]> Cypherpunks repositories - gostls13.git/log
gostls13.git
9 years agoruntime: accept NumGC==0 in TestMemStats
Austin Clements [Mon, 14 Dec 2015 23:04:40 +0000 (18:04 -0500)]
runtime: accept NumGC==0 in TestMemStats

TestMemStats currently requires that NumGC != 0, but GC may
legitimately not have run (for example, if this test runs first, or
GOGC is set high, etc). Accept NumGC == 0 and instead sanity check
NumGC by making sure that all pause times after NumGC are 0.

Fixes #11989.

Change-Id: I4203859fbb83292d59a509f2eeb24d6033e7aabc
Reviewed-on: https://go-review.googlesource.com/17830
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Mikio Hara <mikioh.mikioh@gmail.com>
9 years agocmd/compile: re-vendor math/big so we use latest version in compiler
Robert Griesemer [Wed, 16 Dec 2015 18:11:26 +0000 (10:11 -0800)]
cmd/compile: re-vendor math/big so we use latest version in compiler

This simply copies the current version of math/big into the
compiler directory. The change was created automatically by
running cmd/compile/internal/big/vendor.bash. No other manual
changes.

Change-Id: Ica225d196b3ac10dfd9d4dc1e4e4ef0b22812ff9
Reviewed-on: https://go-review.googlesource.com/17900
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agonet/http: fix Transport race returning bodyless responses and reusing conns
Brad Fitzpatrick [Wed, 16 Dec 2015 16:49:16 +0000 (16:49 +0000)]
net/http: fix Transport race returning bodyless responses and reusing conns

The Transport had a delicate protocol between its readLoop goroutine
and the goroutine calling RoundTrip. The basic concern is that the
caller's RoundTrip goroutine wants to wait for either a
connection-level error (the conn dying) or the response. But sometimes
both happen: there's a valid response (without a body), but the conn
is also going away. Both goroutines' logic dealing with this had grown
large and complicated with hard-to-follow comments over the years.

Simplify and document. Pull some bits into functions and do all
bodyless stuff in one place (it's special enough), rather than having
a bunch of conditionals scattered everywhere. One test is no longer
even applicable since the race it tested is no longer possible (the
code doesn't exist).

The bug that this fixes is that when the Transport reads a bodyless
response from a server, it was returning that response before
returning the persistent connection to the idle pool. As a result,
~1/1000 of serial requests would end up creating a new connection
rather than re-using the just-used connection due to goroutine
scheduling chance. Instead, this now adds bodyless responses'
connections back to the idle pool first, then sends the response to
the RoundTrip goroutine, but making sure that the RoundTrip goroutine
is outside of its select on the connection dying.

There's a new buffered channel involved now, which is a minor
complication, but it's much more self-contained and well-documented
than the previous complexity. (The alternative of making the
responseAndError channel itself unbuffered is too invasive and risky
at this point; it would require a number of changes to avoid
deadlocked goroutines in error cases)

In any case, flakes look to be gone now. We'll see if trybots agree.

Fixes #13633

Change-Id: I95a22942b2aa334ae7c87331fddd751d4cdfdffc
Reviewed-on: https://go-review.googlesource.com/17890
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

9 years agocmd/compile: fix -race nit
Russ Cox [Mon, 14 Dec 2015 18:24:27 +0000 (13:24 -0500)]
cmd/compile: fix -race nit

Fixes #13264.

Change-Id: I74b941164610921a03814733fea08631f18b6178
Reviewed-on: https://go-review.googlesource.com/17815
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agocmd/dist, runtime: make runtime version available as runtime.buildVersion
Shenghou Ma [Sun, 6 Dec 2015 22:35:12 +0000 (17:35 -0500)]
cmd/dist, runtime: make runtime version available as runtime.buildVersion

So that there is a uniformed way to retrieve Go version from a Go
binary, starting from Go 1.4 (see https://golang.org/cl/117040043)

Updates #13507.

Change-Id: Iaa2b14fca2d8c4d883d3824e2efc82b3e6fe2624
Reviewed-on: https://go-review.googlesource.com/17459
Reviewed-by: Keith Randall <khr@golang.org>
9 years agoruntime: change SIGEMT on linux/mips64 to throw
Ian Lance Taylor [Tue, 15 Dec 2015 23:46:48 +0000 (15:46 -0800)]
runtime: change SIGEMT on linux/mips64 to throw

This matches SIGEMT on other systems that use it (SIGEMT is not used
for most linux systems).

Change-Id: If394c06c9ed1cb3ea2564385a8edfbed8b5566d1
Reviewed-on: https://go-review.googlesource.com/17874
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Minux Ma <minux@golang.org>
9 years agonet: retighten test harnesses for dial cancelation
Mikio Hara [Wed, 16 Dec 2015 00:35:23 +0000 (09:35 +0900)]
net: retighten test harnesses for dial cancelation

Updates #11225.

Change-Id: I6c33d577f144643781f370ba2ab0997d1c1a3820
Reviewed-on: https://go-review.googlesource.com/17880
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agogo/constant: use Float.Rat method instead of doing it manually
Robert Griesemer [Tue, 15 Dec 2015 22:17:56 +0000 (14:17 -0800)]
go/constant: use Float.Rat method instead of doing it manually

Also fixed conversion bug and added corresponding test case.

Change-Id: I26f143fbc8d40a6d073ecb095e61b461495f3d68
Reviewed-on: https://go-review.googlesource.com/17872
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agospec: be clearer about which parameter section can be variadic
Robert Griesemer [Tue, 15 Dec 2015 21:13:38 +0000 (13:13 -0800)]
spec: be clearer about which parameter section can be variadic

Fixes #13595.

Change-Id: I870ddc97ea25b7f6f7a1bb1a78e5e4874fba1ddc
Reviewed-on: https://go-review.googlesource.com/17871
Reviewed-by: Rob Pike <r@golang.org>
9 years agonet: add Dialer.Cancel to cancel pending dials
Brad Fitzpatrick [Mon, 14 Dec 2015 22:21:48 +0000 (22:21 +0000)]
net: add Dialer.Cancel to cancel pending dials

Dialer.Cancel is a new optional <-chan struct{} channel whose closure
indicates that the dial should be canceled. It is compatible with the
x/net/context and http.Request.Cancel types.

Tested by hand with:

package main

    import (
            "log"
            "net"
            "time"
    )

    func main() {
            log.Printf("start.")
            var d net.Dialer
            cancel := make(chan struct{})
            time.AfterFunc(2*time.Second, func() {
                    log.Printf("timeout firing")
                    close(cancel)
            })
            d.Cancel = cancel
            c, err := d.Dial("tcp", "192.168.0.1:22")
            if err != nil {
                    log.Print(err)
                    return
            }
            log.Fatalf("unexpected connect: %v", c)
    }

Which says:

    2015/12/14 22:24:58 start.
    2015/12/14 22:25:00 timeout firing
    2015/12/14 22:25:00 dial tcp 192.168.0.1:22: operation was canceled

Fixes #11225

Change-Id: I2ef39e3a540e29fe6bfec03ab7a629a6b187fcb3
Reviewed-on: https://go-review.googlesource.com/17821
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agonet/http: maybe deflake TestCancelRequestMidBody_h2 on linux-noopt builder
Brad Fitzpatrick [Tue, 15 Dec 2015 17:20:05 +0000 (17:20 +0000)]
net/http: maybe deflake TestCancelRequestMidBody_h2 on linux-noopt builder

This might deflake it. Or it'll at least give us more debugging clues.

Fixes #13626 maybe

Change-Id: Ie8cd0375d60dad033ec6a64830a90e7b9152a3d9
Reviewed-on: https://go-review.googlesource.com/17825
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

9 years agonet/http: rework CloseNotifier implementation, clarify expectations in docs
Brad Fitzpatrick [Tue, 1 Dec 2015 19:07:41 +0000 (19:07 +0000)]
net/http: rework CloseNotifier implementation, clarify expectations in docs

CloseNotifier wasn't well specified previously. This CL simplifies its
implementation, clarifies the public documentation on CloseNotifier,
clarifies internal documentation on conn, and fixes two CloseNotifier
bugs in the process.

The main change, though, is tightening the rules and expectations for using
CloseNotifier:

* the caller must consume the Request.Body first (old rule, unwritten)
* the received value is the "true" value (old rule, unwritten)
* no promises for channel sends after Handler returns (old rule, unwritten)
* a subsequent pipelined request fires the CloseNotifier (new behavior;
  previously it never fired and thus effectively deadlocked as in #13165)
* advise that it should only be used without HTTP/1.1 pipelining (use HTTP/2
  or non-idempotent browsers). Not that browsers actually use pipelining.

The main implementation change is that each Handler now gets its own
CloseNotifier channel value, rather than sharing one between the whole
conn. This means Handlers can't affect subsequent requests. This is
how HTTP/2's Server works too. The old docs never clarified a behavior
either way. The other side effect of each request getting its own
CloseNotifier channel is that one handler can't "poison" the
underlying conn preventing subsequent requests on the same connection
from using CloseNotifier (this is #9763).

