Forgot to mail last patch set before committing, repair that.
Change-Id: I1ef72d0d7df56e89369e6fb4d6e5690f254e6aa8
Reviewed-on: https://go-review.googlesource.com/c/go/+/286912
Trust: David Chase <drchase@google.com>
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
It's no longer conditional.
Change-Id: I697bb0e9ffe9644ec4d2766f7e8be8b82d3b0638
Reviewed-on: https://go-review.googlesource.com/c/go/+/286013
Trust: David Chase <drchase@google.com>
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
The recent refactoring of SelectorExpr code to helpers broke the
handling of MethodExprs when there is an embedded field involved (e.g.
test/method7.go, line 48). If there is an embedded field involved, the
node op seen in DotMethod() is an ODOT rather than an OTYPE. Also, the
receiver type of the result should be the original type, but the new
code was using the last type after following the embedding path.
Change-Id: I13f7ea6448b03d3e8f974103ee3a027219ca8388
Reviewed-on: https://go-review.googlesource.com/c/go/+/286176
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Dan Scales <danscales@google.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Convert the syscall package on openbsd/amd64 to use libc rather than performing
direct system calls.
Updates #36435
Change-Id: Ieb5926a91ed34f7c722e3667004ec484c86804ef
Reviewed-on: https://go-review.googlesource.com/c/go/+/270380
Trust: Joel Sing <joel@sing.id.au>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
The test is fine and probably was excluded by mistake.
Change-Id: I98508e603afe01a781ad7c8638830514aa75939c
Reviewed-on: https://go-review.googlesource.com/c/go/+/286732
Trust: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Shortcut matching code if both operands have the same representation.
Change-Id: I9433455acb5b9d0b88d3c81eb1b3b0ae258193e7
Reviewed-on: https://go-review.googlesource.com/c/go/+/286654
Trust: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
go/constant represents a Float constant either as a rational number
(if numerator and denominator are small enough), or, as a "catch-all",
as a arbitrary-precision floating-point number.
This CL cleans up some of these transitions by factoring out more
of the decision logic and documents the rationale between the state
transitions better.
This CL also simplifies some unrelated code that was overly complex.
Updates #43908.
Change-Id: Iccdd2d6b7fb7ed13d68ed5e6d992d1bc56a065bb
Reviewed-on: https://go-review.googlesource.com/c/go/+/286572
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Currently, types2 sometimes produces constant.Values with a Kind
different than the untyped constant type's Is{Integer,Float,Complex}
info, which irgen expects to always match.
While we mull how best to proceed in #43891, this CL adapts irgen to
types2's current behavior. In particular, fixedbugs/issue11945.go now
passes with -G=3.
Updates #43891.
Change-Id: I24823a32ff49af6045a032d3903dbb55cbec6bef
Reviewed-on: https://go-review.googlesource.com/c/go/+/286652
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Trust: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
NewFile requires the file descriptor to be either closed
through the returned File instance, or to stay valid at least
until the finalizer runs during garbage collection.
These requirements are easily violated when file descriptors
are closed via unix.Close, or when the *File returned by
NewFile is garbage collected while the underlying file descriptor is
still in use.
This commit adds further documentation for NewFile and Fd, making it
explicit that using naked file descriptors is subject to constraints
due to garbage collection of File objects.
Fixes#43863
Change-Id: I49ea1f0054eb2d2a72b616450c8e83476f4d07fb
GitHub-Last-Rev: 180d0130ae
GitHub-Pull-Request: golang/go#43867
Reviewed-on: https://go-review.googlesource.com/c/go/+/286032
Trust: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
This merge involved two merge conflicts:
1. walk's ascompatee code has been substantially refactored on
dev.regabi, so CL 285633 is ported to the new style.
2. The os.TestDirFS workaround added in CL 286213 can be removed now
that #42637 has been fixed by CL 285720.
