]> Cypherpunks repositories - gostls13.git/log
gostls13.git
6 years agodoc: add net/http additions to go1.11 notes
Brad Fitzpatrick [Mon, 16 Jul 2018 16:35:04 +0000 (16:35 +0000)]
doc: add net/http additions to go1.11 notes

Change-Id: Ib2da2b0ceb33867a41b0b9b7e4dcecb11f964b73
Reviewed-on: https://go-review.googlesource.com/124035
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agocmd/cgo: add note about bug writing C pointers to uninitialized C memory
Ian Lance Taylor [Thu, 12 Jul 2018 18:57:05 +0000 (11:57 -0700)]
cmd/cgo: add note about bug writing C pointers to uninitialized C memory

Describe the problem as a bug, since it is not implied by the rest of
the pointer passing rules, and it may be possible to fix it.

Updates #19928

Change-Id: I2d336e7336b2a215c0b8cf909a203201ef1b054e
Reviewed-on: https://go-review.googlesource.com/123658
Reviewed-by: Austin Clements <austin@google.com>
6 years agocmd/go: make TestNewReleaseRebuildsStalePackagesInGOPATH pass again
Ian Lance Taylor [Fri, 13 Jul 2018 16:57:17 +0000 (09:57 -0700)]
cmd/go: make TestNewReleaseRebuildsStalePackagesInGOPATH pass again

The test TestNewReleaseRebuildsStalePackagesInGOPATH is not run in
short mode, so people tend to not notice when it fails. It was failing
due to the build cache. Make it pass again by 1) changing it to modify
the package in a way visible to the compiler, so that the change is
not hidden by caching; 2) accepting "not installed but available in
build cache" as always being a valid reason for a stale package, as go
list does not try to figure out an underlying reason for why a package
is stale when it finds it in the build cache but not installed.

Updates #24436

Change-Id: Iaeaa298f153451ec913a653dd4e6da79a33055bb
Reviewed-on: https://go-review.googlesource.com/123815
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
6 years agodoc/go1.11: update with latest output from x/build/cmd/relnote
Andrew Bonventre [Mon, 16 Jul 2018 14:22:18 +0000 (10:22 -0400)]
doc/go1.11: update with latest output from x/build/cmd/relnote

Change-Id: I4055ecc39b1b250cce7b32241f13dbc05567361f
Reviewed-on: https://go-review.googlesource.com/124015
Reviewed-by: Bryan C. Mills <bcmills@google.com>
6 years agoruntime/pprof: add a fake mapping when /proc/self/maps is unavailable
Hana (Hyang-Ah) Kim [Fri, 13 Jul 2018 20:11:24 +0000 (16:11 -0400)]
runtime/pprof: add a fake mapping when /proc/self/maps is unavailable

Profile's Mapping field is currently populated by reading /proc/self/maps.
On systems where /proc/self/maps is not available, the profile generated
by Go's runtime will not have any Mapping entry. Pprof command then adds
a fake entry and links all Location entries in the profile with the fake
entry to be used during symbolization.
https://github.com/google/pprof/blob/a8644067d5a3c9a6386e7c88fa4a3d9d37877ca3/internal/driver/fetch.go#L437

The fake entry is not enough to suppress the error or warning messages
pprof command produces. We need to tell pprof that Location entries are
symbolized already by Go runtime and pprof does not have to attempt to
perform further symbolization.

In #25743, we made Go runtime mark Mapping entries with HasFunctions=true
when all Location entries from the Mapping entries are successfully
symbolized. This change makes the Go runtime add a fake mapping entry,
otherwise the pprof command tool would add, and set the HasFunctions=true
following the same logic taken when the real mapping information is
available.

Updates #19790.
Fixes #26255. Tested pprof doesn't report the error message any more
for pure Go program.

Change-Id: Ib12b62e15073f5d6c80967e44b3e8709277c11bd
Reviewed-on: https://go-review.googlesource.com/123779
Run-TryBot: Hyang-Ah Hana Kim <hyangah@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agodoc: update Design and Types sections of the FAQ
Rob Pike [Sat, 14 Jul 2018 06:46:57 +0000 (16:46 +1000)]
doc: update Design and Types sections of the FAQ

Update #26107.

Change-Id: I8bfa5b01ce953c53f7fd7a866d0ece61ba04c618
Reviewed-on: https://go-review.googlesource.com/123919
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agodoc: update Values, Writing Code, and Pointers and Allocation sections of the FAQ
Rob Pike [Sat, 14 Jul 2018 04:52:31 +0000 (14:52 +1000)]
doc: update Values, Writing Code, and Pointers and Allocation sections of the FAQ

Significant surgery done to the Versioning section, bringing it closer to
modern thinking.

Also add a question about constants.

Update #26107.

Change-Id: Icf70b7228503c6baaeab0b95ee3e6bee921575aa
Reviewed-on: https://go-review.googlesource.com/123918
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agodoc: update Implementation and Performance sections of the FAQ
Rob Pike [Sat, 14 Jul 2018 04:06:28 +0000 (14:06 +1000)]
doc: update Implementation and Performance sections of the FAQ

Changes are mostly about making more about now than about the past,
changing some verb tenses, and mentioning gollvm (which should
be pronounced "gollum" if you ask me).

Update #26107

Change-Id: I6c14f42b9fc2684259d4ba8bc149d7ec9bb83d15
Reviewed-on: https://go-review.googlesource.com/123917
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agodoc: rearrange the description of GOMAXPROCS
Rob Pike [Thu, 12 Jul 2018 23:08:13 +0000 (09:08 +1000)]
doc: rearrange the description of GOMAXPROCS

The old text was written when it was only 1 by default, which
changed a long time ago.

Also add a note that GOMAXPROCS does not limit the total
number of threads.

Change-Id: I104ccd7266d11335320a4d7f5671fb09ed641f88
Reviewed-on: https://go-review.googlesource.com/123916
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agomisc/cgo: fix test on iOS
Keith Randall [Sun, 15 Jul 2018 00:18:45 +0000 (17:18 -0700)]
misc/cgo: fix test on iOS

The test in CL 123715 doesn't work on iOS, it needs to use a different
version scheme to determine whether SecKeyAlgorithm and friends exist.
Restrict the old version test to OSX only.

The same problem occurs on iOS: the functions tested don't exist before
iOS 10.  But we don't have builders below iOS 10, so it isn't a big issue.
If we ever get older builders, or someone wants to run all.bash on an
old iOS, they'll need to figure out the right incantation.

Update #24161
Update #26355

Change-Id: Ia3ace86b00486dc172ed00c0c6d668a95565bff7
Reviewed-on: https://go-review.googlesource.com/123959
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agocmd/vet,sync: check lock values more precisely
Dan Kortschak [Sun, 1 Jul 2018 00:03:55 +0000 (09:33 +0930)]
cmd/vet,sync: check lock values more precisely

Fixes #26165

Change-Id: I1f3bd193af9b6f8461c736330952b6e50d3e00d9
Reviewed-on: https://go-review.googlesource.com/121876
Reviewed-by: Alan Donovan <adonovan@google.com>
Run-TryBot: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agonet/http: don't cancel Request.Context on pipelined Server requests
Brad Fitzpatrick [Fri, 13 Jul 2018 19:31:35 +0000 (19:31 +0000)]
net/http: don't cancel Request.Context on pipelined Server requests

See big comment in code.

Fixes #23921

Change-Id: I2dbd1acc2e9da07a71f9e0640aafe0c59a335627
Reviewed-on: https://go-review.googlesource.com/123875
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agoos/exec: document ExtraFiles is not supported on windows
Adam Shannon [Fri, 13 Jul 2018 18:44:51 +0000 (13:44 -0500)]
os/exec: document ExtraFiles is not supported on windows

Fixes #26182

Change-Id: I1181e191f4742f166c9b67a6f41332a237cf0ede
Reviewed-on: https://go-review.googlesource.com/123855
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agocmd/go: fix module replace flag parsed bug
Baokun Lee [Fri, 13 Jul 2018 20:17:19 +0000 (04:17 +0800)]
cmd/go: fix module replace flag parsed bug

In CL 122404, we change -replace syntax from => to =.
And we also need to change this and the tests.