In the old implementation, once any request on a connection used
ClosedNotifier, the conn's underlying bufio.Reader source was switched
from the TCPConn to the read side of the pipe being fed by a
never-ending copy. Since it was impossible to abort that never-ending
copy, we could never get back to a fresh state where it was possible
to return the underlying TCPConn to callers of Hijack. Now, instead of
a never-ending Copy, the background goroutine doing a Read from the
TCPConn (or *tls.Conn) only reads a single byte. That single byte
can be in the request body, a socket timeout error, io.EOF error, or
the first byte of the second body. In any case, the new *connReader
type stitches sync and async reads together like an io.MultiReader. To
clarify the flow of Read data and combat the complexity of too many
wrapper Reader types, the *connReader absorbs the io.LimitReader
previously used for bounding request header reads.  The
liveSwitchReader type is removed. (an unused switchWriter type is also
removed)

Many fields on *conn are also documented more fully.

Fixes #9763 (CloseNotify + Hijack together)
Fixes #13165 (deadlock with CloseNotify + pipelined requests)

Change-Id: I40abc0a1992d05b294d627d1838c33cbccb9dd65
Reviewed-on: https://go-review.googlesource.com/17750
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoruntime: only trigger forced GC if GC is not running
Austin Clements [Mon, 14 Dec 2015 22:25:28 +0000 (17:25 -0500)]
runtime: only trigger forced GC if GC is not running

Currently, sysmon triggers a forced GC solely based on
memstats.last_gc. However, memstats.last_gc isn't updated until mark
termination, so once sysmon starts triggering forced GC, it will keep
triggering them until GC finishes. The first of these actually starts
a GC; the remainder up to the last print "GC forced", but gcStart
returns immediately because gcphase != _GCoff; then the last may start
another GC if the previous GC finishes (and sets last_gc) between
sysmon triggering it and gcStart checking the GC phase.

Fix this by expanding the condition for starting a forced GC to also
require that no GC is currently running. This, combined with the way
forcegchelper blocks until the GC cycle is started, ensures sysmon
only starts one GC when the time exceeds the forced GC threshold.

Fixes #13458.

Change-Id: Ie6cf841927f6085136be3f45259956cd5cf10d23
Reviewed-on: https://go-review.googlesource.com/17819
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoruntime: simplify sigprof traceback interlocking
Austin Clements [Mon, 23 Nov 2015 20:13:16 +0000 (15:13 -0500)]
runtime: simplify sigprof traceback interlocking

The addition of stack barrier locking to copystack subsumes the
partial fix from commit bbd1a1c for SIGPROF during copystack. With the
stack barrier locking, this commit simplifies the rule in sigprof to:
the user stack can be traced only if sigprof can acquire the stack
barrier lock.

Updates #12932, #13362.

Change-Id: I1c1f80015053d0ac7761e9e0c7437c2aba26663f
Reviewed-on: https://go-review.googlesource.com/17192
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/compile: change dead code into assert
Matthew Dempsky [Mon, 14 Dec 2015 20:37:26 +0000 (12:37 -0800)]
cmd/compile: change dead code into assert

After fixing #13587, I noticed that the "OAS2FUNC in disguise" block
looked like it probably needed write barriers too.  However, testing
revealed the multi-value "return f()" case was already being handled
correctly.

It turns out this block is dead code due to "return f()" already being
transformed into "t1, t2, ..., tN := f(); return t1, t2, ..., tN" by
orderstmt when f is a multi-valued function.

Updates #13587.

Change-Id: Icde46dccc55beda2ea5fd5fcafc9aae26cec1552
Reviewed-on: https://go-review.googlesource.com/17759
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
9 years agoruntime: update triggerRatio in setGCPercent
Austin Clements [Mon, 14 Dec 2015 20:38:12 +0000 (15:38 -0500)]
runtime: update triggerRatio in setGCPercent

Currently, runtime/debug.SetGCPercent does not adjust the controller
trigger ratio. As a result, runtime reductions of GOGC don't take full
effect until after one more concurrent cycle has happened, which
adjusts the trigger ratio to account for the new gcpercent.

Fix this by lowering the trigger ratio if necessary in setGCPercent.

Change-Id: I4d23e0c58d91939b86ac60fa5d53ef91d0d89e0c
Reviewed-on: https://go-review.googlesource.com/17813
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoruntime: print gctrace before releasing worldsema
Austin Clements [Mon, 14 Dec 2015 20:19:07 +0000 (15:19 -0500)]
runtime: print gctrace before releasing worldsema

Currently we drop worldsema and then print the gctrace. We did this so
that if stderr is a pipe or a blocked terminal, blocking on printing
the gctrace would not block another GC from starting. However, this is
a bit of a fool's errand because a blocked runtime print will block
the whole M/P, so after GOMAXPROCS GC cycles, the whole system will
freeze. Furthermore, now this is much less of an issue because
allocation will block indefinitely if it can't start a GC (whereas it
used to be that allocation could run away). Finally, this allows
another GC cycle to start while the previous cycle is printing the
gctrace, which leads to races on reading various statistics to print
them and the next GC cycle overwriting those statistics.

Fix this by moving the release of worldsema after the gctrace print.

Change-Id: I3d044ea0f77d80f3b4050af6b771e7912258662a
Reviewed-on: https://go-review.googlesource.com/17812
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Rick Hudson <rlh@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoruntime: reset sweep stats before starting the world
Austin Clements [Mon, 14 Dec 2015 20:07:40 +0000 (15:07 -0500)]
runtime: reset sweep stats before starting the world

Currently we reset the sweep stats just after gcMarkTermination starts
the world and releases worldsema. However, background sweeping can
start the moment we start the world and, in fact, pause sweeping can
start the moment we release worldsema (because another GC cycle can
start up), so these need to be cleared before starting the world.

Change-Id: I95701e3de6af76bb3fbf2ee65719985bf57d20b2
Reviewed-on: https://go-review.googlesource.com/17811
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoruntime: fix (sometimes major) underestimation of heap_live
Austin Clements [Fri, 11 Dec 2015 22:49:14 +0000 (17:49 -0500)]
runtime: fix (sometimes major) underestimation of heap_live

Currently, we update memstats.heap_live from mcache.local_cachealloc
whenever we lock the heap (e.g., to obtain a fresh span or to release
an unused span). However, under the right circumstances,
local_cachealloc can accumulate allocations up to the size of
the *entire heap* without flushing them to heap_live. Specifically,
since span allocations from an mcentral don't lock the heap, if a
large number of pages are held in an mcentral and the application
continues to use and free objects of that size class (e.g., the
BinaryTree17 benchmark), local_cachealloc won't be flushed until the
mcentral runs out of spans.

This is a problem because, unlike many of the memory statistics that
are purely informative, heap_live is used to determine when the
garbage collector should start and how hard it should work.

This commit eliminates local_cachealloc, instead atomically updating
heap_live directly. To control contention, we do this only when
obtaining a span from an mcentral. Furthermore, we make heap_live
conservative: allocating a span assumes that all free slots in that
span will be used and accounts for these when the span is
allocated, *before* the objects themselves are. This is important
because 1) this triggers the GC earlier than necessary rather than
potentially too late and 2) this leads to a conservative GC rate
rather than a GC rate that is potentially too low.

Alternatively, we could have flushed local_cachealloc when it passed
some threshold, but this would require determining a threshold and
would cause heap_live to underestimate the true value rather than
overestimate.

Fixes #12199.

name                      old time/op    new time/op    delta
BinaryTree17-12              2.88s ± 4%     2.88s ± 1%    ~     (p=0.470 n=19+19)
Fannkuch11-12                2.48s ± 1%     2.48s ± 1%    ~     (p=0.243 n=16+19)
FmtFprintfEmpty-12          50.9ns ± 2%    50.7ns ± 1%    ~     (p=0.238 n=15+14)
FmtFprintfString-12          175ns ± 1%     171ns ± 1%  -2.48%  (p=0.000 n=18+18)
FmtFprintfInt-12             159ns ± 1%     158ns ± 1%  -0.78%  (p=0.000 n=19+18)
FmtFprintfIntInt-12          270ns ± 1%     265ns ± 2%  -1.67%  (p=0.000 n=18+18)
FmtFprintfPrefixedInt-12     235ns ± 1%     234ns ± 0%    ~     (p=0.362 n=18+19)
FmtFprintfFloat-12           309ns ± 1%     308ns ± 1%  -0.41%  (p=0.001 n=18+19)
FmtManyArgs-12              1.10µs ± 1%    1.08µs ± 0%  -1.96%  (p=0.000 n=19+18)
GobDecode-12                7.81ms ± 1%    7.80ms ± 1%    ~     (p=0.425 n=18+19)
GobEncode-12                6.53ms ± 1%    6.53ms ± 1%    ~     (p=0.817 n=19+19)
Gzip-12                      312ms ± 1%     312ms ± 2%    ~     (p=0.967 n=19+20)
Gunzip-12                   42.0ms ± 1%    41.9ms ± 1%    ~     (p=0.172 n=19+19)
HTTPClientServer-12         63.7µs ± 1%    63.8µs ± 1%    ~     (p=0.639 n=19+19)
JSONEncode-12               16.4ms ± 1%    16.4ms ± 1%    ~     (p=0.954 n=19+19)
JSONDecode-12               58.5ms ± 1%    57.8ms ± 1%  -1.27%  (p=0.000 n=18+19)
Mandelbrot200-12            3.86ms ± 1%    3.88ms ± 0%  +0.44%  (p=0.000 n=18+18)
GoParse-12                  3.67ms ± 2%    3.66ms ± 1%  -0.52%  (p=0.001 n=18+19)
RegexpMatchEasy0_32-12       100ns ± 1%     100ns ± 0%    ~     (p=0.257 n=19+18)
RegexpMatchEasy0_1K-12       347ns ± 1%     347ns ± 1%    ~     (p=0.527 n=18+18)
RegexpMatchEasy1_32-12      83.7ns ± 2%    83.1ns ± 2%    ~     (p=0.096 n=18+19)
RegexpMatchEasy1_1K-12       509ns ± 1%     505ns ± 1%  -0.75%  (p=0.000 n=18+19)
RegexpMatchMedium_32-12      130ns ± 2%     129ns ± 1%    ~     (p=0.962 n=20+20)
RegexpMatchMedium_1K-12     39.5µs ± 2%    39.4µs ± 1%    ~     (p=0.376 n=20+19)
RegexpMatchHard_32-12       2.04µs ± 0%    2.04µs ± 1%    ~     (p=0.195 n=18+17)
RegexpMatchHard_1K-12       61.4µs ± 1%    61.4µs ± 1%    ~     (p=0.885 n=19+19)
Revcomp-12                   540ms ± 2%     542ms ± 4%    ~     (p=0.552 n=19+17)
Template-12                 69.6ms ± 1%    71.2ms ± 1%  +2.39%  (p=0.000 n=20+20)
TimeParse-12                 357ns ± 1%     357ns ± 1%    ~     (p=0.883 n=18+20)
TimeFormat-12                379ns ± 1%     362ns ± 1%  -4.53%  (p=0.000 n=18+19)
[Geo mean]                  62.0µs         61.8µs       -0.44%

