -0700)]
cmd/compile/internal/gc: fix the build

Commit acc90c5 passed the trybots, lingered for weeks, and in the
meantime the type of this variable changed to a bool. I didn't rebase
and re-run the trybots before submitting.

Fixes #12832

Change-Id: If24fda227edd8207f8069c67f1c45f08e6ac215a
Reviewed-on: https://go-review.googlesource.com/15286
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agomath/big: check return value from quick.Check() for GCD tests
Damian Gryski [Sat, 3 Oct 2015 10:06:32 +0000 (12:06 +0200)]
math/big: check return value from quick.Check() for GCD tests

Change-Id: I46c12aaaf453365c157604dfb1486605cfefd7af
Reviewed-on: https://go-review.googlesource.com/15263
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agomath: optimize ceil/floor functions on amd64
Ilya Tocar [Mon, 14 Sep 2015 15:42:39 +0000 (18:42 +0300)]
math: optimize ceil/floor functions on amd64

Use SSE 4.1 rounding instruction to perform rounding
Results (haswell):

name      old time/op  new time/op  delta
Floor-48  2.71ns ± 0%  1.87ns ± 1%  -31.17%  (p=0.000 n=16+19)
Ceil-48   3.09ns ± 3%  2.16ns ± 0%  -30.16%  (p=0.000 n=19+12)

Change-Id: If63715879eed6530b1eb4fc96132d827f8f43909
Reviewed-on: https://go-review.googlesource.com/14561
Reviewed-by: Klaus Post <klauspost@gmail.com>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
9 years agocmd/compile: Fix `internal compiler: getinarg: not a func` when returning invalid...
acanino [Sat, 22 Aug 2015 02:24:20 +0000 (22:24 -0400)]
cmd/compile: Fix `internal compiler: getinarg: not a func` when returning invalid interface.

Internal error arose from calling methodfunc on a invalid interface
field during the implements check. int obviously isn't a function,
and errors on getinarg...

for im := iface.Type; im != nil; im = im.Down {
  imtype = methodfunc(im.Type, nil)
  // ...
}

Fix handles the internal compiler error, but does not throw an
additional error, i.e. the following code will error on the I
interface, but type A will pass the implements check since
'Read(string) string' is implemented and 'int' is skipped

type I interface {
  Read(string) string
  int
}

type A struct {
}

func (a *A) Read(s string) string {
  return s
}

func New() I {
  return new(A)
}

Fixes #10975

Change-Id: I4b54013afb2814db3f315515f0c742d8631ca500
Reviewed-on: https://go-review.googlesource.com/13747
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agodoc: update release tag in source directions to go1.5.1
Austin Clements [Fri, 2 Oct 2015 19:25:01 +0000 (15:25 -0400)]
doc: update release tag in source directions to go1.5.1

Change-Id: I8da1c7a86adf6672e5e5c44cbab422706833c1da
Reviewed-on: https://go-review.googlesource.com/15350
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agoruntime: use 4 byte writes in amd64p32 memmove/memclr
Austin Clements [Fri, 2 Oct 2015 22:17:54 +0000 (18:17 -0400)]
runtime: use 4 byte writes in amd64p32 memmove/memclr

Currently, amd64p32's memmove and memclr use 8 byte writes as much as
possible and 1 byte writes for the tail of the object. However, if an
object ends with a 4 byte pointer at an 8 byte aligned offset, this
may copy/zero the pointer field one byte at a time, allowing the
garbage collector to observe a partially copied pointer.

Fix this by using 4 byte writes instead of 8 byte writes.

Updates #12552.

Change-Id: I13324fd05756fb25ae57e812e836f0a975b5595c
Reviewed-on: https://go-review.googlesource.com/15370
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
9 years agoruntime: adjust huge page flags only on huge page granularity
Austin Clements [Wed, 30 Sep 2015 15:52:54 +0000 (11:52 -0400)]
runtime: adjust huge page flags only on huge page granularity

This fixes an issue where the runtime panics with "out of memory" or
"cannot allocate memory" even though there's ample memory by reducing
the number of memory mappings created by the memory allocator.

Commit 7e1b61c worked around issue #8832 where Linux's transparent
huge page support could dramatically increase the RSS of a Go process
by setting the MADV_NOHUGEPAGE flag on any regions of pages released
to the OS with MADV_DONTNEED. This had the side effect of also
increasing the number of VMAs (memory mappings) in a Go address space
because a separate VMA is needed for every region of the virtual
address space with different flags. Unfortunately, by default, Linux
limits the number of VMAs in an address space to 65530, and a large
heap can quickly reach this limit when the runtime starts scavenging
memory.

