Michael Munday [Mon, 17 Oct 2016 21:10:24 +0000 (17:10 -0400)]
runtime: get s390x vector facility availability from AT_HWCAP
This is a more robust method for obtaining the availability of vx.
Since this variable may be checked frequently I've also now
padded it so that it will be in its own cache line.
I've kept the other check (in hash/crc32) the same for now until
I can figure out the best way to update it.
Austin Clements [Mon, 17 Oct 2016 01:22:02 +0000 (21:22 -0400)]
runtime: keep gcMarkRootCheck happy with spare Gs
oneNewExtraM creates a spare M and G for use with cgo callbacks. The G
doesn't run right away, but goes directly into syscall status. For the
garbage collector, it's marked as "scan valid" and not on the rescan
list, but I forgot to also mark it as "scan done". As a result,
gcMarkRootCheck thinks that the goroutine hasn't been scanned and
panics.
This only affects GODEBUG=gccheckmark=1 mode, since we otherwise skip
the gcMarkRootCheck.
runtime: update heap profile stats after world is started
Updating the heap profile stats is one of the most expensive parts of
mark termination other than stack rescanning, but there's really no
need to do this with the world stopped. Move it to right after we've
started the world back up. This creates a *very* small window where
allocations from the next cycle can slip into the profile, but the
exact point where mark termination happens is so non-deterministic
already that a slight reordering here is unimportant.
runtime: remove gcWork flushes in mark termination
The only reason these flushes are still necessary at all is that
gcmarknewobject doesn't flush its gcWork stats like it's supposed to.
By changing gcmarknewobject to follow the standard protocol, the
flushes become completely unnecessary because mark 2 ensures caches
are flushed (and stay flushed) before we ever enter mark termination.
In the garbage benchmark, this takes roughly 50 µs, which is
surprisingly long for doing nothing. We still double-check after
draining that they are in fact empty.
Ian Lance Taylor [Fri, 14 Oct 2016 21:53:59 +0000 (14:53 -0700)]
cmd/cgo: always use a function literal for pointer checking
The pointer checking code needs to know the exact type of the parameter
expected by the C function, so that it can use a type assertion to
convert the empty interface returned by cgoCheckPointer to the correct
type. Previously this was done by using a type conversion, but that
meant that the code accepted arguments that were convertible to the
parameter type, rather than arguments that were assignable as in a
normal function call. In other words, some code that should not have
passed type checking was accepted.
This CL changes cgo to always use a function literal for pointer
checking. Now the argument is passed to the function literal, which has
the correct argument type, so type checking is performed just as for a
function call as it should be.
Since we now always use a function literal, simplify the checking code
to run as a statement by itself. It now no longer needs to return a
value, and we no longer need a type assertion.
This does have the cost of introducing another function call into any
call to a C function that requires pointer checking, but the cost of the
additional call should be minimal compared to the cost of pointer
checking.
Fixes #16591.
Change-Id: I220165564cf69db9fd5f746532d7f977a5b2c989
Reviewed-on: https://go-review.googlesource.com/31233
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Matthew Dempsky [Tue, 18 Oct 2016 23:11:50 +0000 (16:11 -0700)]
cmd/compile: rework mkbuiltin.go to generate code
Generating binary export data requires a working Go compiler. Even
trickier to change the export data format itself requires a careful
bootstrapping procedure.
Instead, simply generate normal Go code that lets us directly
construct the builtin runtime declarations.
Passes toolstash -cmp.
Fixes #17508.
Change-Id: I4f6078a3c7507ba40072580695d57c87a5604baf
Reviewed-on: https://go-review.googlesource.com/31493
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
Joe Tsai [Tue, 18 Oct 2016 23:57:02 +0000 (16:57 -0700)]
archive/tar: fix parsePAX to be POSIX.1-2001 compliant
Relevant PAX specification:
<<<
If the <value> field is zero length, it shall delete any header
block field, previously entered extended header value, or
global extended header value of the same name.
>>>
We don't delete global extender headers since the Reader doesn't
even support global headers (which the specification admits was
a controversial feature).
Russ Cox [Wed, 19 Oct 2016 13:11:16 +0000 (09:11 -0400)]
text/template: add support for reflect.Value args, results in funcs
Add support for passing reflect.Values to and returning reflect.Values from
any registered functions in the FuncMap, much as if they were
interface{} values. Keeping the reflect.Value instead of round-tripping
to interface{} preserves addressability of the value, which is important
for method lookup.
Change index and a few other built-in functions to use reflect.Values,
making a loop using explicit indexing now match the semantics that
range has always had.