Fixes golang/go#26373.

Change-Id: I2d4e85e10c1578540cc7673b93d849270940d776
Reviewed-on: https://go-review.googlesource.com/123778
Reviewed-by: Bryan C. Mills <bcmills@google.com>
Run-TryBot: Bryan C. Mills <bcmills@google.com>

6 years agocmd/go: remove windows Skip in test that wasn't removed
Brad Fitzpatrick [Fri, 13 Jul 2018 20:56:46 +0000 (20:56 +0000)]
cmd/go: remove windows Skip in test that wasn't removed

Was supposed to be removed from CL 123757

Change-Id: I74405adab58be103f8e4a0c2405567f480a0e622
Reviewed-on: https://go-review.googlesource.com/123895
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
6 years agocmd/go: fix TestModFindModulePath on Windows
Russ Cox [Fri, 13 Jul 2018 20:38:24 +0000 (16:38 -0400)]
cmd/go: fix TestModFindModulePath on Windows

The os.RemoveAll(tg.tempdir) was not a good idea.

Change-Id: I6f78cff887044186649cbf3ee04a58abdbcb71e2
Reviewed-on: https://go-review.googlesource.com/123757
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
6 years agoregexp: reword Match documentation to be more like Find
Russ Cox [Fri, 13 Jul 2018 14:28:56 +0000 (10:28 -0400)]
regexp: reword Match documentation to be more like Find

Before:

  // Find returns a slice holding the text of the leftmost match in b of the regular expression.

  // Match checks whether a textual regular expression matches a byte slice.

After:

  // Match reports whether the byte slice b contains any match of the regular expression re.

The use of different wording for Find and Match always makes me think
that Match required the entire string to match while Find clearly allows
a substring to match.

This CL makes the Match wording correspond more closely to Find,
to try to avoid that confusion.

Change-Id: I97fb82d5080d3246ee5cf52abf28d2a2296a5039
Reviewed-on: https://go-review.googlesource.com/123736
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agocmd/link: eliminate a couple of unused DWARF attrs
Than McIntosh [Wed, 11 Jul 2018 16:30:05 +0000 (12:30 -0400)]
cmd/link: eliminate a couple of unused DWARF attrs

The linker's DWARF generation occasionally computes and attaches an
attribute X to a type even though the type's abbrev doesn't have the
specified attr. For example, the DW_TAG_subroutine_type abbrev entry
has no type attribute, but a type attr is given to it (wasting
memory). Similarly there are some places where a byte size attr is
added to a DIE whose abbrev lacks that attr. This patch trims away a
few of these not-needed attrs, saving some very tiny amount of memory.

Updates #26186

Change-Id: I69e853df468ac54b07772a614b4106d7c4dae01d
Reviewed-on: https://go-review.googlesource.com/123296
Reviewed-by: Heschi Kreinick <heschi@google.com>
Run-TryBot: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agoruntime: check tgkill error in Debug tests
Ian Lance Taylor [Wed, 11 Jul 2018 04:07:55 +0000 (21:07 -0700)]
runtime: check tgkill error in Debug tests

Updates #25519

Change-Id: Ibcdf948fd38d8d02d467b62213566ec0d7ce0d6a
Reviewed-on: https://go-review.googlesource.com/123180
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
6 years agohttp/http/httputil: add ReverseProxy.ErrorHandler
Julien Salleyron [Mon, 13 Nov 2017 22:32:07 +0000 (23:32 +0100)]
http/http/httputil: add ReverseProxy.ErrorHandler

This permits specifying an ErrorHandler to customize the RoundTrip
error handling if the backend fails to return a response.

Fixes #22700
Fixes #21255

Change-Id: I8879f0956e2472a07f584660afa10105ef23bf11
Reviewed-on: https://go-review.googlesource.com/77410
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agocmd/go: call flag.Parse to properly initialize test environment variables
Jakub Čajka [Fri, 13 Jul 2018 13:23:12 +0000 (15:23 +0200)]
cmd/go: call flag.Parse to properly initialize test environment variables

Executing tests in src/cmd/go/internal/modfetch/codehost/git_test.go in enviroment
witout outside connectivity in to the internet results in tests failure:

2018/07/13 14:31:14 git clone --mirror https://vcs-test.golang.org/git/gitrepo1 /tmp/gitrepo-test-996701800/gitrepo2 in : exit status 128:
        Cloning into bare repository '/tmp/gitrepo-test-996701800/gitrepo2'...
        fatal: unable to access 'https://vcs-test.golang.org/git/gitrepo1/': Could not resolve host: vcs-test.golang.org
FAIL    cmd/go/internal/modfetch/codehost       0.307s

Fixes #26007

Change-Id: Ia39d8b3215c920dad6c0c58ffabb0b2ab39bb55c
Reviewed-on: https://go-review.googlesource.com/123735
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agonet/http: correct use of byte slice in js syscall
Johan Brandhorst [Thu, 12 Jul 2018 16:32:02 +0000 (16:32 +0000)]
net/http: correct use of byte slice in js syscall

syscall/js does not allow []byte to be used in direct inputs to
its JavaScript manipulation methods since
https://github.com/golang/go/commit/bafe466a9537d8ea5ac5767504628803302ebb12.
Unfortunately, this use of a byte slice was missed, so any
uses of the WASM Roundtripper with a body will panic.
This ensures the byte slice is appropriately converted
before being passed to syscall.

Fixes #26349

Change-Id: I83847645d71ce310c1eee3decddbac990fae166b
GitHub-Last-Rev: 3914bda2ff457deba4b76182a48d26ce8948b4cb
GitHub-Pull-Request: golang/go#26350
Reviewed-on: https://go-review.googlesource.com/123537
Run-TryBot: Emmanuel Odeke <emm.odeke@gmail.com>
Reviewed-by: Richard Musiol <neelance@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agocmd/go/internal/modload: more aggressive symlink analysis in FindModulePath
Dmitry Doroginin [Fri, 13 Jul 2018 12:31:01 +0000 (12:31 +0000)]
cmd/go/internal/modload: more aggressive symlink analysis in FindModulePath

Fixes golang/go#26217.

Change-Id: I0c456047ee31aa78b72acc413446651ca8c3882a
GitHub-Last-Rev: b700554d883b43b57c6619e31a5f8fcb22b1d71f
GitHub-Pull-Request: golang/vgo#5
Reviewed-on: https://go-review.googlesource.com/123755
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
6 years agonet/url: don't escape sub-delims in fragment
Kunpei Sakai [Fri, 8 Sep 2017 15:39:20 +0000 (00:39 +0900)]
net/url: don't escape sub-delims in fragment

According to RFC-3986, the sub-delims chars should not be escaped in
fragment.
So this change fixes current behavior a bit.

Fixes #19917

Change-Id: I1a8deb93255d979532f75bae183c3fb53a05d395
Reviewed-on: https://go-review.googlesource.com/61650
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agomisc/cgo/test: fix issue 24161 test for 1.11 and earlier
Keith Randall [Thu, 12 Jul 2018 23:51:27 +0000 (16:51 -0700)]
misc/cgo/test: fix issue 24161 test for 1.11 and earlier

The test uses functions from C that were introduced in OSX 1.12.
Include stubs for those functions when compiling for 1.11 and earlier.
This test really a compile-time test, it doesn't matter much what the
executed code actually does.
Use a nasty #define hack to work around the fact that cgo doesn't
support static global variables.

Update #24161
Fixes #26355

Change-Id: Icf6f7bc9b6b36cacc81d5d0e033a2ebaff7e0298
Reviewed-on: https://go-review.googlesource.com/123715
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agodoc: update Concurrency, Functions and Methods, and Control Flow sections
Rob Pike [Thu, 12 Jul 2018 05:57:56 +0000 (15:57 +1000)]
doc: update Concurrency, Functions and Methods, and Control Flow sections

Many parts of the FAQ are dusty and need cleaning up.
This is the first of a series of changes to bring it up to date.
Since it was first written at the time of release, some of the
ideas and background have changed, and some historical
remarks are largely irrelevant now.