This commit dramatically reduces the number of VMAs. It does this
primarily by only adjusting the huge page flag at huge page
granularity. With this change, on amd64, even a pessimal heap that
alternates between MADV_NOHUGEPAGE and MADV_HUGEPAGE must reach 128GB
to reach the VMA limit. Because of this rounding to huge page
granularity, this change is also careful to leave large used and
unused regions huge page-enabled.

This change reduces the maximum number of VMAs during the runtime
benchmarks with GODEBUG=scavenge=1 from 692 to 49.

Fixes #12233.

Change-Id: Ic397776d042f20d53783a1cacf122e2e2db00584
Reviewed-on: https://go-review.googlesource.com/15191
Reviewed-by: Keith Randall <khr@golang.org>
9 years agoruntime: remove sweep wait loop in finishsweep_m
Austin Clements [Sat, 26 Sep 2015 18:00:57 +0000 (14:00 -0400)]
runtime: remove sweep wait loop in finishsweep_m

In general, finishsweep_m must block until any spans that are
concurrently being swept have been swept. It accomplishes this by
looping over all spans, which, as in the previous commit, takes
~1ms/heap GB. Unfortunately, we do this during the STW sweep
termination phase, so multi-gigabyte heaps can push our STW time past
10ms.

However, there's no need to do this wait if the world is stopped
because, in effect, stopping the world already had to wait for
anything that was sweeping (and if it didn't, the wait in
finishsweep_m would deadlock). Hence, we can simply skip this loop if
the world is stopped, such as during sweep termination. In fact,
currently all calls to finishsweep_m are STW, but this hasn't always
been the case and may not be the case in the future, so we keep the
logic around.

For 24GB heaps, this reduces max pause time by 75% relative to tip and
by 90% relative to Go 1.5. Notably, all pauses are now well under
10ms. Here are the results for the garbage benchmark:

               ------------- max pause ------------
Heap   Procs   after change   before change   1.5.1
24GB     12        3.8ms          16ms         37ms
24GB      4        3.7ms          16ms         37ms
 4GB      4        3.7ms           3ms        6.9ms

In the 4GB/4P case, it seems the "before change" run got lucky: the
max went up, but the 99%ile pause time went down from 3ms to 2.04ms.

Change-Id: Ica22189559f231d408ef2815019c9dbb5f38bf31
Reviewed-on: https://go-review.googlesource.com/15071
Reviewed-by: Rick Hudson <rlh@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoruntime: remove in-use page count loop from STW
Austin Clements [Sat, 26 Sep 2015 16:31:59 +0000 (12:31 -0400)]
runtime: remove in-use page count loop from STW

In order to compute the sweep ratio, the runtime needs to know how
many pages belong to spans in state _MSpanInUse. Currently it finds
this out by looping over all spans during mark termination. However,
this takes ~1ms/heap GB, so multi-gigabyte heaps can quickly push our
STW time past 10ms.

Replace the loop with an actively maintained count of in-use pages.

For multi-gigabyte heaps, this reduces max mark termination pause time
by 75%–90% relative to tip and by 85%–95% relative to Go 1.5.1. This
shifts the longest pause time for large heaps to the sweep termination
phase, so it only slightly decreases max pause time, though it roughly
halves mean pause time. Here are the results for the garbage
benchmark:

               ---- max mark termination pause ----
Heap   Procs   after change   before change   1.5.1
24GB     12        1.9ms          18ms         37ms
24GB      4        3.7ms          18ms         37ms
 4GB      4        920µs         3.8ms        6.9ms

Fixes #11484.

Change-Id: Ia2d28bb8a1e4f1c3b8ebf79fb203f12b9bf114ac
Reviewed-on: https://go-review.googlesource.com/15070
Reviewed-by: Rick Hudson <rlh@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoruntime: scan objects with finalizers concurrently
Austin Clements [Thu, 24 Sep 2015 18:39:27 +0000 (14:39 -0400)]
runtime: scan objects with finalizers concurrently

This reduces pause time by ~25% relative to tip and by ~50% relative
to Go 1.5.1.

Currently one of the steps of STW mark termination is to loop (in
parallel) over all spans to find objects with finalizers in order to
mark all objects reachable from these objects and to treat the
finalizer special as a root. Unfortunately, even if there are no
finalizers at all, this loop takes roughly 1 ms/heap GB/core, so
multi-gigabyte heaps can quickly push our STW time past 10ms.

Fix this by moving this scan from mark termination to concurrent scan,
where it can run in parallel with mutators. The loop itself could also
be optimized, but this cost is small compared to concurrent marking.