Fixes #14916.
Change-Id: Iae1a2fd9bb426886a7fcd9204f30a2d6ad4646ad
Reviewed-on: https://go-review.googlesource.com/31462 Reviewed-by: Rob Pike <r@golang.org>
Joe Tsai [Wed, 19 Oct 2016 00:51:04 +0000 (17:51 -0700)]
archive/tar: make Reader handle GNU format properly
The GNU format does not have a prefix field, so we should make
no attempt to read it. It does however have atime and ctime fields.
Since Go previously placed incorrect values here, we liberally
read the atime and ctime fields and ignore errors so that old tar
files written by Go can at least be partially read.
This fixes half of #12594. The Writer is much harder to fix.
Russ Cox [Tue, 18 Oct 2016 14:26:07 +0000 (10:26 -0400)]
sync: throw, not panic, for unlock of unlocked mutex
The panic leaves the lock in an unusable state.
Trying to panic with a usable state makes the lock significantly
less efficient and scalable (see early CL patch sets and discussion).
Instead, use runtime.throw, which will crash the program directly.
In general throw is reserved for when the runtime detects truly
serious, unrecoverable problems. This problem is certainly serious,
and, without a significant performance hit, is unrecoverable.
David Crawshaw [Wed, 19 Oct 2016 15:06:36 +0000 (11:06 -0400)]
plugin: mention OS X support and concurrency
Change-Id: I4270bf81511a5bf80ed146f5e66e4f8aeede2aa2
Reviewed-on: https://go-review.googlesource.com/31463 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Robert Griesemer [Wed, 19 Oct 2016 16:56:53 +0000 (09:56 -0700)]
spec: slightly more realistic example for type assertions
For #17428.
Change-Id: Ia902b50cf0c40e3c2167fb573a39d328331c38c7
Reviewed-on: https://go-review.googlesource.com/31449 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Brad Fitzpatrick [Wed, 19 Oct 2016 10:31:15 +0000 (10:31 +0000)]
net/http: make NewRequest set empty Body nil, don't peek Read Body in Transport
This CL makes NewRequest set Body nil for known-zero bodies, and makes
the http1 Transport not peek-Read a byte to determine whether there's
a body.
Background:
Many fields of the Request struct have different meanings for whether
they're outgoing (via the Transport) or incoming (via the Server).
For outgoing requests, ContentLength and Body are documented as:
// Body is the request's body.
//
// For client requests a nil body means the request has no
// body, such as a GET request. The HTTP Client's Transport
// is responsible for calling the Close method.
Body io.ReadCloser
// ContentLength records the length of the associated content.
// The value -1 indicates that the length is unknown.
// Values >= 0 indicate that the given number of bytes may
// be read from Body.
// For client requests, a value of 0 with a non-nil Body is
// also treated as unknown.
ContentLength int64
Because of the ambiguity of what ContentLength==0 means, the http1 and
http2 Transports previously Read the first byte of a non-nil Body when
the ContentLength was 0 to determine whether there was an actual body
(with a non-zero length) and ContentLength just wasn't populated, or
it was actually empty.
That byte-sniff has been problematic and gross (see #17480, #17071)
and was removed for http2 in a previous commit.
... would not send a Content-Length header in their http2 request,
because the size of the reader (even though it was known, being one of
the three common recognized types from NewRequest) was zero, and so
the HTTP Transport thought it was simply unset.
To signal explicitly-zero vs unset-zero, this CL changes NewRequest to
signal explicitly-zero by setting the Body to nil, instead of the
strings.NewReader("") or other zero-byte reader.
This CL also removes the byte sniff from the http1 Transport, like
https://golang.org/cl/31326 did for http2.
Updates #17480
Updates #17071
Change-Id: I329f02f124659bf7d8bc01e2c9951ebdd236b52a
Reviewed-on: https://go-review.googlesource.com/31445
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
- Giving a template definition containing
nothing but spaces has no effect.
- Giving a template definition containing
non-spaces can only be done once per template.
Fixes #16912.
Fixes #16913.
Fixes #17360.
Change-Id: Ie3971b83ab148b7c8bb800fe4a21579566378e3e
Reviewed-on: https://go-review.googlesource.com/31459
Run-TryBot: Russ Cox <rsc@golang.org> Reviewed-by: Rob Pike <r@golang.org> Reviewed-by: Andrew Gerrand <adg@golang.org>
Emmanuel Odeke [Mon, 10 Oct 2016 05:45:17 +0000 (22:45 -0700)]
math/big: implement Float.Scan, type assert fmt interfaces to enforce docs
Implements Float.Scan which satisfies fmt.Scanner interface.