Update #26107.

Change-Id: I1f36df7d8ecc8a1a033d5ac4fa1edeece25ed6b4
Reviewed-on: https://go-review.googlesource.com/123496
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agogo/types: record type information after detecting error
Rebecca Stambler [Thu, 12 Jul 2018 16:52:18 +0000 (12:52 -0400)]
go/types: record type information after detecting error

The existing implementation stops recording type information once it
encounters an error. This results in missing type information that is
needed by various tools. This change handles a few commonly encountered
cases by continuing to check subtrees after errors. Also, add tests for
cases where the package fails to type-check.

Updates #22467

Change-Id: I1bb48d4cb8ae5548dca63bdd785ea2f69329e92b
Reviewed-on: https://go-review.googlesource.com/123578
Run-TryBot: Rebecca Stambler <rstambler@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
6 years agonet/http: update bundled http2
Brad Fitzpatrick [Thu, 12 Jul 2018 19:12:38 +0000 (19:12 +0000)]
net/http: update bundled http2

Updates bundled x/net/http2 to git rev d0887baf81f4 for:

    http2: ignore unknown 1xx responses like HTTP/1
    https://golang.org/cl/123615

    http2: fix bug in earlier CL 123615
    https://golang.org/cl/123675

Also along for the ride, but without any effect:

    http2: export a field of an internal type for use by net/http
    https://golang.org/cl/123656

Fixes #26189
Updates #17739

Change-Id: I1955d844d74113efbcbbdaa7d7a7faebb2225b45
Reviewed-on: https://go-review.googlesource.com/123676
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agocmd/go: fix tests from x/vgo repo
Russ Cox [Thu, 12 Jul 2018 17:52:14 +0000 (13:52 -0400)]
cmd/go: fix tests from x/vgo repo

This CL fixes up tests from the x/vgo repo that are failing
on some of the builders.
It will be submitted together with CL 123576.

Change-Id: I6bec81a93ad4f7116e8edc8c15beafa25747530c
Reviewed-on: https://go-review.googlesource.com/123580
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agocmd/go: merge module support from x/vgo repo
Russ Cox [Thu, 12 Jul 2018 03:55:39 +0000 (23:55 -0400)]
cmd/go: merge module support from x/vgo repo

This CL corresponds to CL 123361, the final manual CL in that repo,
making this the final manual sync.

All future commits will happen in this repo (the main Go repo),
and we'll update x/vgo automatically with a fixed patch+script.

Change-Id: I572243309c1809727604fd704705a23c30e85d1a
Reviewed-on: https://go-review.googlesource.com/123576
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
6 years agonet/http: make Transport.CloseIdleConnections close non-bundled http2.Transport
Brad Fitzpatrick [Thu, 12 Jul 2018 18:47:31 +0000 (18:47 +0000)]
net/http: make Transport.CloseIdleConnections close non-bundled http2.Transport

Previously Transport.CloseIdleConnections only closed the HTTP/2
Transport's idle connections if the HTTP/2 transport was configured
automatically via the bundled copy (in h2_bundle.go).

This makes it also work if the user called http2.ConfigureTransport
themselves using golang.org/x/net/http2 instead of the bundled copy.

No tests because we have no current way to run such cross-repo tests,
at least in any efficient or non-flaky way.

Tested by hand that:

    package main

    import (
        "net/http"

        "golang.org/x/net/http2"
    )

    func main() {
        tr := &http.Transport{}
        http2.ConfigureTransport(tr)
        tr.CloseIdleConnections()
    }

... now works and calls the x/net/http2.Transport.CloseIdleConnections
code. (I threw in a print statement locally)

Fixes #22891 once CL 123656 is also in.

Change-Id: Id697fd3e7877c3a988bc3c3368b88940ba56cfd0
Reviewed-on: https://go-review.googlesource.com/123657
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agocmd/cgo: fix cgo bad typedefs
Keith Randall [Tue, 10 Jul 2018 20:47:15 +0000 (13:47 -0700)]
cmd/cgo: fix cgo bad typedefs

Two fixes:

1) Typedefs of the bad typedefs should also not be rewritten to the
   underlying type.  They shouldn't just be uintptr, though, they should
   retain the C naming structure.  For example, in C:

   typedef const __CFString * CFStringRef;
   typedef CFStringRef SecKeyAlgorithm;

   we want the Go:

   type _Ctype_CFStringRef uintptr
   type _Ctype_SecKeyAlgorithm = _Ctype_CFStringRef

2) We need more types than just function arguments/return values.
   At least we need types of global variables, so when we see a reference to:

   extern const SecKeyAlgorithm kSecKeyAlgorithmECDSASignatureDigestX962SHA1;

   we know that we need to investigate the type SecKeyAlgorithm.
   Might as well just find every typedef and check the badness of all of them.
   This requires looping until a fixed point of known types is reached.
   Usually it takes just 2 iterations, sometimes 3.

Fixes #24161

Change-Id: I32ca7e48eb4d4133c6242e91d1879636f5224ea9
Reviewed-on: https://go-review.googlesource.com/123177
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agocmd/compile: handle degenerate write barrier case
Keith Randall [Thu, 12 Jul 2018 18:44:05 +0000 (11:44 -0700)]
cmd/compile: handle degenerate write barrier case

If both branches of a write barrier test go to the same block,
then there's no unsafe points.

This can only happen if the resulting memory state is somehow dead,
which can only occur in degenerate cases, like infinite loops. No
point in cleaning up the useless branch in these situations.

Fixes #26024.

Change-Id: I93a7df9fdf2fc94c6c4b1fe61180dc4fd4a0871f
Reviewed-on: https://go-review.googlesource.com/123655
Reviewed-by: David Chase <drchase@google.com>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/compile: add LocalAddr that takes SP,mem operands
David Chase [Tue, 3 Jul 2018 15:34:38 +0000 (11:34 -0400)]
cmd/compile: add LocalAddr that takes SP,mem operands

Lack of a well-defined order between VarDef and related
address operations sometimes causes problems with store order
and write barrier transformations; glitches in the order are
made irreparable (by later optimizations) if the two parts of
the glitch straddle a split in the original block caused by
insertion of a write barrier diamond.

Fix this by creating a LocalAddr for addresses of locals
(what VarDef matters for) that takes a memory input to
help make the order explicit.  Addr is modified to only
be legal for SB operand, so there is no overlap between
Addr and LocalAddr uses (there may be some downstream
cleanup from this).

Changes to generic.rules and rewrite.go ensure that codegen
tests continue to pass; CSE of LocalAddr is impaired, not
quite sure of the cost.

Fixes #26105.

Change-Id: Id4192b4440aa4e9d7ba54a465c456df9b530b515
Reviewed-on: https://go-review.googlesource.com/122483
Run-TryBot: David Chase <drchase@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
6 years agocmd/go: delete TestGoBuildDashA*
Russ Cox [Thu, 12 Jul 2018 04:16:00 +0000 (00:16 -0400)]
cmd/go: delete TestGoBuildDashA*

The behavior of -a no longer changes depending on which kind
of branch of Go you are using (the new build cache fixed all that).
These tests are not doing anything useful (and failing).

Change-Id: I1c65120a3e05286e888951d61bca4a903e2c1158
Reviewed-on: https://go-review.googlesource.com/123575
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agosyscall: convert Windows AddrinfoW.Addr from uintptr to syscall.Pointer
Brad Fitzpatrick [Wed, 11 Jul 2018 23:25:38 +0000 (23:25 +0000)]
syscall: convert Windows AddrinfoW.Addr from uintptr to syscall.Pointer

API compatibility violation, but it wasn't safe before and people
should be using golang.org/x/sys/windows instead.