Making this scan concurrent introduces two complications:

1) The scan currently walks the specials list of each span without
locking it, which is safe only with the world stopped. We fix this by
speculatively checking if a span has any specials (the vast majority
won't) and then locking the specials list only if there are specials
to check.

2) An object can have a finalizer set after concurrent scan, in which
case it won't have been marked appropriately by concurrent scan. If
the finalizer is a closure and is only reachable from the special, it
could be swept before it is run. Likewise, if the object is not marked
yet when the finalizer is set and then becomes unreachable before it
is marked, other objects reachable only from it may be swept before
the finalizer function is run. We fix this issue by making
addfinalizer ensure the same marking invariants as markroot does.

For multi-gigabyte heaps, this reduces max pause time by 20%–30%
relative to tip (depending on GOMAXPROCS) and by ~50% relative to Go
1.5.1 (where this loop was neither concurrent nor parallel). Here are
the results for the garbage benchmark:

               ---------------- max pause ----------------
Heap   Procs   Concurrent scan   STW parallel scan   1.5.1
24GB     12         18ms              23ms            37ms
24GB      4         18ms              25ms            37ms
 4GB      4         3.8ms            4.9ms           6.9ms

In all cases, 95%ile pause time is similar to the max pause time. This
also improves mean STW time by 10%–30%.

Fixes #11485.

Change-Id: I9359d8c3d120a51d23d924b52bf853a1299b1dfd
Reviewed-on: https://go-review.googlesource.com/14982
Reviewed-by: Rick Hudson <rlh@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoruntime: introduce gcMode type for GC modes
Austin Clements [Thu, 24 Sep 2015 18:30:09 +0000 (14:30 -0400)]
runtime: introduce gcMode type for GC modes

Currently, the GC modes constants are untyped and functions pass them
around as ints. Clean this up by introducing a proper type for these
constant.

Change-Id: Ibc022447bdfa203644921fbb548312d7e2272e8d
Reviewed-on: https://go-review.googlesource.com/14981
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoruntime: fix out-of-date comment on gcWork usage
Austin Clements [Thu, 24 Sep 2015 18:24:32 +0000 (14:24 -0400)]
runtime: fix out-of-date comment on gcWork usage

Change-Id: I3c21ffa80a5c14911e07238b1f64bec686ed7b72
Reviewed-on: https://go-review.googlesource.com/14980
Reviewed-by: Minux Ma <minux@golang.org>
9 years agosyscall: skip a couple tests when running under Kubernetes
Brad Fitzpatrick [Thu, 1 Oct 2015 22:40:55 +0000 (15:40 -0700)]
syscall: skip a couple tests when running under Kubernetes

Update #12815

Change-Id: I3bf6de74bc8ab07000fe9a4308299839ef20632f
Reviewed-on: https://go-review.googlesource.com/15283
Reviewed-by: Evan Brown <evanbrown@google.com>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agoreadme: emphasize issue tracker is for bugs/proposals
Robert Griesemer [Fri, 2 Oct 2015 17:55:27 +0000 (10:55 -0700)]
readme: emphasize issue tracker is for bugs/proposals

Removed direct link to issue tracker in the README - it makes it too
easy to use it for a question - and it's abused multiple times a day
for questions. It's easy enough to find it if there's a real issue
to report.

Added sentence to point people at golang-nuts and the new forum.

Change-Id: If75bab888cda064aceeefc49ef672fbb964f8f54
Reviewed-on: https://go-review.googlesource.com/15284
Reviewed-by: Jason Buberel <jbuberel@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agodatabase/sql: fix case where Stmt.Close discards error
Ian Gudger [Thu, 1 Oct 2015 10:29:27 +0000 (03:29 -0700)]
database/sql: fix case where Stmt.Close discards error

Fixes a case where the Stmt.Close() function in database/sql discards any error generated by the Close() function of the contained driverStmt.

Fixes #12798

Change-Id: I40384d6165856665b062d15a643e4ecc09d63fda
Reviewed-on: https://go-review.googlesource.com/15178
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agomisc/cgo/testsanitizers: skip test for version of clang before 3.6
Ian Lance Taylor [Fri, 2 Oct 2015 14:04:34 +0000 (07:04 -0700)]
misc/cgo/testsanitizers: skip test for version of clang before 3.6

I've tested with clang 3.6.  The builder is running 3.5, and fails.

Fixes #12814.