Also enforces docs' interface implementation claims with compile time
type assertions, that is:
+ Float always implements fmt.Formatter and fmt.Scanner
+ Int always implements fmt.Formatter and fmt.Scanner
+ Rat always implements fmt.Formatter
which will ensure that the API claims are strictly matched.
Also note that Float.Scan doesn't handle ±Inf.
Fixes #17391
Change-Id: I3d3dfbe7f602066975c7a7794fe25b4c645440ce
Reviewed-on: https://go-review.googlesource.com/30723 Reviewed-by: Robert Griesemer <gri@golang.org>
Caio Marcelo de Oliveira Filho [Wed, 20 Apr 2016 17:29:30 +0000 (14:29 -0300)]
cmd/go, testing: indicate when no tests are run
For example, testing the current directory:
$ go test -run XXX
testing: warning: no tests to run
PASS
ok testing 0.013s
$
And in a summary:
$ go test -run XXX testing
ok testing 0.013s [no tests to run]
$
These make it easy to spot when the -run regexp hasn't matched anything
or there are no tests. Previously the message was printed in the "current directory"
case when there were no tests at all, but not for no matches, and either way
was not surfaced in the directory list summary form.
Fixes #15211.
Change-Id: I1c82a423d6bd429fb991c9ca964c9d26c96fd3c5
Reviewed-on: https://go-review.googlesource.com/22341 Reviewed-by: Marcel van Lohuizen <mpvl@golang.org>
Russ Cox [Tue, 18 Oct 2016 15:00:00 +0000 (11:00 -0400)]
reflect: correct Type.FieldByNameFunc docs
Fixes #16573.
Change-Id: I5a26eaa8b258cb1861190f9690086725532b8a0d
Reviewed-on: https://go-review.googlesource.com/31354 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Hiroshi Ioka [Sun, 16 Oct 2016 04:25:53 +0000 (13:25 +0900)]
os, syscall: fix incorrect offset calculation in Readlink on windows
Current implementation of syscall.Readlink mistakenly calculates
the end offset of the PrintName field.
Also, there are some cases that the PrintName field is empty.
Instead, the CL uses SubstituteName with correct calculation.
Fixes #15978
Fixes #16145
Change-Id: If3257137141129ac1c552d003726d5b9c08bb754
Reviewed-on: https://go-review.googlesource.com/31118 Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Russ Cox [Tue, 18 Oct 2016 16:34:19 +0000 (12:34 -0400)]
os: reject Rename("old", "new") where new is a directory
Unix rejects this when new is a non-empty directory.
Other systems reject this when new is a directory, empty or not.
Make Unix reject empty directory too.
Fixes #14527.
Change-Id: Ice24b8065264c91c22cba24aa73e142386c29c87
Reviewed-on: https://go-review.googlesource.com/31358
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Robert Griesemer [Thu, 10 Dec 2015 19:18:41 +0000 (11:18 -0800)]
spec: require 16 bit minimum exponent in constants rather than 32
A 16bit binary exponent permits a constant range covering roughly the range
from 7e-9865 to 7e9863 which is more than enough for any practical and
hypothetical constant arithmetic.
Furthermore, until recently cmd/compile could not handle very large exponents
correctly anyway; i.e., the chance that any real programs (but for tests that
explore corner cases) are affected are close to zero.
Finally, restricting the minimum supported range significantly reduces the
implementation complexity in an area that hardly matters in reality for new
or alternative spec-compliant implementations that don't or cannot rely on
pre-existing arbitratry precision arithmetic packages that support a 32bit
exponent range.
This is technically a language change but for the reasons mentioned above
this is unlikely to affect any real programs, and certainly not programs
compiled with the gc or gccgo compilers as they currently support up to
32bit exponents.
Matthew Dempsky [Tue, 18 Oct 2016 21:17:05 +0000 (14:17 -0700)]
cmd/compile: handle unsafe builtins like universal builtins
Reuse the same mechanisms for handling universal builtins like len to
handle unsafe.Sizeof, etc. Allows us to drop package unsafe's export
data, and simplifies some code.
Updates #17508.