Fixes #24820

Change-Id: I3b43493f56b3116924b9d848a294899279f314fd
Reviewed-on: https://go-review.googlesource.com/123455
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agoruntime: fix reference to funcdata.go in comment
Xia Bin [Thu, 12 Jul 2018 10:34:34 +0000 (18:34 +0800)]
runtime: fix reference to funcdata.go in comment

Change-Id: I6c8699cd71b41cf8d178a0af3a745a19dcf60905
Reviewed-on: https://go-review.googlesource.com/123536
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agotime: fix typo in Truncate example
Keith Rarick [Thu, 12 Jul 2018 06:08:21 +0000 (23:08 -0700)]
time: fix typo in Truncate example

The existing example code uses variable name d, but
prints t in its output. It's needlessly confusing.

Change-Id: I67bef3c732e84d2d89819f96b4b62663630fd69e
Reviewed-on: https://go-review.googlesource.com/123516
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agonet/http: update bundled http2
Brad Fitzpatrick [Thu, 12 Jul 2018 05:02:36 +0000 (05:02 +0000)]
net/http: update bundled http2

Updates bundled x/net/http2 to git rev cffdcf67 for:

    http2: use GetBody unconditionally on Transport retry, when available
    https://golang.org/cl/123476

    http2: a closed stream cannot receive data
    https://golang.org/cl/111676

Updates #25009
Updates #25023

Change-Id: I84f50cc50c0fa5a3c34f0037a9cb1ef468e5f0d9
Reviewed-on: https://go-review.googlesource.com/123515
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agodoc: ArgsSizeUnknown it's defined in cmd/internal/objabi now
Xia Bin [Thu, 12 Jul 2018 06:35:39 +0000 (14:35 +0800)]
doc: ArgsSizeUnknown it's defined in cmd/internal/objabi now

Change-Id: I877c82788f3edbcb0b334b42049c1a06f36a6477
Reviewed-on: https://go-review.googlesource.com/123517
Reviewed-by: Rob Pike <r@golang.org>
6 years agotest: add test of valid code that gccgo failed to compile
Ian Lance Taylor [Thu, 12 Jul 2018 00:33:01 +0000 (17:33 -0700)]
test: add test of valid code that gccgo failed to compile

Updates #26340

Change-Id: I3bc7cd544ea77df660bbda7de99a009b63d5be1b
Reviewed-on: https://go-review.googlesource.com/123477
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agonet/http: add tests to validate that Client.Timeout closes connections
Brad Fitzpatrick [Wed, 11 Jul 2018 22:55:16 +0000 (22:55 +0000)]
net/http: add tests to validate that Client.Timeout closes connections

For #23399

Change-Id: I9bc7c21fda6bfa89af2e7656e5c85aa9edd4f29e
Reviewed-on: https://go-review.googlesource.com/123435
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agodoc: clarify a sentence about *_js.go
Rob Pike [Thu, 12 Jul 2018 01:37:17 +0000 (11:37 +1000)]
doc: clarify a sentence about *_js.go

Change "have to" to "need to" for clarity and to avoid a
peculiar English idiom.

Change-Id: Iec2b1f841d0353dd7925f8f934fe82d4ed059d7d
Reviewed-on: https://go-review.googlesource.com/123495
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agocmd/compile: fix ICE due to missing inline function body
Matthew Dempsky [Wed, 11 Jul 2018 23:43:51 +0000 (16:43 -0700)]
cmd/compile: fix ICE due to missing inline function body

For golang.org/cl/74110, I forgot that you can use range-based for
loops to extract key values from a map value.

This wasn't a problem for the binary format importer, because it was
more tolerant about missing inline function bodies. However, the
indexed importer is more particular about this.

We could potentially just make it more lenient like the binary
importer, but tweaking the logic here is easy enough and seems like
the preferable solution.

Fixes #26341.

Change-Id: I54564dcd0be60ea393f8a0f6954b7d3d61e96ee5
Reviewed-on: https://go-review.googlesource.com/123475
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Heschi Kreinick <heschi@google.com>
6 years agocmd/dist: enable build cache always
Russ Cox [Wed, 11 Jul 2018 18:31:17 +0000 (14:31 -0400)]
cmd/dist: enable build cache always

GOCACHE=off is not a reliable signal of user intent.

At startup the go command fills in an empty GOCACHE with the effective setting.
If $HOME is set, then GOCACHE ends up somewhere in $HOME/.cache.
But if $HOME is unset, then the go command sets GOCACHE=off explicitly.

That environment is used for invoking "go tool dist".

So if the machine has no $HOME, then go tool dist ends up with the cache
disabled even though the user was not trying to disable the cache.
This affects the linux-ppc64le builder, which appears to be unique
among builders in not having $HOME set. So that builder is running
with no build cache.

Now that there is a cmd/go test that needs the cache to be on,
the linux-ppc64le builder is failing.

In the next release we intend to force the use of the build cache
always. This CL is not doing that: it's only forcing the use of the
build cache during all.bash, which won't affect the majority of
our users (they run pre-build binary releases).

If this is a problem we can roll it back and fix the linux-ppc64le
builders some other way.

While we're here, print a few more useful variables in 'go tool dist env'
and sort the output.

Change-Id: I66548aa8990d0794cbc0f2069b739ab1834898dd
Reviewed-on: https://go-review.googlesource.com/123297
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agodoc: mention that *_js.go files are now ignored and treated like a GOOS
Brad Fitzpatrick [Wed, 11 Jul 2018 21:30:09 +0000 (21:30 +0000)]
doc: mention that *_js.go files are now ignored and treated like a GOOS

Fixes #26329

Change-Id: Id87fd106e69d3d9682653eb753b1de616adeed2b
Reviewed-on: https://go-review.googlesource.com/123416
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agotime: add clarification to Timer.Stop documentation
Caleb Martinez [Wed, 11 Jul 2018 13:40:14 +0000 (13:40 +0000)]
time: add clarification to Timer.Stop documentation

Fixes #26220

Change-Id: I05703912be594d985ad3ccb3a9757ae21ec738ab
GitHub-Last-Rev: e4649847e015e573ee169c7d6db29b9437841afd
GitHub-Pull-Request: golang/go#26293
Reviewed-on: https://go-review.googlesource.com/122715
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agodoc: update the minimum support Windows version
Brad Fitzpatrick [Wed, 11 Jul 2018 21:14:42 +0000 (21:14 +0000)]
doc: update the minimum support Windows version

Also, remove some test code that was trying to work on XP and fix up
some comments referencing XP.

Fixes #26191
Updates #23380

Change-Id: I0b7319fe1954afddb22d396e5ec91d8c960268d8
Reviewed-on: https://go-review.googlesource.com/123415
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agoruntime: don't say "different packages" if they may not be different
Ian Lance Taylor [Wed, 11 Jul 2018 20:51:39 +0000 (13:51 -0700)]
runtime: don't say "different packages" if they may not be different

Fix the panic message produced for an interface conversion error to
only say "types from different packages" if they are definitely from
different packges. If they may be from the same package, say "types
from different scopes."

Updates #18911
Fixes #26094

Change-Id: I0cea50ba31007d88e70c067b4680009ede69bab9
Reviewed-on: https://go-review.googlesource.com/123395
Reviewed-by: Austin Clements <austin@google.com>
6 years agocmd/link: use BestSpeed for DWARF compression
Austin Clements [Wed, 11 Jul 2018 20:13:04 +0000 (16:13 -0400)]
cmd/link: use BestSpeed for DWARF compression

DWARF compression accounts for roughly 30% of the linker's time. This
CL switches from DefaultCompression to BestSpeed, which virtually
eliminates this time. This roughly halves the overhead of handling
DWARF in the linker:

name \ time/op         nodwarf     dwarf        dwarf-speed
BuildCmdGoAll          10.0s ±11%   10.6s ± 5%   10.8s ± 5%
             nodwarf                  +6.41%       +8.03%
               dwarf                                 ~
LinkCmdGo              626ms ± 5%  1096ms ± 2%   860ms ± 2%
             nodwarf                 +75.17%      +37.36%
               dwarf                              -21.59%

Previously, enabling DWARF had a 75% overhead in link time for cmd/go.
This change reduces this overhead to 37% (a 22% reduction).

The effect on binary size is minimal compared to DefaultCompression,
and still substantially better than no compression:

             cmd/go bytes
nodwarf      10106953
dwarf        12159049 nodwarf+20%
dwarf-speed  12408905 nodwarf+23%
dwarf-nozlib 17766473 nodwarf+76%

Updates #26318.