Change-Id: I087fb75c3a24bed7f7fa5e9d7a1444590a316d63
Reviewed-on: https://go-review.googlesource.com/15259
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agoruntime: darwin/386 entrypoint for c-archive
David Crawshaw [Mon, 20 Jul 2015 16:33:39 +0000 (12:33 -0400)]
runtime: darwin/386 entrypoint for c-archive

Change-Id: Ic22597b5e2824cffe9598cb9b506af3426c285fd
Reviewed-on: https://go-review.googlesource.com/12412
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agoruntime: adjust the ppc64x memmove and memclr to copy by word as much as it can
Michael Hudson-Doyle [Tue, 22 Sep 2015 10:35:52 +0000 (22:35 +1200)]
runtime: adjust the ppc64x memmove and memclr to copy by word as much as it can

Issue #12552 can happen on ppc64 too, although much less frequently in my
testing. I'm fairly sure this fixes it (2 out of 200 runs of oracle.test failed
without this change and 0 of 200 failed with it). It's also a lot faster for
large moves/clears:

name           old speed      new speed       delta
Memmove1-6      157MB/s ± 9%    144MB/s ± 0%    -8.20%         (p=0.004 n=10+9)
Memmove2-6      281MB/s ± 1%    249MB/s ± 1%   -11.53%        (p=0.000 n=10+10)
Memmove3-6      376MB/s ± 1%    328MB/s ± 1%   -12.64%        (p=0.000 n=10+10)
Memmove4-6      475MB/s ± 4%    345MB/s ± 1%   -27.28%         (p=0.000 n=10+8)
Memmove5-6      540MB/s ± 1%    393MB/s ± 0%   -27.21%        (p=0.000 n=10+10)
Memmove6-6      609MB/s ± 0%    423MB/s ± 0%   -30.56%         (p=0.000 n=9+10)
Memmove7-6      659MB/s ± 0%    468MB/s ± 0%   -28.99%         (p=0.000 n=8+10)
Memmove8-6      705MB/s ± 0%   1295MB/s ± 1%   +83.73%          (p=0.000 n=9+9)
Memmove9-6      740MB/s ± 1%   1241MB/s ± 1%   +67.61%         (p=0.000 n=10+8)
Memmove10-6     780MB/s ± 0%   1162MB/s ± 1%   +48.95%         (p=0.000 n=10+9)
Memmove11-6     811MB/s ± 0%   1180MB/s ± 0%   +45.58%          (p=0.000 n=8+9)
Memmove12-6     820MB/s ± 1%   1073MB/s ± 1%   +30.83%         (p=0.000 n=10+9)
Memmove13-6     849MB/s ± 0%   1068MB/s ± 1%   +25.87%        (p=0.000 n=10+10)
Memmove14-6     877MB/s ± 0%    911MB/s ± 0%    +3.83%        (p=0.000 n=10+10)
Memmove15-6     893MB/s ± 0%    922MB/s ± 0%    +3.25%         (p=0.000 n=10+9)
Memmove16-6     897MB/s ± 1%   2418MB/s ± 1%  +169.67%         (p=0.000 n=10+9)
Memmove32-6     908MB/s ± 0%   3927MB/s ± 2%  +332.64%         (p=0.000 n=10+8)
Memmove64-6    1.11GB/s ± 0%   5.59GB/s ± 0%  +404.64%          (p=0.000 n=9+9)
Memmove128-6   1.25GB/s ± 0%   6.71GB/s ± 2%  +437.49%         (p=0.000 n=9+10)
Memmove256-6   1.33GB/s ± 0%   7.25GB/s ± 1%  +445.06%        (p=0.000 n=10+10)
Memmove512-6   1.38GB/s ± 0%   8.87GB/s ± 0%  +544.43%        (p=0.000 n=10+10)
Memmove1024-6  1.40GB/s ± 0%  10.00GB/s ± 0%  +613.80%        (p=0.000 n=10+10)
Memmove2048-6  1.41GB/s ± 0%  10.65GB/s ± 0%  +652.95%         (p=0.000 n=9+10)
Memmove4096-6  1.42GB/s ± 0%  11.01GB/s ± 0%  +675.37%         (p=0.000 n=8+10)
Memclr5-6       269MB/s ± 1%    264MB/s ± 0%    -1.80%        (p=0.000 n=10+10)
Memclr16-6      600MB/s ± 0%    887MB/s ± 1%   +47.83%        (p=0.000 n=10+10)
Memclr64-6     1.06GB/s ± 0%   2.91GB/s ± 1%  +174.58%         (p=0.000 n=8+10)
Memclr256-6    1.32GB/s ± 0%   6.58GB/s ± 0%  +399.86%         (p=0.000 n=9+10)
Memclr4096-6   1.42GB/s ± 0%  10.90GB/s ± 0%  +668.03%         (p=0.000 n=8+10)
Memclr65536-6  1.43GB/s ± 0%  11.37GB/s ± 0%  +697.83%          (p=0.000 n=9+8)
GoMemclr5-6     359MB/s ± 0%    360MB/s ± 0%    +0.46%        (p=0.000 n=10+10)
GoMemclr16-6    750MB/s ± 0%   1264MB/s ± 1%   +68.45%        (p=0.000 n=10+10)
GoMemclr64-6   1.17GB/s ± 0%   3.78GB/s ± 1%  +223.58%         (p=0.000 n=10+9)
GoMemclr256-6  1.35GB/s ± 0%   7.47GB/s ± 0%  +452.44%        (p=0.000 n=10+10)