name              old time/op  new time/op  delta
XBenchGarbage-12  5.89ms ± 2%  5.81ms ± 2%  -1.41%  (p=0.000 n=19+18)

Change-Id: I96b31cca6ae77c30693a891cff3fe663fa2447a0
Reviewed-on: https://go-review.googlesource.com/17748
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoruntime: trace sweep completion in gcpacertrace mode
Austin Clements [Fri, 11 Dec 2015 22:50:22 +0000 (17:50 -0500)]
runtime: trace sweep completion in gcpacertrace mode

Change-Id: I7991612e4d064c15492a39c19f753df1db926203
Reviewed-on: https://go-review.googlesource.com/17747
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoruntime: check for spanBytesAlloc underflow
Austin Clements [Fri, 11 Dec 2015 22:50:02 +0000 (17:50 -0500)]
runtime: check for spanBytesAlloc underflow

Change-Id: I5e6739ff0c6c561195ed9891fb90f933b81e7750
Reviewed-on: https://go-review.googlesource.com/17746
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoruntime: deduct correct sweep credit
Austin Clements [Fri, 11 Dec 2015 17:33:34 +0000 (12:33 -0500)]
runtime: deduct correct sweep credit

deductSweepCredit expects the size in bytes of the span being
allocated, but mCentral_CacheSpan passes the size of a single object
in the span. As a result, we don't sweep enough on that call and when
mCentral_CacheSpan later calls reimburseSweepCredit, it's very likely
to underflow mheap_.spanBytesAlloc, which causes the next call to
deductSweepCredit to think it owes a huge number of pages and finish
off the whole sweep.

In addition to causing the occasional allocation that triggers the
full sweep to be potentially extremely expensive relative to other
allocations, this can indirectly slow down many other allocations.
deductSweepCredit uses sweepone to sweep spans, which returns
fully-unused spans to the heap, where these spans are freed and
coalesced with neighboring free spans. On the other hand, when
mCentral_CacheSpan sweeps a span, it does so with the intent to
immediately reuse that span and, as a result, will not return the span
to the heap even if it is fully unused. This saves on the cost of
locking the heap, finding a span, and initializing that span. For
example, before this change, with GOMAXPROCS=1 (or the background
sweeper disabled) BinaryTree17 returned roughly 220K spans to the heap
and allocated new spans from the heap roughly 232K times. After this
change, it returns 1.3K spans to the heap and allocates new spans from
the heap 39K times. (With background sweeping these numbers are
effectively unchanged because the background sweeper sweeps almost all
of the spans with sweepone; however, parallel sweeping saves more than
the cost of allocating spans from the heap.)

Fixes #13535.
Fixes #13589.

name                      old time/op    new time/op    delta
BinaryTree17-12              3.03s ± 1%     2.86s ± 4%  -5.61%  (p=0.000 n=18+20)
Fannkuch11-12                2.48s ± 1%     2.49s ± 1%    ~     (p=0.060 n=17+20)
FmtFprintfEmpty-12          50.7ns ± 1%    50.9ns ± 1%  +0.43%  (p=0.025 n=15+16)
FmtFprintfString-12          174ns ± 2%     174ns ± 2%    ~     (p=0.539 n=19+20)
FmtFprintfInt-12             158ns ± 1%     158ns ± 1%    ~     (p=0.300 n=18+20)
FmtFprintfIntInt-12          269ns ± 2%     269ns ± 2%    ~     (p=0.784 n=20+18)
FmtFprintfPrefixedInt-12     233ns ± 1%     234ns ± 1%    ~     (p=0.389 n=18+18)
FmtFprintfFloat-12           309ns ± 1%     310ns ± 1%  +0.25%  (p=0.048 n=18+18)
FmtManyArgs-12              1.10µs ± 1%    1.10µs ± 1%    ~     (p=0.259 n=18+19)
GobDecode-12                7.81ms ± 1%    7.72ms ± 1%  -1.17%  (p=0.000 n=19+19)
GobEncode-12                6.56ms ± 0%    6.55ms ± 1%    ~     (p=0.433 n=17+19)
Gzip-12                      318ms ± 2%     317ms ± 1%    ~     (p=0.578 n=19+18)
Gunzip-12                   42.1ms ± 2%    42.0ms ± 0%  -0.45%  (p=0.007 n=18+16)
HTTPClientServer-12         63.9µs ± 1%    64.0µs ± 1%    ~     (p=0.146 n=17+19)
JSONEncode-12               16.4ms ± 1%    16.4ms ± 1%    ~     (p=0.271 n=19+19)
JSONDecode-12               58.1ms ± 1%    58.0ms ± 1%    ~     (p=0.152 n=18+18)
Mandelbrot200-12            3.85ms ± 0%    3.85ms ± 0%    ~     (p=0.126 n=19+18)
GoParse-12                  3.71ms ± 1%    3.64ms ± 1%  -1.86%  (p=0.000 n=20+18)
RegexpMatchEasy0_32-12       100ns ± 2%     100ns ± 1%    ~     (p=0.588 n=20+20)
RegexpMatchEasy0_1K-12       346ns ± 1%     347ns ± 1%  +0.27%  (p=0.014 n=17+20)
RegexpMatchEasy1_32-12      82.9ns ± 3%    83.5ns ± 3%    ~     (p=0.096 n=19+20)
RegexpMatchEasy1_1K-12       506ns ± 1%     506ns ± 1%    ~     (p=0.530 n=19+19)
RegexpMatchMedium_32-12      129ns ± 2%     129ns ± 1%    ~     (p=0.566 n=20+19)
RegexpMatchMedium_1K-12     39.4µs ± 1%    39.4µs ± 1%    ~     (p=0.713 n=19+20)
RegexpMatchHard_32-12       2.05µs ± 1%    2.06µs ± 1%  +0.36%  (p=0.008 n=18+20)
RegexpMatchHard_1K-12       61.6µs ± 1%    61.7µs ± 1%    ~     (p=0.286 n=19+20)
Revcomp-12                   538ms ± 1%     541ms ± 2%    ~     (p=0.081 n=18+19)
Template-12                 71.5ms ± 2%    71.6ms ± 1%    ~     (p=0.513 n=20+19)
TimeParse-12                 357ns ± 1%     357ns ± 1%    ~     (p=0.935 n=19+18)
TimeFormat-12                352ns ± 1%     352ns ± 1%    ~     (p=0.293 n=19+20)
[Geo mean]                  62.0µs         61.9µs       -0.21%

name              old time/op  new time/op  delta
XBenchGarbage-12  5.83ms ± 2%  5.86ms ± 3%    ~     (p=0.247 n=19+20)

Change-Id: I790bb530adace27ccf25d372f24a11954b88443c
Reviewed-on: https://go-review.googlesource.com/17745
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoruntime: init argc/argv for android/arm64 c-shared
Péter Szilágyi [Sun, 13 Dec 2015 19:04:06 +0000 (21:04 +0200)]
runtime: init argc/argv for android/arm64 c-shared

Analogous to https://go-review.googlesource.com/#/c/8457/ this
code synthesizes an set of program arguments for Android on the
arm64 architecture.

Change-Id: I851958b4b0944ec79d7a1426a3bb2cfc31746797
Reviewed-on: https://go-review.googlesource.com/17782
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Run-TryBot: David Crawshaw <crawshaw@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agopath/filepath: include test number in TestIssue13582 output
Alex Brainman [Tue, 15 Dec 2015 05:56:18 +0000 (16:56 +1100)]
path/filepath: include test number in TestIssue13582 output

Otherwise it's hard to tell the difference between
link1 and link2 or other tests.