Change-Id: I33bb7caa038a2753c29104501663daf4839e7054
Reviewed-on: https://go-review.googlesource.com/123356
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Heschi Kreinick <heschi@google.com>
6 years agoruntime: document philosophy behind runtime atomic usage
Austin Clements [Wed, 11 Jul 2018 13:02:46 +0000 (09:02 -0400)]
runtime: document philosophy behind runtime atomic usage

Based on Dmitry Vyukov's comments in CL 65210.

Change-Id: I5dce7286b0d180cd43cad3aaf70f537fafcda588
Reviewed-on: https://go-review.googlesource.com/123275
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
6 years agonet/http: fix rare Transport leak, remove incorrect defensive logic
Brad Fitzpatrick [Wed, 11 Jul 2018 17:35:13 +0000 (17:35 +0000)]
net/http: fix rare Transport leak, remove incorrect defensive logic

Remove some incorrect code that was present after since I added
support for idle timeouts in CL 22670.

This code actually caused a bug (a rare goroutine leak) rather than
prevent a bogus connection reuse.

The t.idleMu mutex already protects most the invariants, including an
explicit Stop call. There's only one Stop call on that timer, and it's
guarded by t.idleMu. What idleMu doesn't protect against is the timer
firing on its own. But we don't need code to protect against that case
because the goroutine that is created via AfterFunc when the timer
fires already checks the invariants:

  // closeConnIfStillIdle closes the connection if it's still sitting idle.
  // This is what's called by the persistConn's idleTimer, and is run in its
  // own goroutine.
  func (pc *persistConn) closeConnIfStillIdle() {
     t := pc.t
     t.idleMu.Lock()
     defer t.idleMu.Unlock()
     if _, ok := t.idleLRU.m[pc]; !ok {
       // Not idle.
       return
     }

(note the "Not idle." part).

Tested by hand with the repro code from #25621. No more leaks.

Fixes #25621

Change-Id: Idf011a4cb1fcd01f55a5a6269e4c0ee5f4446786
Reviewed-on: https://go-review.googlesource.com/123315
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agotest: add test case that failed with gccgo
Ian Lance Taylor [Wed, 11 Jul 2018 17:51:13 +0000 (10:51 -0700)]
test: add test case that failed with gccgo

Updates #26335

Change-Id: Ibfb1e232a0c66fa699842c8908ae5ff0f5d2177d
Reviewed-on: https://go-review.googlesource.com/123316
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agoruntime: correct new thread stack for openbsd MAP_STACK
Joel Sing [Mon, 9 Jul 2018 18:54:37 +0000 (04:54 +1000)]
runtime: correct new thread stack for openbsd MAP_STACK

OpenBSD 6.4 will require the stack pointer to be pointing at an area that is
marked as MAP_STACK when entering and exiting syscalls. Adjust the stack pointer
used for a new thread such that it points within the stack, not at the top of
it (i.e. outside).

Fixes #26142

Change-Id: I905bd8e5be3dfc325392e7ac490fb56a7c71b3aa
Reviewed-on: https://go-review.googlesource.com/122735
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agomisc/wasm: free up memory on exit
Richard Musiol [Thu, 5 Jul 2018 13:32:31 +0000 (15:32 +0200)]
misc/wasm: free up memory on exit

Private fields of the Go class are not used any more after the program
has exited. Delete them to allow JavaScript's garbage collection to
clean up the WebAssembly instance.

Updates #26193.

Change-Id: I349784a49eaad0c22ceedd4f859df97132775537
Reviewed-on: https://go-review.googlesource.com/122296
Run-TryBot: Richard Musiol <neelance@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Paul Jolly <paul@myitcv.org.uk>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agocmd/pprof: disable readline UI support for TERM=dumb
Hana (Hyang-Ah) Kim [Tue, 10 Jul 2018 04:44:21 +0000 (00:44 -0400)]
cmd/pprof: disable readline UI support for TERM=dumb

In general, dumb terminal indicates terminal with limited capability.
It may provide no support for special character sequences, e.g., no
handling of ANSI escape sequences. Its input/output handling behavior
may deviate from what's described in termios or terminfo. E.g., in
the shell in emacs, even after successfully setting the terminal to
raw mode, the terminal behaves as if it's still operating in canonical
mode since emacs is doing input processing first.

Readline support can be broken in various ways in dumb terminal mode,
so we want to disable readline or advanced UI features. The easiest
way to detect dumb terminal is to check the environment variable "TERM".

Fixes #26254

Change-Id: I6b652eb555bc03b84405aae08b0b25d111fbb8b0
Reviewed-on: https://go-review.googlesource.com/122879
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agointernal/poll: don't take read lock in SetBlocking
Ian Lance Taylor [Tue, 10 Jul 2018 23:44:56 +0000 (16:44 -0700)]
internal/poll: don't take read lock in SetBlocking

Taking a read lock in SetBlocking could cause SetBlocking to block
waiting for a Read in another goroutine to complete. Since SetBlocking
is called by os.(*File).Fd, that could lead to deadlock if the
goroutine calling Fd is going to use it to unblock the Read.
Use an atomic store instead.

Updates #24481

Change-Id: I79413328e06ddf28b6d5b8af7a0e29d5b4e1e6ff
Reviewed-on: https://go-review.googlesource.com/123176
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agoos: clarify that Close cancels pending I/O
Ian Lance Taylor [Tue, 10 Jul 2018 13:47:44 +0000 (06:47 -0700)]
os: clarify that Close cancels pending I/O

Change-Id: I6be6818d951a999f916c2266a6753a5ce5144ee7
Reviewed-on: https://go-review.googlesource.com/122955
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
6 years agoruntime/cgo: Add initial NetBSD Thread Sanitizer support
Kamil Rytarowski [Sat, 30 Jun 2018 23:29:41 +0000 (23:29 +0000)]
runtime/cgo: Add initial NetBSD Thread Sanitizer support

Recognize NetBSD in:
 - go/internal/work/init.go
 - race.bash
 - runtime/race/race.go

Add __ps_strings symbol in runtime/cgo/netbsd.go as this is
used internally in the TSan library for NetBSD and used for
ReExec().

Tested on NetBSD/amd64 v. 8.99.12.

Around 98% tests are passing for the ./race.bash target.

Updates #19273

Change-Id: Ic0e48d2fb159a7868aab5e17156eeaca1225e513
GitHub-Last-Rev: d6e082707b9b18df1fe63f723666f4d2eb5e6cfe
GitHub-Pull-Request: golang/go#24322
Reviewed-on: https://go-review.googlesource.com/99835
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agonet/http: clarify when it's allowed to reuse a Request
Brad Fitzpatrick [Mon, 9 Jul 2018 22:05:34 +0000 (22:05 +0000)]
net/http: clarify when it's allowed to reuse a Request

Fixes #21780

Change-Id: Ic6fb6a536fff800a05be2d25309f72092604a785
Reviewed-on: https://go-review.googlesource.com/122817
Reviewed-by: Andrew Bonventre <andybons@golang.org>
6 years agotest: add order of evaluation test case that gccgo got wrong
Ian Lance Taylor [Tue, 10 Jul 2018 20:25:46 +0000 (13:25 -0700)]
test: add order of evaluation test case that gccgo got wrong

Updates #23188

Change-Id: Idc5567546d1c4c592f997a4cebbbf483b85331e0
Reviewed-on: https://go-review.googlesource.com/123115
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agodoc/go1.11: mention import path restriction
Russ Cox [Tue, 10 Jul 2018 21:14:14 +0000 (17:14 -0400)]
doc/go1.11: mention import path restriction

Change-Id: I18a6915614a317adb4da710a01268b574300f0e0
Reviewed-on: https://go-review.googlesource.com/123096
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
6 years agocmd/compile: call objabi.PathToPrefix when emitting abstract fn
Than McIntosh [Tue, 10 Jul 2018 16:23:31 +0000 (12:23 -0400)]
cmd/compile: call objabi.PathToPrefix when emitting abstract fn

When generating an abstract function DIE, call objabi.PathToPrefix on
the import path so as to be consistent with how the linker handles
import paths. This is intended to resolve another problem with DWARF
inline info generation in which there are multiple inconsistent
versions of an abstract function DIE for a function whose package path
is rewritten/canonicalized by objabi.PathToPrefix.