Update #12552

Change-Id: I7192e9deb9684a843aed37f58a16a4e29970e893
Reviewed-on: https://go-review.googlesource.com/14840
Reviewed-by: Minux Ma <minux@golang.org>
9 years agoruntime: drop sigfwd from signal forwarding unsupported platforms
Mikio Hara [Sun, 12 Jul 2015 00:23:37 +0000 (09:23 +0900)]
runtime: drop sigfwd from signal forwarding unsupported platforms

This change splits signal_unix.go into signal_unix.go and
signal2_unix.go and removes the fake symbol sigfwd from signal
forwarding unsupported platforms for clarification purpose.

Change-Id: I205eab5cf1930fda8a68659b35cfa9f3a0e67ca6
Reviewed-on: https://go-review.googlesource.com/12062
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agoarchive/tar: make Reader.Read errors persistent
Joe Tsai [Thu, 1 Oct 2015 10:08:18 +0000 (03:08 -0700)]
archive/tar: make Reader.Read errors persistent

If the stream is in an inconsistent state, it does not make sense
that Reader.Read can be called and possibly succeed.

Change-Id: I9d1c5a1300b2c2b45232188aa7999e350809dcf2
Reviewed-on: https://go-review.googlesource.com/15177
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

9 years agonet: make /etc/hosts lookups case-insensitive
Burcu Dogan [Thu, 1 Oct 2015 19:01:50 +0000 (15:01 -0400)]
net: make /etc/hosts lookups case-insensitive

The native Go host resolver was behaving differently than libc
and the entries in the /etc/hosts were handled in a case sensitive
way. In order to be compatible with libc's resolver, /etc/hosts
lookups must be case-insensitive.

Fixes #12806.

Change-Id: I3c14001abffadf7458fd1a027c91e6438a87f285
Reviewed-on: https://go-review.googlesource.com/15321
Run-TryBot: Burcu Dogan <jbd@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agogo/types: fix incorrect comment at Info.Implicits.
Alan Donovan [Thu, 1 Oct 2015 19:06:20 +0000 (15:06 -0400)]
go/types: fix incorrect comment at Info.Implicits.

Change-Id: Ibd24e1567cb03f7f00f3cbe381bedd6c5215af35
Reviewed-on: https://go-review.googlesource.com/15320
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
9 years agocmd/compile: repair escape analysis of range &fixedArray
David Chase [Mon, 28 Sep 2015 20:13:57 +0000 (16:13 -0400)]
cmd/compile: repair escape analysis of range &fixedArray

The existing test did not take into account the implicit
dereference of &fixedArray and thus heap-escaped when it
was not necessary.

Also added a detailed test for this and related cases.

Fixes #12588

Change-Id: I951e9684a093082ccdca47710f69f4366bd6b3cf
Reviewed-on: https://go-review.googlesource.com/15130
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
9 years agoruntime/trace: reduce memory usage for trace stress tests on openbsd/arm
Joel Sing [Thu, 1 Oct 2015 12:34:48 +0000 (22:34 +1000)]
runtime/trace: reduce memory usage for trace stress tests on openbsd/arm

Reduce allocation to avoid running out of memory on the openbsd/arm builder,
until issue/12032 is resolved.

Update issue #12032

Change-Id: Ibd513829ffdbd0db6cd86a0a5409934336131156
Reviewed-on: https://go-review.googlesource.com/15242
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
9 years agoruntime: handle sysReserve failure in mHeap_SysAlloc
Joel Sing [Sat, 26 Sep 2015 17:56:05 +0000 (03:56 +1000)]
runtime: handle sysReserve failure in mHeap_SysAlloc

sysReserve will return nil on failure - correctly handle this case and return
nil to the caller. Currently, a failure will result in h.arena_end being set
to psize, h.arena_used being set to zero and fun times ensue.