Conflicts:
- src/cmd/compile/internal/gc/walk.go
- src/os/os_test.go
Merge List:
+ 2021-01-25 bf0f7c9d78 doc/go1.16: mention os.DirFS in os section
+ 2021-01-25 deaf29a8a8 cmd/compile: fix order-of-assignment issue w/ defers
+ 2021-01-25 ad2ca26a52 doc/go1.16: mention os.DirEntry and types moved from os to io/fs
+ 2021-01-25 a51921fa5b doc/go1.16: mention new testing/iotest functions
+ 2021-01-25 e6b6d107f7 doc/go1.16: mention deprecation of io/ioutil
+ 2021-01-25 96a276363b doc/go1.16: mention go/build changes
+ 2021-01-25 3d85c69a0b html/template: revert "avoid race when escaping updates template"
+ 2021-01-25 54514c6b28 cmd/go: fix TestScript/cgo_path, cgo_path_space when CC set
+ 2021-01-25 6de8443f3b doc/asm: add a section on go_asm.h, clean up go_tls.h section
+ 2021-01-25 54b251f542 lib/time, time/tzdata: update tzdata to 2021a
+ 2021-01-25 ff82cc971a os: force consistent mtime before running fstest on directory on Windows
+ 2021-01-25 044f937a73 doc/go1.16: fix WalkDir and Walk links
+ 2021-01-23 b634f5d97a doc/go1.16: add crypto/x509 memory optimization
+ 2021-01-23 9897655c61 doc/go1.16: reword ambiguously parsable sentence
+ 2021-01-23 cd99385ff4 cmd/internal/obj/arm64: fix VMOVQ instruction encoding error
+ 2021-01-23 66ee8b158f runtime: restore cgo_import_dynamic for libc.so on openbsd
+ 2021-01-22 25c39e4fb5 io/ioutil: fix example test for WriteFile to allow it to run in the playground
+ 2021-01-22 eb21b31e48 runtime: define dummy msanmove
+ 2021-01-22 3a778ff50f runtime: check for g0 stack last in signal handler
+ 2021-01-22 a2cef9b544 cmd/go: don't lookup the path for CC when invoking cgo
Change-Id: I651949f9eb18b57e3c996c4f3b2b3bf458bc5d97
For #42026
Change-Id: I51e3ce9d3a4729cfac44bd3ff3f3ec80dcd5abb5
Reviewed-on: https://go-review.googlesource.com/c/go/+/285376
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Dmitri Shuralyov <dmitshur@golang.org>
The race reported in issue #41167 was detected only because the
ReadWriter used in test code happened to be a bytes.Buffer whose
Read and Write operate (unsafely) on shared state. This is not the
case in any realistic scenario where the FastCGI protocol is spoken
over sockets or pairs of pipes.
Since tests that use nopWriteCloser don't care about any output
generate by child.Serve(), we change nopWriteCloser to provide
a dummy Write method.
Remove the locking added in CL 252417, since it causes a deadlock
during write as reported in #43901. The race in tests no longer
happens thanks to the aforementioned change to nopWriteCloser.
Fixes#43901.
Updates #41167.
Change-Id: I8cf31088a71253c34056698f8e2ad0bee9fcf6c6
GitHub-Last-Rev: b06d8377fd
GitHub-Pull-Request: golang/go#43027
Reviewed-on: https://go-review.googlesource.com/c/go/+/275692
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Trust: Dmitri Shuralyov <dmitshur@golang.org>
CL 261677 fixed a logic issue in walk's alias detection, where it was
checking the RHS expression instead of the LHS expression when trying
to determine the kind of assignment. However, correcting this exposed
a latent issue with assigning to result parameters in functions with
defers, where an assignment could become visible earlier than intended
if a later expression could panic.
Fixes#43835.
Change-Id: I061ced125e3896e26d65f45b28c99db2c8a74a8c
Reviewed-on: https://go-review.googlesource.com/c/go/+/285633
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Cuong Manh Le <cuong.manhle.vn@gmail.com>
Reviewed-by: Keith Randall <khr@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Floating-point constants are represented as rational numbers when
possible (i.e., when numerators and denominators are not too large).
If we convert to floats when not necessary, we risk losing precision.
This is the minimal fix for the specific issue, but it's too aggressive:
If the numbers are too large, we still want to convert to floats.
Will address in a separate CL that also does a few related cleanups.
Fixes#43908.
Change-Id: Id575e34fa18361a347c43701cfb4dd7221997f66
Reviewed-on: https://go-review.googlesource.com/c/go/+/286552
Trust: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Adds some extra validation that the frontend is setting flags like
Addrtaken correctly.
Change-Id: Iffde83e32ba1c4c917ab8cb3fe410a4f623cf635
Reviewed-on: https://go-review.googlesource.com/c/go/+/286434
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
This reverts CLs 274450 and 279492, except for the new tests.
The new race test is changed to skip, as it now fails.
We can try again for 1.17.
Original CL descriptions:
html/template: attach functions to namespace
The text/template functions are stored in a data structure shared by
all related templates, so do the same with the original, unwrapped,
functions on the html/template side.
html/template: avoid race when escaping updates template
For #39807Fixes#43855
Change-Id: I2ce91321ada06ea496a982aefe170eb5af9ba847
Reviewed-on: https://go-review.googlesource.com/c/go/+/285957
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Emmanuel Odeke <emmanuel@orijtech.com>
These tests failed if CC was set to a path containing a separator
during make.bash. They now set CC explicitly.