Change-Id: I36c153cccb10959535595938dfbc49db930b9fac
Reviewed-on: https://go-review.googlesource.com/17851
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agonet/http: updated bundled http2 copy, enable some tests
Brad Fitzpatrick [Tue, 15 Dec 2015 00:52:58 +0000 (00:52 +0000)]
net/http: updated bundled http2 copy, enable some tests

Updates bundled copy of x/net/http2 to include
https://golang.org/cl/17823 (catching panics in Handlers)

Fixes #13555

Change-Id: I08e4e38e736a8d93f5ec200e8041c143fc6eafce
Reviewed-on: https://go-review.googlesource.com/17824
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agonet/http/httputil: make DumpRequest use Request.RequestURI when available
Brad Fitzpatrick [Tue, 20 Oct 2015 22:52:11 +0000 (22:52 +0000)]
net/http/httputil: make DumpRequest use Request.RequestURI when available

Fixes #10912

Change-Id: If04e3205d5cc43ebfd6864bc59340c8697cbc0af
Reviewed-on: https://go-review.googlesource.com/17592
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agonet/http: skip TestClientTimeout_Headers in HTTP/2 mode
Burcu Dogan [Tue, 15 Dec 2015 00:39:09 +0000 (16:39 -0800)]
net/http: skip TestClientTimeout_Headers in HTTP/2 mode

Change-Id: I3533b557cd6c7127ab4efbe8766184b51ce260c9
Reviewed-on: https://go-review.googlesource.com/17768
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

9 years agonet/http: test client timeout against HTTP/2
Burcu Dogan [Tue, 15 Dec 2015 00:17:21 +0000 (16:17 -0800)]
net/http: test client timeout against HTTP/2

Change-Id: Id511855da1c663250a4ffb149277a3f4a7f38360
Reviewed-on: https://go-review.googlesource.com/17766
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agogo/constant: switch to floating-point representation when fractions become too large
Robert Griesemer [Thu, 3 Dec 2015 01:41:39 +0000 (17:41 -0800)]
go/constant: switch to floating-point representation when fractions become too large

Use two internal representations for Float values (similar to what is done
for Int values). Transparently switch to a big.Float representation when
big.Rat values become unwieldy. This is almost never needed for real-world
programs but it is trivial to create test cases that cannot be handled with
rational arithmetic alone.

As a consequence, the go/constant API semantics changes slightly: Until now,
a value could always be represented in its "smallest" form (e.g., float values
that happened to be integers would be represented as integers). Now, constant
Kind depends on how the value was created, rather than its actual value. (The
reason why we cannot automatically "normalize" values to their smallest form
anymore is because floating-point numbers are not exact in general; and thus
normalization is often not possible in the first place, or would throw away
precision when it is not desired.) This has repercussions as to how constant
Values are used go/types and required corresponding adjustments.

Details of the changes:

go/constant package:
- use big.Rat and big.Float values to represent floating-point values
  (internal change)
- changed semantic of Value.Kind accordingly
- String now returns a short, human-readable form of a value
  (this leads to better error messages in go/types)
- added ToInt, ToFloat, and ToComplex conversion functions
- added ExactString to obtain an exact string form of a value

go/types:
- adjusted and simplified implementation of representableConst
- adjusted various places where Value.Kind was expected to be "smallest"
  by calling the respective ToInt/Float/Complex conversion functions
- enabled 5 disabled tests in stdlib_test.go that now work

api checker:
- print all constant values in a short human-readable form (floats are
  printed in floating-point form), but also print an exact form if it
  is different from the short form
- adjusted test golden file and go.1.1.text reference file

Fixes #11327.

Change-Id: I492b704aae5b0238e5b7cee13e18ffce61193587
Reviewed-on: https://go-review.googlesource.com/17360
Reviewed-by: Alan Donovan <adonovan@google.com>
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agonet/http: add Transport tests for using Request.Cancel mid-body
Brad Fitzpatrick [Mon, 14 Dec 2015 18:45:34 +0000 (18:45 +0000)]
net/http: add Transport tests for using Request.Cancel mid-body

This CL also updates the bundled http2 package with the h2 fix from
https://golang.org/cl/17757

Fixes #13159

Change-Id: If0e3b4bd04d0dceed67d1b416ed838c9f1961576
Reviewed-on: https://go-review.googlesource.com/17758
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoos: make TestLchown actually test Lchown.
Rahul Chaudhry [Thu, 3 Dec 2015 21:19:48 +0000 (13:19 -0800)]
os: make TestLchown actually test Lchown.

TestLchown was creating a hard-link instead of a symlink. It would
have passed if you replaced all Lchown() calls in it with Chown().

Change-Id: I3a108948ec25fcbac8ea890a6eaf5bac094f0800
Reviewed-on: https://go-review.googlesource.com/17397
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/cgo: use _Complex in rewritten Go code too
Matthew Dempsky [Mon, 14 Dec 2015 21:05:25 +0000 (13:05 -0800)]
cmd/cgo: use _Complex in rewritten Go code too

Followup to CL 17716, which updated cgo's boilerplate prologue code to
use standard C's _Complex instead of GCC's __complex extension.

Change-Id: I74f29b0cc3d13cab2853441cafbfe77853bba4f9
Reviewed-on: https://go-review.googlesource.com/17820
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agoruntime: recycle large stack spans
Austin Clements [Mon, 14 Dec 2015 19:30:25 +0000 (14:30 -0500)]
runtime: recycle large stack spans

To prevent races with the garbage collector, stack spans cannot be
reused as heap spans during a GC. We deal with this by caching stack
spans during GC and releasing them at the end of mark termination.
However, while our cache lets us reuse small stack spans, currently
large stack spans are *not* reused. This can cause significant memory
growth in programs that allocate large stacks rapidly, but grow the
heap slowly (such as in issue #13552).

Fix this by adding logic to reuse large stack spans for other stacks.

Fixes #11466.

Fixes #13552. Without this change, the program in this issue creeps to
over 1GB of memory over the course of a few hours. With this change,
it stays rock solid at around 30MB.

Change-Id: If8b2d85464aa80c96230a1990715e39aa803904f
Reviewed-on: https://go-review.googlesource.com/17814
Reviewed-by: Keith Randall <khr@golang.org>
9 years agocmd/internal/obj: remove 3 incorrect copyright notices
Russ Cox [Mon, 14 Dec 2015 17:12:57 +0000 (12:12 -0500)]
cmd/internal/obj: remove 3 incorrect copyright notices

These three files contain only code written for Go
(and trivial amounts at that), not any code ported
from Inferno or Plan 9.

Remove the incorrect Inferno/Plan 9 notices.

Fixes #13576.

Change-Id: Ib9901fb360232282aae5ee0f4aa527bd6f4eaaed
Reviewed-on: https://go-review.googlesource.com/17779
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agonet: make LookupPort with empty service mean 0
Brad Fitzpatrick [Mon, 14 Dec 2015 18:22:23 +0000 (18:22 +0000)]
net: make LookupPort with empty service mean 0

Fixes #13610

Change-Id: I9c8f924dc1ad515a9697291e981ece34fdbec8b7
Reviewed-on: https://go-review.googlesource.com/17755
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agonet: add test for CL 17458
Russ Cox [Fri, 11 Dec 2015 04:51:36 +0000 (23:51 -0500)]
net: add test for CL 17458

I thought that we avoided creating on-disk Unix sockets,
but I was mistaken. Use one to test CL 17458.

Fixes #11826.

Change-Id: Iaa1fb007b95fa6be48200586522a6d4789ecd346
Reviewed-on: https://go-review.googlesource.com/17725
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocrypto/tls: document lack of Lucky13 hardening
Brad Fitzpatrick [Tue, 8 Dec 2015 16:49:17 +0000 (16:49 +0000)]
crypto/tls: document lack of Lucky13 hardening

Updates #13385

Change-Id: I9c2edf8c02adc388c48760b29e63dfa2966262d6
Reviewed-on: https://go-review.googlesource.com/17532
Reviewed-by: Tim Cooijmans <timcooijmans@gmail.com>
Reviewed-by: Adam Langley <agl@golang.org>
9 years agonet/http: update bundled http2 copy from x/net/http2
Brad Fitzpatrick [Mon, 14 Dec 2015 18:30:24 +0000 (18:30 +0000)]
net/http: update bundled http2 copy from x/net/http2

Updates to x/net git rev 6c105c0a

Fixes #13598

Change-Id: I207d4c78d744f0fd83cb5acd8bd6e5987e59a4f7
Reviewed-on: https://go-review.googlesource.com/17756
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/cgo: use standard C syntax for complex types
Ian Lance Taylor [Fri, 11 Dec 2015 03:38:14 +0000 (19:38 -0800)]
cmd/cgo: use standard C syntax for complex types

(instead of using a GCC extension).