On the openbsd/arm builder this has resulted in:

  runtime: address space conflict: map(0x0) = 0x40946000
  fatal error: runtime: address space conflict

When it should be reporting out of memory instead.

Change-Id: Iba828d5ee48ee1946de75eba409e0cfb04f089d4
Reviewed-on: https://go-review.googlesource.com/15056
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
9 years agoruntime: update comment to match function name
Jeremy Schlatter [Thu, 1 Oct 2015 07:55:51 +0000 (07:55 +0000)]
runtime: update comment to match function name

Change-Id: I8f22434ade576cc7e3e6d9f357bba12c1296e3d1
Reviewed-on: https://go-review.googlesource.com/15250
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agotext/template: change IsTrue to take interface{} instead of reflect.Value.
David Symonds [Thu, 1 Oct 2015 04:21:37 +0000 (14:21 +1000)]
text/template: change IsTrue to take interface{} instead of reflect.Value.

This is a follow-up to a326c3e to avoid reflect being in the API.

Fixes #12801.

Change-Id: Ic4c2e592e2c35b5911f75d88f1d9c44787c80f30
Reviewed-on: https://go-review.googlesource.com/15240
Run-TryBot: David Symonds <dsymonds@golang.org>
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agoarchive/tar: fix bugs with sparseFileReader
Joe Tsai [Mon, 28 Sep 2015 23:38:16 +0000 (16:38 -0700)]
archive/tar: fix bugs with sparseFileReader

The sparseFileReader is prone to two different forms of
denial-of-service attacks:
* A malicious tar file can cause an infinite loop
* A malicious tar file can cause arbitrary panics

This results because of poor error checking/handling, which this
CL fixes. While we are at it, add a plethora of unit tests to
test for possible malicious inputs.

Change-Id: I2f9446539d189f3c1738a1608b0ad4859c1be929
Reviewed-on: https://go-review.googlesource.com/15115
Reviewed-by: Andrew Gerrand <adg@golang.org>
Run-TryBot: Andrew Gerrand <adg@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocompress/zlib: detect truncated streams
Joe Tsai [Tue, 22 Sep 2015 08:17:01 +0000 (01:17 -0700)]
compress/zlib: detect truncated streams

Reader failed to detect truncated streams since calls to
io.ReadFull did not check if the error is io.EOF.

Change-Id: I86c497519daaaccefc6eb5617ddcd8fd3b99f51b
Reviewed-on: https://go-review.googlesource.com/14835
Reviewed-by: Nigel Tao <nigeltao@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agoruntime, runtime/cgo: support using msan on cgo code
Ian Lance Taylor [Wed, 30 Sep 2015 04:24:13 +0000 (21:24 -0700)]
runtime, runtime/cgo: support using msan on cgo code

The memory sanitizer (msan) is a nice compiler feature that can
dynamically check for memory errors in C code.  It's not useful for Go
code, since Go is memory safe.  But it is useful to be able to use the
memory sanitizer on C code that is linked into a Go program via cgo.
Without this change it does not work, as msan considers memory passed
from Go to C as uninitialized.

To make this work, change the runtime to call the C mmap function when
using cgo.  When using msan the mmap call will be intercepted and marked
as returning initialized memory.

Work around what appears to be an msan bug by calling malloc before we
call mmap.

Change-Id: I8ab7286d7595ae84782f68a98bef6d3688b946f9
Reviewed-on: https://go-review.googlesource.com/15170
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
9 years agoencoding/binary: document that Read returns io.EOF iff zero bytes are read
Joe Tsai [Wed, 30 Sep 2015 20:35:03 +0000 (13:35 -0700)]
encoding/binary: document that Read returns io.EOF iff zero bytes are read

Also add a unit test to lock this behavior into the API.

Fixes #12016

Change-Id: Ib6ec6e7948f0705f3504ede9143b5dc4e790fc44
Reviewed-on: https://go-review.googlesource.com/15171
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoruntime: test that periodic GC works
Austin Clements [Wed, 5 Aug 2015 15:35:28 +0000 (11:35 -0400)]
runtime: test that periodic GC works

We've broken periodic GC a few times without noticing because there's
no test for it, partly because you have to wait two minutes to see if
it happens. This exposes control of the periodic GC timeout to runtime
tests and adds a test that cranks it down to zero and sleeps for a bit
to make sure periodic GCs happen.