Fixes#43897
Change-Id: Ic6e7f192fcb363f0ac9f45b329113255453bf76f
Reviewed-on: https://go-review.googlesource.com/c/go/+/286292
Run-TryBot: Jay Conrod <jayconrod@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Jay Conrod <jayconrod@google.com>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
This CL expands internal-abi.md to cover Go's memory layout rules and
then uses this to specify the calling convention more precisely.
Change-Id: Ifeef9e49d9ccc8c7333dec81bdd47b511b028469
Reviewed-on: https://go-review.googlesource.com/c/go/+/286073
Trust: Austin Clements <austin@google.com>
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
This adds a document specifying the internal ABI (specifically the
calling convention). This document lives in the Go tree (rather than
the proposal repository) because the intent is for it to track the
reality in the compiler.
Updates #40724.
Change-Id: I583190080cd7d8cb1084f616fd1384d0f1f25725
Reviewed-on: https://go-review.googlesource.com/c/go/+/285292
Trust: Austin Clements <austin@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Currently the only mention of go_asm.h is buried in a confusing
section about the runtime-specific go_tls.h header. We actually want
people to use go_asm.h, so this CL adds a section with a proper
discussion of this header. As part of this, we remove the discussion
of go_asm.h from the go_tls.h section and clean up what remains.
I stumbled on this when working on the internal ABI specification. I
wanted to refer to stable documentation on how to access struct fields
from assembly and found there was none.
Change-Id: I0d53741e7685e65794611939e76285f7c82e1d65
Reviewed-on: https://go-review.googlesource.com/c/go/+/286052
Trust: Austin Clements <austin@google.com>
Reviewed-by: Jeremy Faller <jeremy@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
This allows more precision and matches types2's behavior.
For backwards compatibility with gcimporter, for now we still need to
write out declared constants as limited-precision floating-point
values. To ensure consistent behavior of constant arithmetic whether
it spans package boundaries or not, we include the full-precision
rational representation in the compiler's extension section of the
export data.
Also, this CL simply uses the math/big.Rat.String text representation
as the encoding. This is inefficient, but because it's only in the
compiler's extension section, we can easily revisit this in the
future.
Declaring exported untyped float and complex constants isn't very
common anyway. Within the standard library, only package math declares
any at all, containing just 15. And those 15 are only imported a total
of 12 times elsewhere in the standard library.
Change-Id: I85ea23ab712e93fd3b68e52d60cbedce9be696a0
Reviewed-on: https://go-review.googlesource.com/c/go/+/286215
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
This test is causing nearly every trybot run on dev.regabi and
dev.typeparams to fail. It's already a release blocker for Go 1.16, so
the failures on the development branches is entirely noise; and
because it causes the trybots to short-circuit, it risks masking
actual Windows-specific failures.
This CL disables the test until a proper solution is decided upon and
implemented for Go 1.16.
Updates #42637.
Change-Id: Ibc85edaed591f1c125cf0b210867aa89d2b0a4b6
Reviewed-on: https://go-review.googlesource.com/c/go/+/286213
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Fixes "GO_GCFLAGS=-G=3 go run run.go -- blank.go interface/fail.go".
Change-Id: I669ab06ae29366ce96e2948c89a5c1620afd53db
Reviewed-on: https://go-review.googlesource.com/c/go/+/286214
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Change-Id: I170e4f9c5a1db4bad02a5fe4bddc65d4c75f51e8
Reviewed-on: https://go-review.googlesource.com/c/go/+/286232
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
The race detector on ppc64le corrupts command-line arguments lists if
they contain an empty string, and cmd/go often generates compiler
argument lists containing `-D ""`. Since this is equivalent to not
specifying the `-D` flag at all, just do that. This allows using a
race-detector-enabled cmd/compile on ppc64le.
Updates #43883.
Change-Id: Ifac5cd9a44932129438b9b0b3ecc6101ad3716b2
Reviewed-on: https://go-review.googlesource.com/c/go/+/286173
Trust: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
Reviewed-by: Jay Conrod <jayconrod@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Discovered difficluties posed by earlier design, these modifications
should work better. Updated tests, also added some helper functions
for use in call lowering.
Change-Id: I459f0f71ad8a6730c571244925c3f395e1df28de
Reviewed-on: https://go-review.googlesource.com/c/go/+/285392
Trust: David Chase <drchase@google.com>
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
FindFileNext sometimes returns a different mtime than looking at the
file directly, because the MFT on NTFS is written to lazily. In order to
keep these in sync, we use GetFileInformationByHandle to get the actual
mtime, and then write it back to the file explicitly.
Fixes#42637.
Change-Id: I774016d3ac55d0dc9b0f9c1b681516c33ba0d28a
Reviewed-on: https://go-review.googlesource.com/c/go/+/285720
Run-TryBot: Jason A. Donenfeld <Jason@zx2c4.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
Trust: Jason A. Donenfeld <Jason@zx2c4.com>
Eager re-sync-branch to keep Git history reasonably accurate, since
Git lacks a better way of encoding partial merges like CL 286172.