Change-Id: I110dc45bfe5f1377fe3453070eccde283b5cc161
Reviewed-on: https://go-review.googlesource.com/17716
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agopath/filepath: keep walking if EvalSymlinks returns symlink
Alex Brainman [Sun, 13 Dec 2015 23:33:56 +0000 (10:33 +1100)]
path/filepath: keep walking if EvalSymlinks returns symlink

Fixes #13582

Change-Id: I220f3c7b9511b3c080874f5c42f2a431fdddcbb7
Reviewed-on: https://go-review.googlesource.com/17794
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agodoc: fix typo "heirarchy"
Russ Cox [Fri, 11 Dec 2015 16:35:36 +0000 (11:35 -0500)]
doc: fix typo "heirarchy"

Change-Id: Iae2bf44ec15975f440d026fd52fcccfbd9c598d9
Reviewed-on: https://go-review.googlesource.com/17740
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agonet/http: fix race in TimeoutHandler
Brad Fitzpatrick [Mon, 14 Dec 2015 01:04:07 +0000 (01:04 +0000)]
net/http: fix race in TimeoutHandler

New implementation of TimeoutHandler: buffer everything to memory.

All or nothing: either the handler finishes completely within the
timeout (in which case the wrapper writes it all), or it misses the
timeout and none of it gets written, in which case handler wrapper can
reliably print the error response without fear that some of the
wrapped Handler's code already wrote to the output.

Now the goroutine running the wrapped Handler has its own write buffer
and Header copy.

Document the limitations.

Fixes #9162

Change-Id: Ia058c1d62cefd11843e7a2fc1ae1609d75de2441
Reviewed-on: https://go-review.googlesource.com/17752
Reviewed-by: David Symonds <dsymonds@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

9 years agonet: fix typo
Mikio Hara [Mon, 14 Dec 2015 03:16:13 +0000 (12:16 +0900)]
net: fix typo

Change-Id: I064a7e21bdc55d7f72147bcf6cfb91b96dbb77bb
Reviewed-on: https://go-review.googlesource.com/17795
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
9 years agonet: add missing File{Conn,Listener,PacketConn} read/write tests
Mikio Hara [Mon, 7 Dec 2015 08:45:19 +0000 (17:45 +0900)]
net: add missing File{Conn,Listener,PacketConn} read/write tests

At present, the series of File{Conn,Listener,PacketConn} APIs are the
only way to configure platform-specific socket options such as
SO_REUSE{ADDR,PORT}, TCP_FASTOPEN. This change adds missing test cases
that test read and write operations on connections created by File APIs
and removes redundant parameter tests which are already tested in
server_test.go.

Also adds comment on full stack test cases for IPConn.

Fixes #10730.

Change-Id: I67abb083781b602e876f72a6775a593c0f363c38
Reviewed-on: https://go-review.googlesource.com/17476
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agonet/http: annotate equivalent h2 tests for h1
Emmanuel Odeke [Wed, 9 Dec 2015 02:29:22 +0000 (19:29 -0700)]
net/http: annotate equivalent h2 tests for h1

Change-Id: Ic639443d4a8536dd54c5b0cf5a9b6d1474448f77
Reviewed-on: https://go-review.googlesource.com/17593
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/compile: add missing write barriers for return statements
Matthew Dempsky [Sat, 12 Dec 2015 03:11:54 +0000 (19:11 -0800)]
cmd/compile: add missing write barriers for return statements

Copying return values to heap-escaped result parameters requires write
barriers.

Fixes #13587.

Change-Id: Ifa04ff7fa4adcc6393acdd82e527beb8f2a00a8b
Reviewed-on: https://go-review.googlesource.com/17762
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoCONTRIBUTORS: add Arkadi Pyuro <arkadi@google.com>
Sameer Ajmani [Thu, 25 Jun 2015 17:11:25 +0000 (13:11 -0400)]
CONTRIBUTORS: add Arkadi Pyuro <arkadi@google.com>

Change-Id: I88fea011e68a8f0d351d2a9dd81d6aa48bc7c7c7
Reviewed-on: https://go-review.googlesource.com/17731
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agomath/big: fix carry propagation in Int.Exp Montgomery code
Russ Cox [Wed, 9 Dec 2015 16:49:53 +0000 (11:49 -0500)]
math/big: fix carry propagation in Int.Exp Montgomery code

Fixes #13515.

Change-Id: I7dd5fbc816e5ea135f7d81f6735e7601f636fe4f
Reviewed-on: https://go-review.googlesource.com/17672
Reviewed-by: Robert Griesemer <gri@golang.org>
9 years agodoc: many updates to go1.6.txt
Russ Cox [Fri, 11 Dec 2015 16:52:57 +0000 (11:52 -0500)]
doc: many updates to go1.6.txt

Change-Id: I97d2315a1f978fbc4fd9e7f5f860f8e29ae43470
Reviewed-on: https://go-review.googlesource.com/17743
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agomath/rand: improve uniformity of rand.Float64,Float32
David Chase [Wed, 9 Dec 2015 16:45:36 +0000 (11:45 -0500)]
math/rand: improve uniformity of rand.Float64,Float32

Replaced code that substituted 0 for rounded-up 1 with
code to try again.  This has minimal effect on the existing
stream of random numbers, but restores uniformity.

Fixes #12290.

Change-Id: Ib68f0b0a4a173339bcd0274cc16509f7b0977de8
Reviewed-on: https://go-review.googlesource.com/17670
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agomisc/cgo/gmp: change pi.go to use standard Go copyright header
Ian Lance Taylor [Fri, 11 Dec 2015 03:24:06 +0000 (19:24 -0800)]
misc/cgo/gmp: change pi.go to use standard Go copyright header

This file is not part of the benchmark shootout, and we wrote it, so use
the usual copyright header, not a partial version of the shootout
license.

Fixes #13575.

Change-Id: Ib610e2ad82914b4ef096a2424cfffe3383db2d5b
Reviewed-on: https://go-review.googlesource.com/17715
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoruntime: remove unnecessary wakeups of worker threads
Dmitry Vyukov [Tue, 8 Dec 2015 14:11:27 +0000 (15:11 +0100)]
runtime: remove unnecessary wakeups of worker threads

Currently we wake up new worker threads whenever we pass
through the scheduler with nmspinning==0. This leads to
lots of unnecessary thread wake ups.
Instead let only spinning threads wake up new spinning threads.

For the following program:

package main
import "runtime"
func main() {
for i := 0; i < 1e7; i++ {
runtime.Gosched()
}
}

Before:
$ time ./test
real 0m4.278s
user 0m7.634s
sys 0m1.423s

$ strace -c ./test
% time     seconds  usecs/call     calls    errors syscall
 99.93    9.314936           3   2685009     17536 futex

After:
$ time ./test
real 0m1.200s
user 0m1.181s
sys 0m0.024s

$ strace -c ./test
% time     seconds  usecs/call     calls    errors syscall
  3.11    0.000049          25         2           futex

Fixes #13527

Change-Id: Ia1f5bf8a896dcc25d8b04beb1f4317aa9ff16f74
Reviewed-on: https://go-review.googlesource.com/17540
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Dmitry Vyukov <dvyukov@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agodoc/go1.6.txt: remove fmt.Scanf change (reverted)
Russ Cox [Fri, 11 Dec 2015 06:34:16 +0000 (01:34 -0500)]
doc/go1.6.txt: remove fmt.Scanf change (reverted)

Change-Id: I26fc120e66f559e4da90883ed5c8457a99426a25
Reviewed-on: https://go-review.googlesource.com/17729
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoRevert "fmt: check newline in the end of input"
Russ Cox [Fri, 11 Dec 2015 03:28:15 +0000 (03:28 +0000)]
Revert "fmt: check newline in the end of input"

This change can break real code. There are other newline-related bugs in this code, and fixing them will also break real code. If we're going to break real code, let's fix all the bugs together and just break things once.

This reverts commit 8331f19d9700457d74bf377755ca270a32d2faa3.

Change-Id: Ie4b3022f3a305c3e1f78cc208e50beed212608e6
Reviewed-on: https://go-review.googlesource.com/17724
Reviewed-by: Rob Pike <r@golang.org>
9 years agoreflect: document DeepEqual, fix case of maps with NaN keys
Russ Cox [Wed, 25 Nov 2015 15:45:59 +0000 (10:45 -0500)]
reflect: document DeepEqual, fix case of maps with NaN keys

The meaning of DeepEqual has never been specified.
Do that.

Also fix bug involving maps with NaN keys.

Except for the map bug fix, there should be no semantic changes here.

Fixes #12025.

Change-Id: Ied562cf543a22ec645d42bdb9b41d451c16b1f21
Reviewed-on: https://go-review.googlesource.com/17450
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>

9 years agonet/url: reject space in host; do not escape < > " in host
Russ Cox [Fri, 4 Dec 2015 17:43:25 +0000 (12:43 -0500)]
net/url: reject space in host; do not escape < > " in host

Host names in URLs must not use %-escaping for ASCII bytes, per RFC 3986.

url.Parse has historically allowed spaces and < > " in the URL host.
In Go 1.5, URL's String method started escaping those,
but then Parse would rejects the escaped form.
This CL is an attempt at some consistency between Parse and String
as far as the accepted host characters and the encoding of host characters,
so that if Parse succeeds, then Parse -> String -> Parse also succeeds.