Change-Id: I3ec44e967e99f4eda752f85c329eebd18b87709e
Reviewed-on: https://go-review.googlesource.com/13169
Reviewed-by: Rick Hudson <rlh@golang.org>
Run-TryBot: Austin Clements <austin@google.com>

9 years agogo/format: handle whitespace-only input correctly
Robert Griesemer [Wed, 30 Sep 2015 00:34:32 +0000 (17:34 -0700)]
go/format: handle whitespace-only input correctly

Applied identical change to cmd/gofmt/internal.go.

Fixes #11275.

Change-Id: Icb4bf0460c94c9e2830dd0d62c69376774cbda30
Reviewed-on: https://go-review.googlesource.com/15154
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agogo/format, cmd/gofmt: avoid dependency on internal package format
Robert Griesemer [Mon, 28 Sep 2015 21:42:21 +0000 (14:42 -0700)]
go/format, cmd/gofmt: avoid dependency on internal package format

Fixes #11844.

Change-Id: I32edd39e79f7c9bdc132c49bd06081f35dac245d
Reviewed-on: https://go-review.googlesource.com/15114
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agocrypto/x509: parse CSRs with a critical flag in the requested extensions.
Adam Langley [Mon, 27 Jul 2015 23:54:30 +0000 (16:54 -0700)]
crypto/x509: parse CSRs with a critical flag in the requested extensions.

The format for a CSR is horribly underspecified and we had a mistake.
The code was parsing the attributes from the CSR as a
pkix.AttributeTypeAndValueSET, which is only almost correct: it works so
long as the requested extensions don't contain the optional “critical”
flag.

Unfortunately this mistake is exported somewhat in the API and the
Attributes field of a CSR actually has the wrong type. I've moved this
field to the bottom of the structure and updated the comment to reflect
this.

The Extensions and other fields of the CSR structure can be saved
however and this change does that.

Fixes #11897.

Change-Id: If8e2f5c21934800b72b041e38691efc3e897ecf1
Reviewed-on: https://go-review.googlesource.com/12717
Reviewed-by: Rob Pike <r@golang.org>
9 years agocrypto/x509: make verification of an empty certificate consistent across platforms.
Adam Langley [Sun, 30 Aug 2015 16:45:26 +0000 (09:45 -0700)]
crypto/x509: make verification of an empty certificate consistent across platforms.

Platform-specific verification needs the ASN.1 contents of a certificate
but that might not be provided if the Certificate was not created by
ParseCertificate. In order to avoid a panic on Windows, and to make
behaviour consistent across platforms, this change causes verification
to fail when the ASN.1 contents of a certificate are not available.

Fixes #12184

Change-Id: I4395d74934e675c179eaf4cded1094a756e478bb
Reviewed-on: https://go-review.googlesource.com/14053
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agomath/big: correct documentation for ProbablyPrime.
Adam Langley [Sun, 30 Aug 2015 16:21:35 +0000 (09:21 -0700)]
math/big: correct documentation for ProbablyPrime.

As akalin points out in the bug, the comment previously claimed that the
probability that the input is prime given that the function returned
true is 1 - ¼ⁿ. But that's wrong: the correct statement is that the
probability of the function returning false given a composite input is
1 - ¼ⁿ.

This is not nearly as helpful, but at least it's truthful. A number of
other (correct) expressions are suggested on the bug, but I think that
the simplier one is preferable.

This change also notes that the function is not suitable for
adversarial inputs since it's deterministic.

Fixes #12274.

Change-Id: I6a0871d103b126ee5a5a922a8c6993055cb7b1ed
Reviewed-on: https://go-review.googlesource.com/14052
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocrypto/tls: better error messages when PEM inputs are switched.
Adam Langley [Sun, 30 Aug 2015 17:23:30 +0000 (10:23 -0700)]
crypto/tls: better error messages when PEM inputs are switched.

This change causes the types of skipped PEM blocks to be recorded when
no certificate or private-key data is found in a PEM input. This allows
for better error messages to be return in the case of common errors like
switching the certifiate and key inputs to X509KeyPair.

Fixes #11092

Change-Id: Ifc155a811cdcddd93b5787fe16a84c972011f2f7
Reviewed-on: https://go-review.googlesource.com/14054
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agogo/types: move gotype command into this directory
Robert Griesemer [Tue, 29 Sep 2015 22:34:29 +0000 (15:34 -0700)]
go/types: move gotype command into this directory

This is a copy of x/tools/cmd/gotype/gotype.go with the corresponding
x/tools/cmd/gotype/doc.go prepended and including a build tag (ignore).

This way, go/types can be built unaffected. If we need the gotype command,
it is trivially built in the go/types directory with: go build gotype.go .

Fixes #12303.