Conflicts:
- src/cmd/compile/internal/inline/inl.go
- src/cmd/compile/internal/noder/import.go
- src/cmd/compile/internal/noder/noder.go
Merge List:
+ 2021-01-25 063c72f06d [dev.regabi] cmd/compile: backport changes from dev.typeparams (9456804)
+ 2021-01-23 d05d6fab32 [dev.regabi] cmd/compile: replace ir.Name map with ir.NameSet for SSA 2
+ 2021-01-23 48badc5fa8 [dev.regabi] cmd/compile: fix escape analysis problem with closures
+ 2021-01-23 51e1819a8d [dev.regabi] cmd/compile: scan body of closure in tooHairy to check for disallowed nodes
Change-Id: I48c0435f7aaf56f4aec26518a7459e9d95a51e9c
This CL backports a bunch of changes that landed on dev.typeparams,
but are not dependent on types2 or generics. By backporting, we reduce
the divergence between development branches, hopefully improving test
coverage and reducing risk of merge conflicts.
Updates #43866.
Change-Id: I382510855c9b5fac52b17066e44a00bd07fe86f5
Reviewed-on: https://go-review.googlesource.com/c/go/+/286172
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
The VMOVQ instruction moves a 128-bit constant into a V register, as 128-bit
constant can't be loaded into a register directly, we split it into two 64-bit
constants and load it from constant pool. Currently we add the 128-bit constant
to literal pool by calling the 'addpool' function twice, this is not the right
way because it doesn't guarantee the two DWORD instructions are consecutive,
and the second call of addpool will overwrite the p.Pool field,resulting in a
wrong PC-relative offset value of the Prog.
This CL renames the flag LFROM3 to LFROM128, and adds a new function addpool128
to add a 128-bit constant to the literal pool.
Change-Id: I616f043c99a9a18a663f8768842cc980de2e6f79
Reviewed-on: https://go-review.googlesource.com/c/go/+/282334
Reviewed-by: eric fang <eric.fang@arm.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: eric fang <eric.fang@arm.com>
Trust: eric fang <eric.fang@arm.com>
Same as CL 284897, the last one.
Passes toolstash -cmp.
Updates #43819
Change-Id: I0bd8958b3717fb58a5a6576f1819a85f33b76e2d
Reviewed-on: https://go-review.googlesource.com/c/go/+/285913
Run-TryBot: Baokun Lee <bk@golangcn.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Trust: Baokun Lee <bk@golangcn.org>
Change-Id: I9591f7aeab0448aca661560bf3064e057b48293e
Reviewed-on: https://go-review.googlesource.com/c/go/+/286012
Trust: Robert Griesemer <gri@golang.org>
Trust: Dan Scales <danscales@google.com>
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
This was removed in change 285692, however we need to explicitly pull libc.so
in when libpthread.so is being used. The current code works on openbsd/amd64
since we pull libc.so in via runtime/sys_openbsd2.go, however openbsd/arm64
does not do this currently.
Change-Id: Ibe93d936a22e69e2fe12620f6d27ccca7a91dba5
Reviewed-on: https://go-review.googlesource.com/c/go/+/285912
Trust: Joel Sing <joel@sing.id.au>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
In reflect.methodWrapper, we call escape analysis without including the
full batch of dependent functions, including the closure functions.
Because of this, we haven't created locations for the params/local
variables of a closure when we are processing a function that
inlines that closure. (Whereas in the normal compilation of the
function, we do call with the full batch.) To deal with this, I am
creating locations for the params/local variables of a closure when
needed.
Without this fix, the new test closure6.go would fail.
Updates #43818
Change-Id: I5f91cfb6f35efe2937ef88cbcc468e403e0da9ad
Reviewed-on: https://go-review.googlesource.com/c/go/+/285677
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Dan Scales <danscales@google.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
We can always re-introduce it if we decide to make use of it.
Change-Id: Ia939fdae978568edc58e21d1af732c6137744aab
Reviewed-on: https://go-review.googlesource.com/c/go/+/285678
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Note: This invalidates the implementation of MethodSet further (it
also has not been updated to accomodate for type parameters). But
types2 doesn't make use of it. We should remove it.
Change-Id: Ia2601bdd59b3f3ee0b72bc2512153c42bf5053b5
Reviewed-on: https://go-review.googlesource.com/c/go/+/285994
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
This CL refactors the SelectorExpr-handling code added in CL 285373
into helper functions that can eventually be reused by iimport.
Change-Id: I15b4a96c242f63cb370d7492ed08168550724f47
Reviewed-on: https://go-review.googlesource.com/c/go/+/285953
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