Change-Id: I620e0617c24e57e8a2d7cccd0e2de34608779656
Reviewed-on: https://go-review.googlesource.com/31433
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
Mohit Agarwal [Tue, 18 Oct 2016 08:30:30 +0000 (14:00 +0530)]
math: speed up Gamma(+Inf)
Add special case for Gamma(+∞) which speeds it up:
benchmark old ns/op new ns/op delta
BenchmarkGamma-4 14.5 7.44 -48.69%
The documentation for math.Gamma already specifies it as a special
case:
Gamma(+Inf) = +Inf
The original C code that has been used as the reference implementation
(as mentioned in the comments in gamma.go) also treats Gamma(+∞) as a
special case:
if( x == INFINITY )
return(x);
Change-Id: Idac36e19192b440475aec0796faa2d2c7f8abe0b
Reviewed-on: https://go-review.googlesource.com/31370 Reviewed-by: Robert Griesemer <gri@golang.org>
Quentin Smith [Mon, 17 Oct 2016 22:40:18 +0000 (18:40 -0400)]
cmd/go: print more env variables in "go env"
"go env" previously only printed a subset of the documented environment
variables; now it includes everything, such as GO386 and CGO_*.
This also fixes the CGO_CFLAGS environment variable to always have the
same default. According to iant@ and confirmed by testing, cgo can now
understand the default value of CGO_CFLAGS.
Fixes #17191.
Change-Id: Icf75055446dd250b6256ef1139e9ce848f4a9d3b
Reviewed-on: https://go-review.googlesource.com/31330
TryBot-Result: Gobot Gobot <gobot@golang.org>
Run-TryBot: Quentin Smith <quentin@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Dhananjay Nakrani [Mon, 17 Oct 2016 21:17:46 +0000 (14:17 -0700)]
cmd/compile: fix code duplication in race-instrumentation
instrumentnode() accidentally copies parent's already-instrumented nodes
into child's Ninit block. This generates repeated code in race-instrumentation.
This case surfaces only when it duplicates inline-labels, because of
compile time error. In other cases, it silently generates incorrect
instrumented code. This change prevents it from doing so.
Fixes #17449.
Change-Id: Icddf2198990442166307e176b7e20aa0cf6c171c
Reviewed-on: https://go-review.googlesource.com/31317 Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Matthew Dempsky [Mon, 17 Oct 2016 23:03:27 +0000 (16:03 -0700)]
cmd/compile: stop treating interface methods like actual functions
Interface methods can't have function bodies, so there's no need to
process their parameter lists as variable declarations. The only
possible reason would be to check for duplicate parameter names and/or
invalid types, but we do that anyway, and have regression tests for it
(test/funcdup.go).
Michael Munday [Tue, 18 Oct 2016 15:47:42 +0000 (11:47 -0400)]
cmd/internal/obj/{ppc64,s390x}: mark functions with small stacks NOSPLIT
This change omits the stack check on ppc64 and s390x when the size of
a stack frame is less than obj.StackSmall. This is an optimization
x86 already performs.
The effect on s390x isn't huge because we were already omitting the
stack check when the frame size was 0 (it shaves about 1K from the
size of bin/go). On ppc64 however this change reduces the size of the
.text section in bin/go by 33K (1%).
Updates #13379 (for ppc64).
Change-Id: I6af0eb987646bea47fcaf0a812db3496bab0f680
Reviewed-on: https://go-review.googlesource.com/31357 Reviewed-by: David Chase <drchase@google.com>
David du Colombier [Tue, 18 Oct 2016 13:21:46 +0000 (15:21 +0200)]
net: always wake up the readers on close on Plan 9
Previously, in acceptPlan9 we set netFD.ctl to the listener's
/net/tcp/*/listen file instead of the accepted connection's
/net/tcp/*/ctl file.
In netFD.Read, we write "close" to netFD.ctl to close the
connection and wake up the readers. However, in the
case of an accepted connection, we got the error
"write /net/tcp/*/listen: inappropriate use of fd"
because the /net/tcp/*/listen doesn't handle the "close" message.
In this case, the connection wasn't closed and the readers
weren't awake.
We modified the netFD structure so that netFD.ctl represents
the accepted connection and netFD.listen represents the
listener.
Change-Id: Ie38c7dbaeaf77fe9ff7da293f09e86d1a01b3e1e
Reviewed-on: https://go-review.googlesource.com/31390
Run-TryBot: David du Colombier <0intro@gmail.com> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Russ Cox [Thu, 13 Oct 2016 02:58:47 +0000 (22:58 -0400)]
encoding/xml: add wildcard support for collecting all attributes
- Like ",any" for elements, add ",any,attr" for attributes to allow
a mop-up field that gets any otherwise unmapped attributes.
- Map attributes to fields of type slice by extending the slice,
just like for elements.
- Allow storing an attribute into an xml.Attr directly, to provide
a way to record the name.
Combined, these three independent features allow
AllAttrs []Attr `xml:",any,attr"`
to collect all attributes not otherwise spoken for in a particular struct.