Change-Id: I2d792fcb39719cc5cc300f657e4735901cd20faa
Reviewed-on: https://go-review.googlesource.com/15152
Reviewed-by: Rob Pike <r@golang.org>
9 years agogo/types: clarify doc string for types.Check
Robert Griesemer [Tue, 29 Sep 2015 20:17:39 +0000 (13:17 -0700)]
go/types: clarify doc string for types.Check

For #12787.

Change-Id: I921d01c8d7d97f3453b25e6d2241a43c5d64f53b
Reviewed-on: https://go-review.googlesource.com/15150
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agocmd/dist: build packages in parallel, make code more Go-like
Russ Cox [Fri, 28 Aug 2015 21:08:51 +0000 (17:08 -0400)]
cmd/dist: build packages in parallel, make code more Go-like

(Changed modified by bradfitz from original rsc version)

Change-Id: I8ea40044c325f333a13d48b59b4795b02c579533
Reviewed-on: https://go-review.googlesource.com/14026
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/compile/internal/gc: remove stringsCompare
Håvard Haugen [Wed, 23 Sep 2015 21:31:17 +0000 (23:31 +0200)]
cmd/compile/internal/gc: remove stringsCompare

Inlined the last occurrence of stringsCompare into exprcmp.

Passes go build -a -toolexec 'toolstash -cmp' std cmd.

Change-Id: I8fd99e3fbffc84283cc269368595cba950533066
Reviewed-on: https://go-review.googlesource.com/14872
Reviewed-by: Dave Cheney <dave@cheney.net>
Run-TryBot: Dave Cheney <dave@cheney.net>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/doc: fix pretty printing of paths
Rob Pike [Mon, 28 Sep 2015 20:45:57 +0000 (13:45 -0700)]
cmd/doc: fix pretty printing of paths

The code to strip GOROOT and GOPATH had a bug: it assumed there
were bytes after the GOROOT prefix but there might not be.
Fix this and other issues by taking care the prefix is really a
file name prefix for the path, not just a string prefix, and
handle the case where GOROOT==path.

Change-Id: I8066865fd05f938bb6dbf3bb8ab1fc58e5cf6bb5
Reviewed-on: https://go-review.googlesource.com/15112
Run-TryBot: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agomath/big: more documentation
Robert Griesemer [Mon, 28 Sep 2015 20:39:04 +0000 (13:39 -0700)]
math/big: more documentation

Good enough for now.

Fixes #11241.

Change-Id: Ieb50809f104d20bcbe14daecac503f72486bec92
Reviewed-on: https://go-review.googlesource.com/15111
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agomath/big: clean up *Int encoding tests
Robert Griesemer [Mon, 28 Sep 2015 18:31:53 +0000 (11:31 -0700)]
math/big: clean up *Int encoding tests

- more uniform naming
- test sign more deliberately
- remove superfluous test (JSON encoder always uses the JSON marshaler if present)

Change-Id: I37b1e367c01fc8bae1e06adbdb72dd366c08d5ce
Reviewed-on: https://go-review.googlesource.com/15110
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agomath/big: implement Float.Text(Un)Marshaler
Robert Griesemer [Fri, 25 Sep 2015 23:35:54 +0000 (16:35 -0700)]
math/big: implement Float.Text(Un)Marshaler

Fixes #12256.

Change-Id: Ie4a3337996da5c060b27530b076048ffead85f3b
Reviewed-on: https://go-review.googlesource.com/15040
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agotext/template, html/template: fix block example name
Andrew Gerrand [Tue, 29 Sep 2015 00:16:09 +0000 (10:16 +1000)]
text/template, html/template: fix block example name

Change-Id: I004a43842430201296363a9745480bee94920041
Reviewed-on: https://go-review.googlesource.com/15084
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agodoc: add text/template blocks and redefinition to go1.6.txt
Andrew Gerrand [Mon, 28 Sep 2015 22:43:07 +0000 (08:43 +1000)]
doc: add text/template blocks and redefinition to go1.6.txt

Change-Id: Ide82ac98dc7cb1035ceb9d461ed95af899f8f983
Reviewed-on: https://go-review.googlesource.com/15081
Reviewed-by: Andrew Gerrand <adg@golang.org>
9 years agoencoding/json: document that encoding.TextMarshaler is used if no (json) Marshaler...
Robert Griesemer [Mon, 28 Sep 2015 17:47:39 +0000 (10:47 -0700)]
encoding/json: document that encoding.TextMarshaler is used if no (json) Marshaler is present

Change-Id: I63da54832548c325e47dc54aaa5b5112e1f3b3ba
Reviewed-on: https://go-review.googlesource.com/15048
Reviewed-by: Rob Pike <r@golang.org>