Fixes #26237

Change-Id: I4b64c090ae43a1ad87f47587a1a71f19bc5fc8e8
Reviewed-on: https://go-review.googlesource.com/123036
Run-TryBot: Than McIntosh <thanm@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agointernal/bytealg: specify argmaps for exported functions
Keith Randall [Mon, 9 Jul 2018 18:09:42 +0000 (11:09 -0700)]
internal/bytealg: specify argmaps for exported functions

Functions exported on behalf of other packages need to have their
argument stack maps specified explicitly.  They don't get an implicit
map because they are not in the local package, and if they get defer'd
they need argument maps.

Fixes #24419

Change-Id: I35b7d8b4a03d4770ba88699e1007cb3fcb5397a9
Reviewed-on: https://go-review.googlesource.com/122676
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
6 years agogo/types: ignore artificial cycles introduced via method declarations
Robert Griesemer [Fri, 29 Jun 2018 21:46:57 +0000 (14:46 -0700)]
go/types: ignore artificial cycles introduced via method declarations

At the moment, method declarations are type-checked together with
they receiver base types. This is a known problem (to be fixed early
for Go 1.12) but with the new cycle detection algorithm now also
introduced artifical type cycles.

This change pushes a special marker on the cycle path in those cases
so that these cycles can be ignored.

Fixes #26124.

Change-Id: I64da4ccc32d4ae293da48880c892154a1c6ac3fe
Reviewed-on: https://go-review.googlesource.com/121757
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
6 years agocmd/go, cmd/cgo: only set TERM=dumb when running the compiler
Ian Lance Taylor [Tue, 10 Jul 2018 14:19:17 +0000 (07:19 -0700)]
cmd/go, cmd/cgo: only set TERM=dumb when running the compiler

The clang compiler on some terminals will issue colored error
messages, which can confuse tools like cgo. To avoid this we used to
set TERM=dumb for all programs started by the go tool. However, that
confuses the pprof tool, which doesn't know whether to support fancy
editing and colors itself.

Instead, change the go tool and the cgo tool to set TERM=dumb where it
matters--when invoking the C compiler--rather than in all cases.

Updates #26254

Change-Id: I95174f961ac269a50a83f5f9d268219043cba968
Reviewed-on: https://go-review.googlesource.com/122975
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agotest/codegen: add more test cases for arm64
Ben Shi [Sun, 8 Jul 2018 10:02:47 +0000 (10:02 +0000)]
test/codegen: add more test cases for arm64

More test cases of combined load for arm64.

Change-Id: I7a9f4dcec6930f161cbded1f47dbf7fcef1db4f1
Reviewed-on: https://go-review.googlesource.com/122582
Run-TryBot: Ben Shi <powerman1st@163.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
6 years agocmd/go: add ImportMap to go list package struct
Russ Cox [Tue, 10 Jul 2018 03:00:35 +0000 (23:00 -0400)]
cmd/go: add ImportMap to go list package struct

Also populate Imports for test main with go list -test.

Update comment in internal/load/test.go about
p.Imports, p.Internal.RawImports, and p.Imports
being perfectly aligned. The first two are,
but the third is not, as evidenced by CL 111175.

Since p.Imports is not aligned, don't assume that anymore.

Fixes #25949.

Change-Id: Icbfbc881bc01d1e195a759648fbd1c978ddbc161
Reviewed-on: https://go-review.googlesource.com/122878
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
6 years agogo/types: correctly compute cycle length
Robert Griesemer [Fri, 29 Jun 2018 21:11:46 +0000 (14:11 -0700)]
go/types: correctly compute cycle length

The existing algorithm assumed that the length of a cycle was simply
the number of elements in the cycle slice starting at the start object.
However, we use a special "indir" indirection object to indicate
pointer and other indirections that break the inline placement of
types in composite types. These indirection objects don't exist as
true named type objects, so don't count them anymore.

This removes an unnecessary cycle error in one of the existing tests
(testdata/issues.src:100).

Also:
- added more tracing support (only active if tracing is enabled)
- better documentation in parts
- r/check.typ/check.typExpr/ in a few of places where we don't
  need to record a type indirection

Found while investigating #26124.

Change-Id: I45341743225d979a72af3fbecfa05012b32fab67
Reviewed-on: https://go-review.googlesource.com/121755
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
6 years agoruntime/trace: comment newlines to restore correct doc summary
Alberto Donizetti [Tue, 10 Jul 2018 08:02:02 +0000 (10:02 +0200)]
runtime/trace: comment newlines to restore correct doc summary

Fixes #26309

Change-Id: I0e0b61b885817e514aa46e299b00833f16e98b2a
Reviewed-on: https://go-review.googlesource.com/122898
Reviewed-by: Айнар Гарипов <gugl.zadolbal@gmail.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agonet/http: update bundled http2
Michael Fraenkel [Tue, 10 Jul 2018 03:24:18 +0000 (23:24 -0400)]
net/http: update bundled http2

Updates http2 to x/net/http2 git rev 292b43b for:

    http2: reject incoming HEADERS in Server on half-closed streams
    https://golang.org/cl/111677

Updates #25023

Change-Id: I479ae9b5b899fb0202e6301d02535fb6aeb4997a
Reviewed-on: https://go-review.googlesource.com/122877
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/cover: invoke go command to find packages
Russ Cox [Fri, 6 Jul 2018 05:29:47 +0000 (01:29 -0400)]
cmd/cover: invoke go command to find packages

cmd/cover has always assumed that package x/y/z can be
found in $GOPATH/src/x/y/z (roughly; by using go/build).
That won't be true for too much longer. Instead, run the
go command to find out where packages are.

This will make 'go tool cover' safe for use with Go modules
when they are in use in Go 1.11, and it continues to work
with the existing Go toolchains too.

An alternative would be to modify the cover profile format
to record file names directly, but that would require also
updating golang.org/x/tools/cover/profile and any tools
that use it, which seems not worth the trouble.
(That fork of the code does not contain any code to resolve
package names to directory locations, so it's unaffected.)

No new test here: cmd/go's TestCoverageFunc tests this code.

Fixes #25318 (when people use Go 1.11 instead of vgo).

Change-Id: I8769b15107aecf25f7aaf8692b724cf7d0f073d0
Reviewed-on: https://go-review.googlesource.com/122478
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Alan Donovan <adonovan@google.com>
6 years agocmd/go: add test for tests with no tests
Russ Cox [Mon, 9 Jul 2018 14:46:28 +0000 (10:46 -0400)]
cmd/go: add test for tests with no tests

CL 122518 rolled back an earlier CL that made "go test"
start running test binaries for packages with no *_test.go files.
Add a test as another roadblock to reintroducing that behavior
in the future.

For #26462.

Change-Id: I898103064efee8d6ae65bcf74f4dffc830eae7e9
Reviewed-on: https://go-review.googlesource.com/122595
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agocmd/compile: make sure alg functions are generated when we call them
Cherry Zhang [Fri, 6 Jul 2018 00:31:33 +0000 (20:31 -0400)]
cmd/compile: make sure alg functions are generated when we call them

When DWARF is disabled, some alg functions were not generated.
Make sure they are generated when we about to generate calls to
them.

Fixes #23546.

Change-Id: Iecfa0eea830e42ee92e55268167cefb1540980b2
Reviewed-on: https://go-review.googlesource.com/122403
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
6 years agotest: add test for gccgo bug #26248
Cherry Zhang [Mon, 9 Jul 2018 20:15:01 +0000 (16:15 -0400)]
test: add test for gccgo bug #26248

The fix is CL 122756.

Updates #26248.