Tests based on CL 16292 by Charles Weill.
Fixes #3633.
Change-Id: I2d75817f17ca8752d7df188080a407836af92611
Reviewed-on: https://go-review.googlesource.com/30946 Reviewed-by: Quentin Smith <quentin@golang.org>
Russ Cox [Tue, 18 Oct 2016 01:08:48 +0000 (21:08 -0400)]
bufio: read from underlying reader at most once in Read
Fixes #17059.
Change-Id: I5c7ee46604399f7dc3c3c49f964cbb1aa6c0d621
Reviewed-on: https://go-review.googlesource.com/31320 Reviewed-by: Ian Lance Taylor <iant@golang.org>
David du Colombier [Mon, 17 Oct 2016 17:15:37 +0000 (19:15 +0200)]
net: close the connection gracefully on Plan 9
Previously, we used to write the "hangup" message to
the TCP connection control file to be able to close
a connection, while waking up the readers.
The "hangup" message closes the TCP connection with a
RST message. This is a problem when closing a connection
consecutively to a write, because the reader may not have
time to acknowledge the message before the connection is
closed, resulting in loss of data.
We use a "close" message, newly implemented in the Plan 9
kernel to be able to close a TCP connection gracefully with a FIN.
Updates #15464.
Change-Id: I2050cc72fdf7a350bc6c9128bae7d14af11e599c
Reviewed-on: https://go-review.googlesource.com/31271
Run-TryBot: David du Colombier <0intro@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Russ Cox [Tue, 18 Oct 2016 03:30:38 +0000 (23:30 -0400)]
testing: fix flag usage messages
Fixes #16404.
Change-Id: Iabaeeef3eff2fff6e5ed2d6bc9ef9c2f6d1cb5e7
Reviewed-on: https://go-review.googlesource.com/31332 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Russ Cox [Mon, 17 Oct 2016 21:20:48 +0000 (17:20 -0400)]
os/exec: allow simultaneous cmd.Wait and Write of cmd.StdinPipe
cmd.StdinPipe returns an io.WriteCloser.
It's reasonable to expect the caller not to call Write and Close simultaneously,
but there is an implicit Close in cmd.Wait that's not obvious.
We already synchronize the implicit Close in cmd.Wait against
any explicit Close from the caller. Also synchronize that implicit
Close against any explicit Write from the caller.
Fixes #9307.
Change-Id: I8561e9369d6e5ac88dfbca1175549f6dfa04b8ac
Reviewed-on: https://go-review.googlesource.com/31148 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Russ Cox [Mon, 17 Oct 2016 19:47:18 +0000 (15:47 -0400)]
time: be consistent about representation of UTC location in Time struct
In the zero Time, the (not user visible) nil *Location indicates UTC.
In the result of t.UTC() and other ways to create times in specific
zones, UTC is indicated by a non-nil *Location, specifically &utcLoc.
This creates a representation ambiguity exposed by comparison with ==
or reflect.DeepEqual or the like.
Change time.Time representation to use only nil, never &utcLoc,
to represent UTC. This eliminates the ambiguity.
Mikio Hara [Tue, 18 Oct 2016 10:43:04 +0000 (19:43 +0900)]
net: update docs on network interface API
This change documents that the InterfaceAddrs function is less usable on
multi-homed IP nodes because of the lack of network interface
identification information.
Also updates documentation on exposed network interface API.
Brad Fitzpatrick [Fri, 14 Oct 2016 10:45:59 +0000 (11:45 +0100)]
net/http: make Server Handler's Request.Context be done on conn errors
This CL changes how the http1 Server reads from the client.
The goal of this change is to make the Request.Context given to Server
Handlers become done when the TCP connection dies (has seen any read
or write error). I didn't finish that for Go 1.7 when Context was
added to http package.
We can't notice the peer disconnect unless we're blocked in a Read
call, though, and previously we were only doing read calls as needed,
when reading the body or the next request. One exception to that was
the old pre-context CloseNotifier mechanism.
The implementation of CloseNotifier has always been tricky. The past
few releases have contained the complexity and moved the
reading-from-TCP-conn logic into the "connReader" type. This CL
extends connReader to make sure that it's always blocked in a Read
call, at least once the request body has been fully consumed.
In the process, this deletes all the old CloseNotify code and unifies
it with the context cancelation code. The two notification mechanisms
are nearly identical, except the CloseNotify path always notifies on
the arrival of pipelined HTTP/1 requests. We might want to change that
in a subsequent commit. I left a TODO for that. For now there's no
change in behavior except that the context now cancels as it was
supposed to.