Allowing space seems like a mistake, so reject that in Parse.
(Similarly, reject \t, \x01, and so on, all of which were being allowed.)

Allowing < > " doesn't seem awful, so continue to do that,
and go back to the Go 1.4 behavior of not escaping them in String.

Fixes #11302.

Change-Id: I0bf65b874cd936598f20694574364352a5abbe5f
Reviewed-on: https://go-review.googlesource.com/17387
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agonet: do not unlink unix socket in UnixListener created from fd
Russ Cox [Sat, 5 Dec 2015 06:15:26 +0000 (01:15 -0500)]
net: do not unlink unix socket in UnixListener created from fd

Fixes #11826.

Change-Id: Id220dd558ca8d8d78c01975087122d27757deea0
Reviewed-on: https://go-review.googlesource.com/17458
Reviewed-by: Ian Lance Taylor <iant@golang.org>
9 years agoruntime: fix GODEBUG=schedtrace=X delay handling.
Rahul Chaudhry [Thu, 10 Dec 2015 23:06:42 +0000 (15:06 -0800)]
runtime: fix GODEBUG=schedtrace=X delay handling.

debug.schedtrace is an int32. Convert it to int64 before
multiplying with constant 1000000. Otherwise, schedtrace
values more than 2147 result in int32 overflow causing
incorrect delays between traces.

Change-Id: I064e8d7b432c1e892a705ee1f31a2e8cdd2c3ea3
Reviewed-on: https://go-review.googlesource.com/17712
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Ian Lance Taylor <iant@golang.org>

9 years agomath/big: fix misuse of Unicode
Russ Cox [Mon, 7 Dec 2015 14:52:31 +0000 (09:52 -0500)]
math/big: fix misuse of Unicode

ˆ (U+02C6) is a circumflex accent, not an exponentiation operator.
In the rest of the source code for this package, exponentation is
written as **, so do the same here.

Change-Id: I107b85be242ab79d152eb8a6fcf3ca2b197d7658
Reviewed-on: https://go-review.googlesource.com/17671
Reviewed-by: Robert Griesemer <gri@golang.org>
9 years agosyscall: fix duplicated copyright header in mksysnum_plan9.sh
Brad Fitzpatrick [Thu, 10 Dec 2015 19:57:46 +0000 (19:57 +0000)]
syscall: fix duplicated copyright header in mksysnum_plan9.sh

Fixes #13569

Change-Id: I04c9a44437414177e2fee260518cfe53a67e6364
Reviewed-on: https://go-review.googlesource.com/17684
Reviewed-by: Rob Pike <r@golang.org>
9 years agonet/url: add tests for RequestURI returning //paths
Russ Cox [Fri, 4 Dec 2015 17:28:15 +0000 (12:28 -0500)]
net/url: add tests for RequestURI returning //paths

Fixes #10433.

Change-Id: I2a8b54e3f07488f204b1d14eb2603842caaf1e5a
Reviewed-on: https://go-review.googlesource.com/17386
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/compile: captureless closures are constants
Keith Randall [Thu, 3 Dec 2015 21:20:58 +0000 (13:20 -0800)]
cmd/compile: captureless closures are constants

In particular, we can initialize globals with them at link time instead
of generating code for them in an init() function.  Less code, less
startup cost.

But the real reason for this change is binary size.  This change reduces
the binary size of hello world by ~4%.

The culprit is fmt.ssFree, a global variable which is a sync.Pool of
scratch scan states.  It is initalized with a captureless closure as the
pool's New action.  That action in turn references all the scanf code.

If you never call any of the fmt.Scanf* routines, ssFree is never used.
But before this change, ssFree is still referenced by fmt's init
function.  That keeps ssFree and all the code it references in the
binary.  With this change, ssFree is initialized at link time.  As a
result, fmt.init never mentions ssFree.  If you don't call fmt.Scanf*,
ssFree is unreferenced and it and the scanf code are not included.

This change is an easy fix for what is generally a much harder problem,
the unnecessary initializing of unused globals (and retention of code
that they reference).  Ideally we should have separate init code for
each global and only include that code if the corresponding global is
live.  (We'd need to make sure that the initializing code has no side
effects, except on the global being initialized.)  That is a much harder
change.

Update #6853

Change-Id: I19d1e33992287882c83efea6ce113b7cfc504b67
Reviewed-on: https://go-review.googlesource.com/17398
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agotime: reject invalid day of month in Parse
Rob Pike [Thu, 10 Dec 2015 18:41:08 +0000 (10:41 -0800)]
time: reject invalid day of month in Parse

There was back-and-forth on this but it has been decided to fix the original
complaint, which was easy.

Fixes #7268.

Change-Id: I6b607c49ad44579086aba2c4f4c5424b97fbed64
Reviewed-on: https://go-review.googlesource.com/17710
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agonet/http/httptest: don't crash in mime sniffing if HeaderMap is nil
Brad Fitzpatrick [Thu, 10 Dec 2015 18:11:15 +0000 (10:11 -0800)]
net/http/httptest: don't crash in mime sniffing if HeaderMap is nil

Fixes some failing Google tests when run under Go tip (1.6).

Updates #12986

Change-Id: I0ca4d20f6103d10ea9464e45730085401336dada
Reviewed-on: https://go-review.googlesource.com/17698
Reviewed-by: Russ Cox <rsc@golang.org>
Reviewed-by: Nodir Turakulov <nodir@google.com>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

9 years agonet/http: make NewRequest with empty method mean GET
Brad Fitzpatrick [Thu, 10 Dec 2015 18:24:03 +0000 (10:24 -0800)]
net/http: make NewRequest with empty method mean GET

Until recently, we always permitted an empty string to NewRequest.
Keep that property, since it broke tests within in Google when trying
out Go 1.6, and probably would've broken others too.

Change-Id: Idddab1ae7b9423d5caac00af2c897fe1065b600b
Reviewed-on: https://go-review.googlesource.com/17699
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agodoc: go1.6.txt: go doc searches in breadth-first order
Rob Pike [Thu, 10 Dec 2015 18:09:55 +0000 (10:09 -0800)]
doc: go1.6.txt: go doc searches in breadth-first order

Change-Id: I12a43b15e81a5e0b174c1c49e77f8307c567233b
Reviewed-on: https://go-review.googlesource.com/17697
Reviewed-by: Rob Pike <r@golang.org>
9 years agocmd/doc: search the tree in breadth-first order
Rob Pike [Wed, 9 Dec 2015 22:39:26 +0000 (14:39 -0800)]
cmd/doc: search the tree in breadth-first order

This is a simple change to the command that should resolve problems like finding
vendored packages before their non-vendored siblings. By searching in breadth-first
order, we find the matching package lowest in the hierarchy, which is more likely
to be correct than the deeper one, such as a vendored package, that will be found
in a depth-first scan.

This may be sufficient to resolve the issue, and has the merit that it is very easy
to explain. I will leave the issue open for now in case my intuition is wrong.

Update #12423

Change-Id: Icf69e8beb1845277203fcb7d19ffb7cca9fa41f5
Reviewed-on: https://go-review.googlesource.com/17691
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/compile: don't truncate tiny float constants to 0 in error messages
Robert Griesemer [Thu, 10 Dec 2015 01:22:48 +0000 (17:22 -0800)]
cmd/compile: don't truncate tiny float constants to 0 in error messages

Fixes #13559.

Change-Id: I6fe8b5083192e8eb6c1b3ca1919fde81a00ccb7e
Reviewed-on: https://go-review.googlesource.com/17695
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
9 years agonet, internal/syscall/windows: fix interface and address identification on windows
Mikio Hara [Fri, 4 Dec 2015 10:06:01 +0000 (19:06 +0900)]
net, internal/syscall/windows: fix interface and address identification on windows

The current implementation including Go 1.5 through 1.5.2 misuses
Windows API and mishandles the returned values from GetAdapterAddresses
on Windows. This change fixes various issues related to network facility
information by readjusting interface and interface address parsers.

Updates #5395.
Updates #10530.
Updates #12301.
Updates #12551.
Updates #13542.
Fixes #12691.
Fixes #12811.
Fixes #13476.
Fixes #13544.

Also fixes fragile screen scraping test cases in net_windows_test.go.

Additional information for reviewers:

It seems like almost all the issues above have the same root cause and
it is misunderstanding of Windows API. If my interpretation of the
information on MSDN is correctly, current implementation contains the
following bugs:

- SIO_GET_INTERFACE_LIST should not be used for IPv6. The behavior of
  SIO_GET_INTERFACE_LIST is different on kernels and probably it doesn't
  work correctly for IPv6 on old kernels such as Windows XP w/ SP2.
  Unfortunately MSDN doesn't describe the detail of
  SIO_GET_INTERFACE_LIST, but information on the net suggests so.

- Fetching IP_ADAPTER_ADDRESSES structures with fixed size area may not
  work when using IPv6. IPv6 generates ton of interface addresses for
  various addressing scopes. We need to adjust the area appropriately.

- PhysicalAddress field of IP_ADAPTER_ADDRESSES structure may have extra
  space. We cannot ignore PhysicalAddressLength field of
  IP_ADAPTER_ADDRESS structure.

- Flags field of IP_ADAPTER_ADDRESSES structure doesn't represent any of
  administratively and operatinal statuses. It just represents settings
  for windows network adapter.