Change-Id: Ic4250ab5d01da9f65d0bc033e2306343d9c87a99
Reviewed-on: https://go-review.googlesource.com/122757
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agonet/http: remove dead code noted in post-submit review of CL 81778
Brad Fitzpatrick [Mon, 9 Jul 2018 23:47:02 +0000 (23:47 +0000)]
net/http: remove dead code noted in post-submit review of CL 81778

Per comments in https://github.com/golang/go/issues/20239#issuecomment-402199944

Updates #20239
Updates #26303

Change-Id: Iddf34c0452bd30ca9111b951bca48d1e011bd85a
Reviewed-on: https://go-review.googlesource.com/122820
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agonet/http: update bundled http2
Brad Fitzpatrick [Mon, 9 Jul 2018 04:54:59 +0000 (04:54 +0000)]
net/http: update bundled http2

Adds tests for #122590 and updates x/net/http2 to git rev 6a8eb5e2b1 for:

     http2: call httptrace.ClientTrace.GetConn in Transport when needed
     https://golang.org/cl/122590

     http2: fire httptrace.ClientTrace.WroteHeaderField if set
     https://golang.org/cl/122816

     http2: compare Connection header value case-insensitively
     https://golang.org/cl/122588

This also includes the code for graceful shutdown, but it has no
public API surface via net/http, and should not affect any existing
code paths, as it's purely new stuff:

     http2: implement client initiated graceful shutdown
     https://golang.org/cl/30076

Fixes #19761
Fixes #23041

Change-Id: I5558a84591014554cad15ee3852a349ed717530f
Reviewed-on: https://go-review.googlesource.com/122591
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agoruntime: clarify SetFinalizer docs
Ian Lance Taylor [Sat, 7 Jul 2018 00:06:55 +0000 (17:06 -0700)]
runtime: clarify SetFinalizer docs

Fixes #24480

Change-Id: I7db721fb71a17f07472ec7f216478e7887435639
Reviewed-on: https://go-review.googlesource.com/122557
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agonet/http/httputil: don't panic in ReverseProxy unless running under a Server
Brad Fitzpatrick [Mon, 9 Jul 2018 22:29:00 +0000 (22:29 +0000)]
net/http/httputil: don't panic in ReverseProxy unless running under a Server

Prior to the fix to #23643, the ReverseProxy didn't panic with
ErrAbortHandler when the copy to a client failed.

During Go 1.11 beta testing, we found plenty of code using
ReverseProxy in tests that were unprepared for a panic.

Change the behavior to only panic when running under the http.Server
that'll handle the panic.

Updates #23643

Change-Id: Ic1fa8405fd54c858ce8c797cec79d006833a9f7d
Reviewed-on: https://go-review.googlesource.com/122819
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agocmd/go: add LDFLAGS to cache ID when using cgo
Ian Lance Taylor [Fri, 6 Jul 2018 20:59:06 +0000 (13:59 -0700)]
cmd/go: add LDFLAGS to cache ID when using cgo

The cgo tool records the value of the CGO_LDFLAGS environment variable
in the generated file, so that the linker can later read and use it.
Therefore, we must add CGO_LDFLAGS to the cache ID, as otherwise
changing CGO_LDFLAGS may cause a build result to be incorrectly read
from the cache, producing a different final program.

Change-Id: Ic89c1edc4069837451a36376710ca9b56fb87455
Reviewed-on: https://go-review.googlesource.com/122520
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agosyscall/js: improve panic messages
Richard Musiol [Thu, 5 Jul 2018 17:55:08 +0000 (19:55 +0200)]
syscall/js: improve panic messages

This commit adds the actual type to the panic message when calling
a method of Value on a Value with a bad type. It also adds better
panic messages to Value.Invoke and Value.Call.

Change-Id: Ic4b3aa29d3bef8e357be40cd07664ad602ffab12
Reviewed-on: https://go-review.googlesource.com/122376
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agosyscall/js: add Value.Type
Richard Musiol [Thu, 5 Jul 2018 17:30:00 +0000 (19:30 +0200)]
syscall/js: add Value.Type

This commits adds Value.Type(), which returns the JavaScript type of
a Value.

The implementation uses two previously unused bits of the NaN payload
to encode type information.

Change-Id: I568609569983791d50d35b8d80c44f3472203511
Reviewed-on: https://go-review.googlesource.com/122375
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agocmd/compile/internal/ssa: fix partsByVarOffset.Less method
Iskander Sharipov [Mon, 9 Jul 2018 20:08:22 +0000 (23:08 +0300)]
cmd/compile/internal/ssa: fix partsByVarOffset.Less method

Fix duplicated index in LHS and RHS of the < operator.

Found using https://go-critic.github.io/overview#dupSubExpr-ref

Change-Id: I9a5a40bbd436b32e8117579a01bc50afe3608c97
Reviewed-on: https://go-review.googlesource.com/122776
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/cgo: check function argument/return types for bad C pointer types
Keith Randall [Sat, 7 Jul 2018 04:38:31 +0000 (21:38 -0700)]
cmd/cgo: check function argument/return types for bad C pointer types

We need to determine whether arguments to and return values from C
functions are "bad" typedef'd pointer types which need to be uintptr
on the Go side.

The type of those arguments are not specified explicitly. As a result,
we never look through the C declarations for the GetTypeID functions
associated with that type, and never realize that they are bad.
However, in another function in the same package there might be an
explicit reference. Then we end up with the declaration being uintptr
in one file and *struct{...} in another file. Badness ensues.

Fix this by doing a 2-pass algorithm. In the first pass, we run as
normal, but record all the argument and result types we see. In the
second pass, we include those argument types also when reading the C
types.

Fixes #24161

Change-Id: I8d727e73a2fbc88cb9d9899f8719ae405f59f753
Reviewed-on: https://go-review.googlesource.com/122575
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agoapi: add some more API to go1.11.txt
Brad Fitzpatrick [Mon, 9 Jul 2018 21:19:11 +0000 (21:19 +0000)]
api: add some more API to go1.11.txt

Added since last update.

Change-Id: Ic5be0e2e379d422ef72b956a794d65613a0dd7be
Reviewed-on: https://go-review.googlesource.com/122815
Reviewed-by: Andrew Bonventre <andybons@golang.org>
6 years agohtml/template: ignore untyped nil arguments to default escapers
Ian Lance Taylor [Sat, 30 Jun 2018 00:31:37 +0000 (17:31 -0700)]
html/template: ignore untyped nil arguments to default escapers

CL 95215 changed text/template so that untyped nil arguments were no
longer ignored, but were instead passed to functions as expected.
This had an unexpected effect on html/template, where all data is
implicitly passed to functions: originally untyped nil arguments were
not passed and were thus effectively ignored, but after CL 95215 they
were passed and were printed, typically as an escaped version of "<nil>".

This CL restores some of the behavior of html/template by ignoring
untyped nil arguments passed implicitly to escaper functions.

While eliminating one change to html/template relative to earlier
releases, this unfortunately introduces a different one: originally
values of interface type with the value nil were printed as an escaped
version of "<nil>". With this CL they are ignored as though they were
untyped nil values. My judgement is that this is a less common case.
We'll see.

This CL adds some tests of typed and untyped nil values to
html/template and text/template to capture the current behavior.

Updates #18716
Fixes #25875

Change-Id: I5912983ca32b31ece29e929e72d503b54d7b0cac
Reviewed-on: https://go-review.googlesource.com/121815
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/link/internal/sym: add sizeof tests
Josh Bleecher Snyder [Mon, 9 Jul 2018 19:53:25 +0000 (12:53 -0700)]
cmd/link/internal/sym: add sizeof tests

CL 121916 showed that sym.Symbol matters for linker performance.
Prevent accidental regression.

Change-Id: I5fd998c91fdeef9e721bc3f6e30f775b81103e95
Reviewed-on: https://go-review.googlesource.com/122716
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agonet/http: add support for SameSite option in http.Cookie
Stephan Renatus [Mon, 27 Nov 2017 11:41:10 +0000 (12:41 +0100)]
net/http: add support for SameSite option in http.Cookie

The same-site cookie attribute prevents a cookie from being sent along with
cross-site requests. The main goal is mitigate the risk of cross-origin
information leakage and provides some protection against cross-site request
forgery attacks.