As a bonus that fell out for free, a Handler can now use CloseNotifier
and Hijack together in the same request now.
Fixes #15224 (make http1 Server always in a Read, like http2)
Fixes #15927 (cancel context when underlying connection closes)
Updates #9763 (CloseNotifier + Hijack)
Change-Id: I972cf6ecbab7f1230efe8cc971e89f8e6e56196b
Reviewed-on: https://go-review.googlesource.com/31173
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Alberto Donizetti [Mon, 17 Oct 2016 19:59:10 +0000 (21:59 +0200)]
math/big: add benchmarks for big.Float String
In addition to the DecimalConversion benchmark, that exercises the
String method of the internal decimal type on a range of small shifts,
add a few benchmarks for the big.Float String method. They can be used
to obtain more realistic data on the real-world performance of
big.Float printing.
Matthew Dempsky [Mon, 17 Oct 2016 21:08:36 +0000 (14:08 -0700)]
cmd/link: remove some unnecessary comments
The comments about pcln functions are obsolete since those functions
now live in cmd/internal/obj. The copyright header is redundant with
the existing one at the top of the file.
Change-Id: I568fd3d259253a0d8eb3b0a157d008df1b5de106
Reviewed-on: https://go-review.googlesource.com/31315 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Cherry Zhang [Mon, 17 Oct 2016 17:40:18 +0000 (13:40 -0400)]
cmd/link: trampoline support for external linking on ARM
all.bash passes with -debugtramp=2 (except the unavoidable
disassembly test as we change instructions). And successfully
build k8s.io/kubernetes/cmd/hyperkube in both internal linking
and external linking mode.
Fixes #17028.
Change-Id: Ic8fac6a394488155c5eba9215662db1c1086e24b
Reviewed-on: https://go-review.googlesource.com/31143 Reviewed-by: David Chase <drchase@google.com>
Adam Langley [Wed, 12 Oct 2016 18:20:27 +0000 (11:20 -0700)]
crypto/tls: only store a single nonce for AES-GCM.
Although an AEAD, in general, can be used concurrently in both the seal
and open directions, TLS is easier. Since the transport keys are
different for different directions in TLS, an AEAD will only ever be
used in one direction. Thus we don't need separate buffers for seal and
open because they can never happen concurrently.
Also, fix the nonce size to twelve bytes since the fixed-prefix
construction for AEADs is superseded and will never be used for anything
else now.
Adam Langley [Mon, 17 Oct 2016 21:26:57 +0000 (14:26 -0700)]
Revert "crypto/tls: add CloseWrite method to Conn"
This reverts commit c6185aa63217c84a1a73c578c155e7d4dec6cec8. That
commit seems to be causing flaky failures on the builders. See
discussion on the original thread: https://golang.org/cl/25159.
Change-Id: I26e72d962d4efdcee28a0bc61a53f246b046df77
Reviewed-on: https://go-review.googlesource.com/31316
Run-TryBot: Adam Langley <agl@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Russ Cox [Mon, 10 Oct 2016 20:18:43 +0000 (16:18 -0400)]
math/big: add (*Int).Sqrt
This is needed for some of the more complex primality tests
(to filter out exact squares), and while the code is simple the
boundary conditions are not obvious, so it seems worth having
in the library.
Michael Munday [Wed, 14 Sep 2016 14:42:14 +0000 (10:42 -0400)]
cmd/compile: merge loads into operations on s390x
Adds the new canMergeLoad function which can be used by rules to
decide whether a load can be merged into an operation. The function
ensures that the merge will not reorder the load relative to memory
operations (for example, stores) in such a way that the block can no
longer be scheduled.
This new function enables transformations such as:
MOVD 0(R1), R2
ADD R2, R3
to:
ADD 0(R1), R3
The two-operand form of the following instructions can now read a
single memory operand:
- ADD
- ADDC
- ADDW
- MULLD
- MULLW
- SUB
- SUBC
- SUBE
- SUBW
- AND
- ANDW
- OR
- ORW
- XOR
- XORW
Improves SHA3 performance by 6-8%.
Updates #15054.
Change-Id: Ibcb9122126cd1a26f2c01c0dfdbb42fe5e7b5b94
Reviewed-on: https://go-review.googlesource.com/29272
Run-TryBot: Michael Munday <munday@ca.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
Inspired by Alberto Donizetti's observations in
https://go-review.googlesource.com/#/c/30099/.
name old time/op new time/op delta
DecimalConversion-8 138µs ± 1% 136µs ± 2% -1.85% (p=0.000 n=10+10)
10 runs each, measured on a Mac Mini, 2.3 GHz Intel Core i7.