- MTU field of IP_ADAPTER_ADDRESSES structure may have a uint32(-1) on
  64-bit platform. We need to convert the value to interger
  appropriately.

- IfType field of IP_ADAPTER_ADDRESSES structure is not a bit field.
  Bitwire operation for the field is completely wrong.

- OperStatus field of IP_ADAPTER_ADDRESSES structure is not a bit field.
  Bitwire operation for the field is completely wrong.

- IPv6IfIndex field of IP_ADAPTER_ADDRESSES structure is just a
  substitute for IfIndex field. We cannot prefer IPv6IfIndex to IfIndex.

- Windows XP, 2003 server and below don't set OnLinkPrefixLength field
  of IP_ADAPTER_UNICAST_ADDRESS structure. We cannot rely on the field
  on old kernels. We can use FirstPrefix field of IP_ADAPTER_ADDRESSES
  structure and IP_ADAPTER_PREFIX structure instead.

- Length field of IP_ADAPTER_{UNICAST,ANYCAST,MULTICAST}_ADDRESS
  sturecures doesn't represent an address prefix length. It just
  represents a socket address length.

Change-Id: Icabdaf7bd1d41360a981d2dad0b830b02b584528
Reviewed-on: https://go-review.googlesource.com/17412
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
9 years agonet/http: run more tests in http2 mode
Brad Fitzpatrick [Wed, 9 Dec 2015 22:02:46 +0000 (22:02 +0000)]
net/http: run more tests in http2 mode

Failing ones are marked skipped.

Fixes #13543 (was just a test issue)
Updates #13555 (to be fixed later)
Updates #13556 (to be fixed later)
Updates #13557 (to be fixed later)
Fixes bug in golang.org/cl/17428 (http1 now uses HTTP status 431, not 413)

Change-Id: I8f7431fee35f2fc081cfe2c232ae75a00800a60b
Reviewed-on: https://go-review.googlesource.com/17683
Reviewed-by: Blake Mizerany <blake.mizerany@gmail.com>
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
Reviewed-by: Burcu Dogan <jbd@google.com>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/compile: remove unneeded error message cleanup pass
Matthew Dempsky [Wed, 9 Dec 2015 20:53:28 +0000 (12:53 -0800)]
cmd/compile: remove unneeded error message cleanup pass

This code used to be necessary because of the error messages generated
by the YACC-based parser, but they're no longer relevant under the new
recursive descent parser:

  - LBRACE no longer exists, so "{ or {" can never occur.

  - The parser never generates error messages about "@" or "?" now
    (except in import sections, where they're actually legitimate).

  - The s/LLITERAL/litbuf/ substitution is handled in p.syntax_error.

Change-Id: Id39f747e4aa492c5830d14a47b161920bd4589ad
Reviewed-on: https://go-review.googlesource.com/17690
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocmd/compile, cmd/internal/obj: ignore AUSEFIELD
Ian Lance Taylor [Wed, 9 Dec 2015 02:29:26 +0000 (18:29 -0800)]
cmd/compile, cmd/internal/obj: ignore AUSEFIELD

When using GOEXPERIMENT=fieldtrack, we can see AUSEFIELD instructions.
We generally want to ignore them.

No tests because as far as I can tell there are no tests for
GOEXPERIMENT=fieldtrack.

Change-Id: Iee26f25592158e5db691a36cf8d77fc54d051314
Reviewed-on: https://go-review.googlesource.com/17610
Reviewed-by: David Symonds <dsymonds@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agoruntime/race: update two stale references
Brad Fitzpatrick [Wed, 9 Dec 2015 17:51:23 +0000 (17:51 +0000)]
runtime/race: update two stale references

Fixes #13550

Change-Id: I407daad8b94f6773d7949ba27981d26cbfd2cdf4
Reviewed-on: https://go-review.googlesource.com/17682
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agonet/http: clarify some RoundTripper behaviors
Brad Fitzpatrick [Wed, 9 Dec 2015 17:07:44 +0000 (17:07 +0000)]
net/http: clarify some RoundTripper behaviors

Fixes #12796
Updates #13444

Change-Id: I56840c0baf9b32a683086a80f5db1c5ea0a7aedf
Reviewed-on: https://go-review.googlesource.com/17680
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agomath/big: fix typo
Brad Fitzpatrick [Wed, 9 Dec 2015 14:55:10 +0000 (14:55 +0000)]
math/big: fix typo

Found by github user asukakenji.

Change-Id: I4c76316b69e8a243fb6bf280283f3722e728d853
Reviewed-on: https://go-review.googlesource.com/17641
Reviewed-by: Robert Griesemer <gri@golang.org>
9 years agonet/http: make testCaseSensitiveMethod also test http2
Emmanuel Odeke [Wed, 9 Dec 2015 03:17:02 +0000 (20:17 -0700)]
net/http: make testCaseSensitiveMethod also test http2

Change-Id: I6427b0914fc4ccd9b841eae3aefe3b7a355fc436
Reviewed-on: https://go-review.googlesource.com/17596
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agonet: add TestInterfaceHardwareAddrWithGetmac
Alex Brainman [Wed, 9 Dec 2015 04:53:45 +0000 (15:53 +1100)]
net: add TestInterfaceHardwareAddrWithGetmac

Use Windows getmac command to verify interface
MAC addresses net package returns.

The test is to be enabled once issue #12691 is fixed.

Updates #12691

Change-Id: Ic28c83303590cb4d48ee025250d4b6e30683bfd4
Reviewed-on: https://go-review.googlesource.com/17632
Reviewed-by: Mikio Hara <mikioh.mikioh@gmail.com>
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agopath/filepath: remove code working around Join bug
Alex Brainman [Wed, 9 Dec 2015 01:53:57 +0000 (12:53 +1100)]
path/filepath: remove code working around Join bug

EvalSymlinks code assumes that Join has a bug
(see issue #11551 for details). But issue #11551 has
been fixed. Remove the workaround so it does not
confuses us when we read code next time.

Change-Id: I06bea20189f01f9922237c05516847353d8e4736
Reviewed-on: https://go-review.googlesource.com/17620
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agodoc: update go1.6.txt
Dave Cheney [Wed, 9 Dec 2015 03:20:36 +0000 (14:20 +1100)]
doc: update go1.6.txt

Change-Id: I1164c7a76cf6e6c48ca5153d4c163f5962c4f0cd
Reviewed-on: https://go-review.googlesource.com/17622
Reviewed-by: Dave Cheney <dave@cheney.net>
9 years agonet: add TestInterfaceAddrsWithNetsh
Alex Brainman [Tue, 8 Dec 2015 04:13:19 +0000 (15:13 +1100)]
net: add TestInterfaceAddrsWithNetsh

Use windows netsh command to verify interface
addresses and netmasks net package returns.

The test is to be enabled once issue #12811
is fixed.

Updates #12811

Change-Id: I191e350a1403e5133791d4ec59561fefa24f5c61
Reviewed-on: https://go-review.googlesource.com/17478
Reviewed-by: Mikio Hara <mikioh.mikioh@gmail.com>
Run-TryBot: Mikio Hara <mikioh.mikioh@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agocmd/compile: recognize labels even if they have the same name as packages
Robert Griesemer [Wed, 9 Dec 2015 01:23:23 +0000 (17:23 -0800)]
cmd/compile: recognize labels even if they have the same name as packages

Another (historic) artifact due to partially resolving symbols too early.

Fixes #13539.

Change-Id: Ie720c491cfa399599454f384b3a9735e75d4e8f1
Reviewed-on: https://go-review.googlesource.com/17600
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Damien Neil <dneil@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

9 years agonet/http: testClientHead now in http2 mode
Emmanuel Odeke [Tue, 8 Dec 2015 08:11:30 +0000 (01:11 -0700)]
net/http: testClientHead now in http2 mode

Change-Id: I21317e3f60502759efc026cdfdb0b1d5f411cae9
Reviewed-on: https://go-review.googlesource.com/17521
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agonet/http: sniffTest for 5953 set in http2 mode
Emmanuel Odeke [Sat, 5 Dec 2015 00:21:08 +0000 (17:21 -0700)]
net/http: sniffTest for 5953 set in http2 mode

Change-Id: I345b6ea7024a1c85c805b2d2e22f9f79aeca5816
Reviewed-on: https://go-review.googlesource.com/17462
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agonet/http: clientHeadContentLength test now in http2 mode
Emmanuel Odeke [Tue, 8 Dec 2015 09:18:57 +0000 (02:18 -0700)]
net/http: clientHeadContentLength test now in http2 mode

Change-Id: If6cc1298ad550790ee981c1b5dc618a14947dc4a
Reviewed-on: https://go-review.googlesource.com/17526
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agonet/http: testClientRedirectEatsBody test now in http2 mode
Emmanuel Odeke [Tue, 8 Dec 2015 09:48:10 +0000 (02:48 -0700)]
net/http: testClientRedirectEatsBody test now in http2 mode

Change-Id: I181346a8d2a8a96fd790cde087c8fd73af911298
Reviewed-on: https://go-review.googlesource.com/17529
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agonet/http: update bundled http2 and add tests for two fixed issues
Brad Fitzpatrick [Tue, 8 Dec 2015 23:01:40 +0000 (23:01 +0000)]
net/http: update bundled http2 and add tests for two fixed issues