This change adds the option to http.Cookie so it can be stored and
passed to HTTP clients.

Spec: https://tools.ietf.org/html/draft-ietf-httpbis-cookie-same-site-00

Fixes #15867

Based on
https://github.com/reedloden/go/commit/eb31a0f063c80058bbb3abff4ca09b3565985500
by Reed Loden <reed@hackerone.com>

Change-Id: I98c8a9a92358b2f632990576879759e3aff38cff
Reviewed-on: https://go-review.googlesource.com/79919
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agovendor: update vendored x/net/http/httpproxy
Brad Fitzpatrick [Mon, 9 Jul 2018 17:30:10 +0000 (17:30 +0000)]
vendor: update vendored x/net/http/httpproxy

This updates x/net/http/httpproxy to git rev c21de06a for:

    http/httpproxy: support CIDR notation and ports with NO_PROXY
    https://golang.org/cl/115255

Fixes #16704

Change-Id: Ic96a0a36828779f88e68cd715bd076f36fd45e7a
Reviewed-on: https://go-review.googlesource.com/122655
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Fraenkel <michael.fraenkel@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agonet/http: comment handleReadError more, superficially use its argument
Brad Fitzpatrick [Mon, 9 Jul 2018 17:56:56 +0000 (17:56 +0000)]
net/http: comment handleReadError more, superficially use its argument

Fixes #24201

Change-Id: Ib970c4eeaa90489d014482276a7e5afa94a50741
Reviewed-on: https://go-review.googlesource.com/122675
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agonet/http: add Transport.MaxConnsPerHost knob
Mark Fischer [Sun, 22 Apr 2018 05:16:46 +0000 (01:16 -0400)]
net/http: add Transport.MaxConnsPerHost knob

Add field to http.Transport which limits connections per host,
including dial-in-progress, in-use and idle (keep-alive) connections.

For HTTP/2, this field only controls the number of dials in progress.

Fixes #13957

Change-Id: I7a5e045b4d4793c6b5b1a7191e1342cd7df78e6c
Reviewed-on: https://go-review.googlesource.com/71272
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/compile: ensure that loop condition is detected correctly
Keith Randall [Mon, 2 Jul 2018 22:21:35 +0000 (15:21 -0700)]
cmd/compile: ensure that loop condition is detected correctly

We need to make sure that the terminating comparison has the right
sense given the increment direction. If the increment is positive,
the terminating comparsion must be < or <=. If the increment is
negative, the terminating comparison must be > or >=.

Do a few cleanups,  like constant-folding entry==0, adding comments,
removing unused "exported" fields.

Fixes #26116

Change-Id: I14230ee8126054b750e2a1f2b18eb8f09873dbd5
Reviewed-on: https://go-review.googlesource.com/121940
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Heschi Kreinick <heschi@google.com>
6 years agoruntime: only run TestMemStats sanity tests once
Ian Lance Taylor [Sun, 8 Jul 2018 23:42:33 +0000 (16:42 -0700)]
runtime: only run TestMemStats sanity tests once

Fixes #22696

Change-Id: Ibe4628f71d64a2b36b655ea69710a925924b12a3
Reviewed-on: https://go-review.googlesource.com/122586
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
6 years agoregexp: revert "use sync.Pool to cache regexp.machine objects"
Russ Cox [Mon, 9 Jul 2018 14:55:36 +0000 (10:55 -0400)]
regexp: revert "use sync.Pool to cache regexp.machine objects"

Revert CL 101715.

The size of a sync.Pool scales linearly with GOMAXPROCS,
making it inappropriate to put a sync.Pool in any individually
allocated object, as the sync.Pool documentation explains.
The change also broke DeepEqual on regexps.

I have a cleaner way to do this with global sync.Pools but it's
too late in the cycle. Will revisit in Go 1.12. For now, revert.

Fixes #26219.

Change-Id: Ie632e709eb3caf489d85efceac0e4b130ec2019f
Reviewed-on: https://go-review.googlesource.com/122596
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

6 years agocmd/go: revert "output coverage report even if there are no test files"
Ian Lance Taylor [Fri, 6 Jul 2018 20:15:01 +0000 (13:15 -0700)]
cmd/go: revert "output coverage report even if there are no test files"

Original CL description:

    When using test -cover or -coverprofile the output for "no test files"
    is the same format as for "no tests to run".

Reverting because this CL changed cmd/go to build test binaries for
packages that have no tests, leading to extra work and confusion.

Updates #24570
Fixes #25789
Fixes #26157
Fixes #26242

Change-Id: Ibab1307d39dfaec0de9359d6d96706e3910c8efd
Reviewed-on: https://go-review.googlesource.com/122518
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
6 years agoruntime: scale timeout in TestStackGrowth
Ian Lance Taylor [Mon, 9 Jul 2018 00:02:48 +0000 (17:02 -0700)]
runtime: scale timeout in TestStackGrowth

Updates #19381

Change-Id: I62b8b0cd7170941af77281eb3aada3802623ec27
Reviewed-on: https://go-review.googlesource.com/122587
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agoruntime: skip TestG0StackOverflow on Android
Austin Clements [Sun, 8 Jul 2018 19:28:21 +0000 (15:28 -0400)]
runtime: skip TestG0StackOverflow on Android

This test is skipped on Linux and should be skipped on Android for the
same reason.

Change-Id: I753c4788d935bd58874554b455c0d5be2315b794
Reviewed-on: https://go-review.googlesource.com/122585
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
6 years agoruntime: fix TestAbort on non-x86 arches
Austin Clements [Sun, 8 Jul 2018 04:21:27 +0000 (00:21 -0400)]
runtime: fix TestAbort on non-x86 arches

CL 122515 tightened TestAbort to look for breakpoint exceptions and
not just general signal crashes, but this only applies on x86 arches.
On non-x86 arches we use a nil pointer dereference to abort, so the
test is now failing.

This CL re-loosens TestAbort on non-x86 arches to only expect a signal
traceback.

Should fix the build on linux/arm, linux/arm64, linux/ppc64, and
linux/s390x.

Change-Id: I1065341180ab5ab4da63b406c641dcde93c9490b
Reviewed-on: https://go-review.googlesource.com/122580
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
6 years agoruntime: fix TestAbort on Plan 9
David du Colombier [Sat, 7 Jul 2018 23:05:18 +0000 (01:05 +0200)]
runtime: fix TestAbort on Plan 9

Since CL 122515, TestAbort is failing on Plan 9
because there is no SIGTRAP signal on Plan 9,
but a note containing the "sys: breakpoint" string.

This change fixes the TestAbort test by handling
the Plan 9 case.

Fixes #26265.

Change-Id: I2fae00130bcee1cf946d8cc9d147a77f951be390
Reviewed-on: https://go-review.googlesource.com/122464
Run-TryBot: David du Colombier <0intro@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
6 years agoruntime: handle g0 stack overflows gracefully
Austin Clements [Mon, 25 Jun 2018 22:00:43 +0000 (18:00 -0400)]
runtime: handle g0 stack overflows gracefully

Currently, if the runtime overflows the g0 stack on Windows, it leads
to an infinite recursion:

1. Something overflows the g0 stack bounds and calls morestack.

2. morestack determines it's on the g0 stack and hence cannot grow the
stack, so it calls badmorestackg0 (which prints "fatal: morestack on
g0") followed by abort.

3. abort performs an INT $3, which turns into a Windows
_EXCEPTION_BREAKPOINT exception.

4. This enters the Windows sigtramp, which ensures we're on the g0
stack and calls exceptionhandler.

5. exceptionhandler has a stack check prologue, so it determines that
it's out of stack and calls morestack.

6. goto 2

Fix this by making the exception handler avoid stack checks until it
has ruled out an abort and by blowing away the stack bounds in
lastcontinuehandler before we print the final fatal traceback (which
itself involves a lot of stack bounds checks).

Fixes #21382.

Change-Id: Ie66e91f708e18d131d97f22b43f9ac26f3aece5a
Reviewed-on: https://go-review.googlesource.com/120857
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>