Performance improvements varied between -1.25% to -4.4%; -1.85% is
about in the middle of the observed improvement. The generated code
is slightly shorter in the inner loops of the conversion code.
Change-Id: I10fb3b2843da527691c39ad5e5e5bd37ed63e2fa
Reviewed-on: https://go-review.googlesource.com/31250 Reviewed-by: Alan Donovan <adonovan@google.com>
Austin Clements [Mon, 10 Oct 2016 16:18:00 +0000 (12:18 -0400)]
runtime: fix GC assist retry path
GC assists retry if preempted or if they fail to park. However, on the
retry path they currently use stale statistics. In particular, the
retry can use "debtBytes", but debtBytes isn't updated when the debt
changes (since other than retries it is only used once). Also, though
less of a problem, the if the assist ratio has changed while the
assist was blocked, the retry will still use the old assist ratio.
Fix all of this by simply making the retry jump back to where we
compute these statistics, rather than just after.
Change-Id: I2ed8b4f0fc9f008ff060aa926f4334b662ac7d3f
Reviewed-on: https://go-review.googlesource.com/30701 Reviewed-by: Rick Hudson <rlh@golang.org>
Austin Clements [Sun, 16 Oct 2016 22:23:39 +0000 (18:23 -0400)]
runtime: fix getArgInfo for deferred reflection calls
getArgInfo for reflect.makeFuncStub and reflect.methodValueCall is
necessarily special. These have dynamically determined argument maps
that are stored in their context (that is, their *funcval). These
functions are written to store this context at 0(SP) when called, and
getArgInfo retrieves it from there.
This technique works if getArgInfo is passed an active call frame for
one of these functions. However, getArgInfo is also used in
tracebackdefers, where the "call" is not a true call with an active
stack frame, but a deferred call. In this situation, getArgInfo
currently crashes because tracebackdefers passes a frame with sp set
to 0. However, the entire approach used by getArgInfo is flawed in
this situation because the wrapper has not actually executed, and
hence hasn't saved this metadata to any stack frame.
In the defer case, we know the *funcval from the _defer itself, so we
can fix this by teaching getArgInfo to use the *funcval context
directly when its available, and otherwise get it from the active call
frame.
While we're here, this commit simplifies getArgInfo a bit by making it
play more nicely with the type system. Rather than decoding the
*reflect.methodValue that is the wrapper's context as a *[2]uintptr,
just write out a copy of the reflect.methodValue type in the runtime.
Austin Clements [Thu, 13 Oct 2016 14:44:57 +0000 (10:44 -0400)]
runtime: print a message on bad morestack
If morestack runs on the g0 or gsignal stack, it currently performs
some abort operation that typically produces a signal (e.g., it does
an INT $3 on x86). This is useful if you're running in a debugger, but
if you're not, the runtime tries to trap this signal, which is likely
to send the program into a deeper spiral of collapse and lead to very
confusing diagnostic output.
Help out people trying to debug without a debugger by making morestack
print an informative message before blowing up.
Change-Id: I2814c64509b137bfe20a00091d8551d18c2c4749
Reviewed-on: https://go-review.googlesource.com/31133
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Rick Hudson <rlh@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Russ Cox [Mon, 17 Oct 2016 17:29:31 +0000 (13:29 -0400)]
cmd/dist, go/build: make CGO_ENABLED during make.bash sticky
Per discussion on #12808, it's a bit odd that if you do
CGO_ENABLED=0 ./make.bash
then you get a toolchain that still tries to use cgo.
So make the CGO_ENABLED setting propagate into
the resulting toolchain as the default setting for that
environment variable, like we do with other variables
like CC and GOROOT.
No reasonable way to test automatically, but I did
test by hand that after the above command, 'go env'
shows CGO_ENABLED=0; before it showed CGO_ENABLED=1.
Lynn Boger [Thu, 13 Oct 2016 17:59:07 +0000 (12:59 -0500)]
bytes: improve performance for bytes.Compare on ppc64x
This improves the performance for byte.Compare by rewriting
the cmpbody function in runtime/asm_ppc64x.s. The previous code
had a simple loop which loaded a pair of bytes and compared them,
which is inefficient for long buffers. The updated function checks
for 8 or 32 byte chunks and then loads and compares double words where
possible.
Because the byte.Compare result indicates greater or less than,
the doubleword loads must take endianness into account, using a
byte reversed load in the little endian case.