Updates to http2's golang.org/cl/17590 (git rev 1796f9b8b)

Fixes #13495
Fixes #13532

Change-Id: I9b95ab438e1d895c75d031d8fcf2605921182a5e
Reviewed-on: https://go-review.googlesource.com/17591
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

9 years agonet/http: zeroLengthPostAndResponse test now in http2 mode
Emmanuel Odeke [Tue, 8 Dec 2015 10:10:45 +0000 (03:10 -0700)]
net/http: zeroLengthPostAndResponse test now in http2 mode

Change-Id: I282a399657c71f7b6406fcdde8afc82a13bdfe5f
Reviewed-on: https://go-review.googlesource.com/17530
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agonet/http: testStreamingGet now in http2 mode
Emmanuel Odeke [Tue, 8 Dec 2015 09:01:03 +0000 (02:01 -0700)]
net/http: testStreamingGet now in http2 mode

Change-Id: I9a1fe057f5bf008fa16577a7d71064050aea47e9
Reviewed-on: https://go-review.googlesource.com/17525
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agonet/http: update bundled http2 copy
Brad Fitzpatrick [Tue, 8 Dec 2015 21:42:52 +0000 (21:42 +0000)]
net/http: update bundled http2 copy

Updates golang.org/x/net/http2 to git rev 438097d76

Fixes #13444

Change-Id: I699ac02d23b56db3e8a27d3f599ae56cd0a5b4b2
Reviewed-on: https://go-review.googlesource.com/17570
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>

9 years agoencoding/pem: make TestFuzz testing/quick safe
Matt T. Proud [Tue, 8 Dec 2015 12:02:17 +0000 (13:02 +0100)]
encoding/pem: make TestFuzz testing/quick safe

This adapts pem.TestFuzz to sanitize the generated Block fields,
because the encoder and wireformat do not differentiate between nil
and empty slices and maps, while reflect.DeepEqual rightfully does.
In the commit mentioned below, we adapt quick.Value in
testing/quick to generate these value states, which had heretofore
been impossible with the standard library fuzz test facility.

This commit is a piecemeal extraction from ...

  https://go-review.googlesource.com/#/c/16470

..., which rsc requested to be separated from the nil slice and map
generations.

Change-Id: Iec751a2b0082af6e672a09dc9b7f4b4fb309e8a8
Reviewed-on: https://go-review.googlesource.com/17499
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoencoding/json: document Indent's preservation of trailing spaces
Russ Cox [Tue, 8 Dec 2015 19:09:59 +0000 (14:09 -0500)]
encoding/json: document Indent's preservation of trailing spaces

Fixes #13520.

Change-Id: Ia70cc44be3912167b369d7f74d3436331975c300
Reviewed-on: https://go-review.googlesource.com/17561
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agocrypto/elliptic: resample private keys if out of range.
Adam Langley [Fri, 4 Dec 2015 22:17:03 +0000 (14:17 -0800)]
crypto/elliptic: resample private keys if out of range.

The orders of the curves in crypto/elliptic are all very close to a
power of two. None the less, there is a tiny bias in the private key
selection.

This change makes the distribution uniform by resampling in the case
that a private key is >= to the order of the curve. (It also switches
from using BitSize to Params().N.BitLen() because, although they're the
same value here, the latter is technically the correct thing to do.)

The private key sampling and nonce sampling in crypto/ecdsa don't have
this issue.

Fixes #11082.

Change-Id: Ie2aad563209a529fa1cab522abaf5fd505c7269a
Reviewed-on: https://go-review.googlesource.com/17460
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agodoc: add heading IDs to Code of Conduct
Chris Broadfoot [Mon, 7 Dec 2015 21:45:06 +0000 (13:45 -0800)]
doc: add heading IDs to Code of Conduct

Fixes #13514

Change-Id: I3903d3926ed4f5d54cfb77209d93c950b832b933
Reviewed-on: https://go-review.googlesource.com/17511
Reviewed-by: Francesc Campoy Flores <campoy@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
9 years agogo/parser, go/types: report invalid else branch in if statements
Robert Griesemer [Fri, 4 Dec 2015 01:28:46 +0000 (17:28 -0800)]
go/parser, go/types: report invalid else branch in if statements

- Only accept valid if statement syntax in go/parser.

- Check AST again in go/types since it may have been modified and the
  AST doesn't preclude other statements in the else branch of an if
  statement.

- Removed a test from gofmt which verified that old-style if statements
  permitting any statement in the else branch were correctly reformatted.
  It's been years since we switched to the current syntax; no need to
  support this anymore.

- Added a comment to go/printer.

Fixes #13475.

Change-Id: Id2c8fbcc68b719cd511027d0412a37266cceed6b
Reviewed-on: https://go-review.googlesource.com/17408
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/go: fix bad shared lib name with buildmode=shared
Aleksandr Demakin [Tue, 25 Aug 2015 16:54:57 +0000 (19:54 +0300)]
cmd/go: fix bad shared lib name with buildmode=shared

Use import paths of packages to build a shared lib name.
Use arguments for meta-packages 'std', 'cmd', and 'all'.

Fixes #12236

Change-Id: If274d63301686ef34e198287eb012f9062541ea0
Reviewed-on: https://go-review.googlesource.com/13921
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 agoencoding/gob: document behavior of zero-valued arrays, slices, and maps
Rob Pike [Thu, 3 Dec 2015 21:06:04 +0000 (13:06 -0800)]
encoding/gob: document behavior of zero-valued arrays, slices, and maps

The documentation was inconsistent. It said zero values were not sent, but
that zero-valued elements of arrays and arrays were sent. But which rule
applies if the array is all zero elements, and is therefore itself a zero value?

The answer is: the array is transmitted. In principle the other choice could
be made, but there would be considerable expense and complexity required
to implement this behavior now, not to mention worries about changes of
behavior.

Therefore we just document the situation: Arrays, slices, and maps are
always encoded. It would perhaps be nice to have sorted this out earlier,
but it was a missed opportunity.

Fixes #13378

Change-Id: I8fae345edfa707fcfa7a3e0160d87ff1ac5cc5a2
Reviewed-on: https://go-review.googlesource.com/17394
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/compile/internal/gc: fix internal compiler error on invalid declaration
Didier Spezia [Sun, 4 Oct 2015 20:33:02 +0000 (20:33 +0000)]
cmd/compile/internal/gc: fix internal compiler error on invalid declaration

Following an empty import, a declaration involving a ? symbol
generates an internal compiler error when the name of the
symbol (in newname function).

package a
import""
var?

go.go:2: import path is empty
go.go:3: internal compiler error: newname nil

Make sure dclname is not called when the symbol is nil.
The error message is now:

go.go:2: import path is empty
go.go:3: invalid declaration
go.go:4: syntax error: unexpected EOF

This CL was initially meant to be applied to the old parser,
and has been updated to apply to the new parser.

Fixes #11610

Change-Id: I75e07622fb3af1d104e3a38c89d9e128e3b94522
Reviewed-on: https://go-review.googlesource.com/15268
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agocmd/compile: avoid converting huge floats to integers
Robert Griesemer [Thu, 3 Dec 2015 23:51:03 +0000 (15:51 -0800)]
cmd/compile: avoid converting huge floats to integers

Fixes #13471.

Change-Id: I232ad1729343d020254e313cfff182695ad6fc54
Reviewed-on: https://go-review.googlesource.com/17401
Reviewed-by: Russ Cox <rsc@golang.org>
9 years agoruntime: best-effort detection of concurrent misuse of maps
Russ Cox [Mon, 7 Dec 2015 19:22:08 +0000 (14:22 -0500)]
runtime: best-effort detection of concurrent misuse of maps

If reports like #13062 are really concurrent misuse of maps,
we can detect that, at least some of the time, with a cheap check.

There is an extra pair of memory writes for writing to a map,
but to the same cache line as h.count, which is often being modified anyway,
and there is an extra memory read for reading from a map,
but to the same cache line as h.count, which is always being read anyway.
So the check should be basically invisible and may help reduce the
number of "mysterious runtime crash due to map misuse" reports.

Change-Id: I0e71b0d92eaa3b7bef48bf41b0f5ab790092487e
Reviewed-on: https://go-review.googlesource.com/17501
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
9 years agocmd/compile/internal/gc: fix panic in Type Stringer
Didier Spezia [Sat, 29 Aug 2015 11:30:10 +0000 (11:30 +0000)]
cmd/compile/internal/gc: fix panic in Type Stringer

The following code:

func n() {(interface{int})}

generates:

3: interface contains embedded non-interface int
3: type %!v(PANIC=runtime error: invalid memory address or nil pointer dereference) is not an expression

It is because the corresponding symbol (Sym field in Type object)
is nil, resulting in a panic in typefmt.

Just skip the symbol if it is nil, so that the error message becomes:

3: interface contains embedded non-interface int
3: type interface { int } is not an expression

Fixes #11614

Change-Id: I219ae7eb01edca264fad1d4a1bd261d026294b00
Reviewed-on: https://go-review.googlesource.com/14015
Reviewed-by: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>