This change imports the chacha20poly1305 and poly1305 packages from
x/crypto at 5f4e837b98443e9e7a65072235205993af565d85. These packages
will be used to support the ChaCha20-Poly1305 AEAD in crypto/tls.
Russ Cox [Thu, 13 Oct 2016 17:45:31 +0000 (13:45 -0400)]
cmd/go: accept plain file for .vcs (instead of directory)
Sometimes .git is a plain file; maybe others will follow.
This CL matches CL 21430, made in x/tools/go/vcs.
The change in the Swift test case makes the test case
pass by changing the test to match current behavior,
which I assume is better than the reverse.
(The test only runs locally and without -short, so the
builders are not seeing this particular failure.)
For #10322.
Change-Id: Iccd08819a01c5609a2880b9d8a99af936e20faff
Reviewed-on: https://go-review.googlesource.com/30948
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Lynn Boger [Tue, 11 Oct 2016 14:26:40 +0000 (09:26 -0500)]
cmd/link: add trampolines for too far calls in ppc64x
This change adds support for trampolines on ppc64x when using
internal linking, in the case where the offset to the branch
target is larger than what fits in the field provided by the
branch instruction.
Then they can't use (*MyStr)(nil) as an argument to an SQL call via
database/sql, because *MyStr also implements driver.Value, but via a
compiler-generated wrapper which checks whether the pointer is nil and
panics if so.
We now accept (*MyStr)(nil) and map it to "nil" (an SQL "NULL")
if the Valuer method is implemented on MyStr instead of *MyStr.
If a user implements the driver.Value interface with a pointer
receiver, they retain full control of what nil means:
type MyStr string
func (s *MyStr) Value() (driver.Value, error) {
if s == nil {
return "missing MyStr", nil
}
return strings.ToUpper(string(*s)), nil
}
Russ Cox [Thu, 13 Oct 2016 19:27:04 +0000 (15:27 -0400)]
cmd/dist: use debug/pe directly for cmd/link
Delete vendored copy.
Change-Id: I06e9d3b709553a1a8d06275e99bd8f617aac5788
Reviewed-on: https://go-review.googlesource.com/31011 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Russ Cox [Thu, 13 Oct 2016 19:13:41 +0000 (15:13 -0400)]
cmd/dist: copy needed packages from standard library during bootstrap
This allows use of newer math/big (and later debug/pe)
without maintaining a vendored copy somewhere in cmd.
Use for math/big, deleting cmd/compile/internal/big.
Change-Id: I2bffa7a9ef115015be29fafdb02acc3e7a665d11
Reviewed-on: https://go-review.googlesource.com/31010 Reviewed-by: Minux Ma <minux@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Russ Cox [Wed, 12 Oct 2016 20:54:02 +0000 (16:54 -0400)]
encoding/json: fix decoding of null into Unmarshaler, TextUnmarshaler
1. Define behavior for Unmarshal of JSON null into Unmarshaler and
TextUnmarshaler. Specifically, an Unmarshaler will be given the
literal null and can decide what to do (because otherwise
json.RawMessage is impossible to implement), and a TextUnmarshaler
will be skipped over (because there is no text to unmarshal), like
most other inappropriate types. Document this in Unmarshal, with a
reminder in UnmarshalJSON about handling null.
2. Test all this.
3. Fix the TextUnmarshaler case, which was returning an unmarshalling
error, to match the definition.
4. Fix the error that had been used for the TextUnmarshaler, since it
was claiming that there was a JSON string when in fact the problem was
NOT having a string.
5. Adjust time.Time and big.Int's UnmarshalJSON to ignore null, as is
conventional.
Ben Burkert [Sun, 24 Jul 2016 22:13:56 +0000 (15:13 -0700)]
crypto/tls: add CloseWrite method to Conn
The CloseWrite method sends a close_notify alert record to the other
side of the connection. This record indicates that the sender has
finished sending on the connection. Unlike the Close method, the sender
may still read from the connection until it recieves a close_notify
record (or the underlying connection is closed). This is analogous to a
TCP half-close.
Martin Möhrmann [Fri, 2 Sep 2016 15:04:41 +0000 (17:04 +0200)]
runtime: speed up non-ASCII rune decoding
Copies utf8 constants and EncodeRune implementation from unicode/utf8.
Adds a new decoderune implementation that is used by the compiler
in code generated for ranging over strings. It does not handle
ASCII runes since these are handled directly before calls to decoderune.
The DecodeRuneInString implementation from unicode/utf8 is not used
since it uses a lookup table that would increase the use of cpu caches.
Adds more tests that check decoding of valid and invalid utf8 sequences.