TestTicker is sensitive to overloaded or slow systems, where a 20ms
ticker running for 10 ticks has a total run time out of the range
[110ms, 290ms]. To counter this flakiness, it tries five times to
get a successful result. This is insufficient--an overloaded test
machine can introduce more than 100ms of delay across the test.
Reduce the five attempts to two, but use a 1s ticker for 8 ticks
in the second attempt.
Updates #46474.
Updates #35692.
Change-Id: Ibd5187b00ccceeb981b652f2af9a1c3766357b78
Reviewed-on: https://go-review.googlesource.com/c/go/+/339892
Trust: Damien Neil <dneil@google.com>
Run-TryBot: Damien Neil <dneil@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Use typecheck.Conv; it does the same thing.
Change-Id: I6bf27b8e929f8016ffc8350931c8320e36584a3d
Reviewed-on: https://go-review.googlesource.com/c/go/+/354350
Trust: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
The tree has package main files scattered around
in it for the purposes of running go generate.
They're all marked "// +build ignore",
which gets special handling in TestDependencies.
It would be nice to be able to use other build tags,
such as "generate", as suggested by the go generate
design doc. Plus the build tag syntax is changing.
This change skips all "package main" files.
By definition these aren't importable,
so they can't contribute to the dependency tree.
We can't quite eliminate the "// +build ignore"
check, as it is used by packages runtime and syscall.
But it's still a step in the right direction.
Change-Id: Ib9449acfdba75f570b87a4200afe944910d76222
Reviewed-on: https://go-review.googlesource.com/c/go/+/339592
Trust: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Accept a uint32 instead of a uintptr to make call sites simpler.
Do less work in the common case in which len(textsectmap) == 1.
Change-Id: Idd6cdc3fdad7a9356864c83790463b5d3000171b
Reviewed-on: https://go-review.googlesource.com/c/go/+/354132
Trust: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Somehow CL 353390 managed to update the vendored code to d2e5035098b3,
but not the go.mod and other version references. Fix.
Change-Id: Ic265c10f7bd7ec982671b46c5d9ae50636a13309
Reviewed-on: https://go-review.googlesource.com/c/go/+/354391
Trust: Damien Neil <dneil@google.com>
Run-TryBot: Damien Neil <dneil@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
The encoding/binary little- and big-endian load and store routines are
frequently used in performance sensitive code. They look fairly complex
to the inliner. Though the routines themselves can be inlined,
code using them typically cannot be.
Yet they typically compile down to an instruction or two
on architectures that support merging such loads.
This change teaches the inliner to treat calls to these methods as cheap,
so that code using them will be more inlineable.
It'd be better to teach the inliner that this pattern of code is cheap,
rather than these particular methods. However, that is difficult to do
robustly when working with the IR representation. And the broader project
of which that would be a part, namely to model the rest of the compiler
in the inliner, is probably a non-starter. By way of contrast, imperfect
though it is, this change is an easy, cheap, and useful heuristic.
If/when we base inlining decisions on more accurate information obtained
later in the compilation process, or on PGO/FGO, we can remove this
and other such heuristics.
Newly inlineable functions in the standard library:
crypto/cipher.gcmInc32
crypto/sha512.appendUint64
crypto/md5.appendUint64
crypto/sha1.appendUint64
crypto/sha256.appendUint64
vendor/golang.org/x/crypto/poly1305.initialize
encoding/gob.(*encoderState).encodeUint
vendor/golang.org/x/text/unicode/norm.buildRecompMap
net/http.(*http2SettingsFrame).Setting
net/http.http2parseGoAwayFrame
net/http.http2parseWindowUpdateFrame
Benchmark impact for encoding/gob (the only package I measured):
name old time/op new time/op delta
EndToEndPipe-8 2.25µs ± 1% 2.21µs ± 3% -1.79% (p=0.000 n=28+27)
EndToEndByteBuffer-8 93.3ns ± 5% 94.2ns ± 5% ~ (p=0.174 n=30+30)
EndToEndSliceByteBuffer-8 10.5µs ± 1% 10.6µs ± 1% +0.87% (p=0.000 n=30+30)
EncodeComplex128Slice-8 1.81µs ± 0% 1.75µs ± 1% -3.23% (p=0.000 n=28+30)
EncodeFloat64Slice-8 900ns ± 1% 847ns ± 0% -5.91% (p=0.000 n=29+28)
EncodeInt32Slice-8 1.02µs ± 0% 0.90µs ± 0% -11.82% (p=0.000 n=28+26)
EncodeStringSlice-8 1.16µs ± 1% 1.04µs ± 1% -10.20% (p=0.000 n=29+26)
EncodeInterfaceSlice-8 28.7µs ± 3% 29.2µs ± 6% ~ (p=0.067 n=29+30)
DecodeComplex128Slice-8 7.98µs ± 1% 7.96µs ± 1% -0.27% (p=0.017 n=30+30)
DecodeFloat64Slice-8 4.33µs ± 1% 4.34µs ± 1% +0.24% (p=0.022 n=30+29)
DecodeInt32Slice-8 4.18µs ± 1% 4.18µs ± 0% ~ (p=0.074 n=30+28)
DecodeStringSlice-8 13.2µs ± 1% 13.1µs ± 1% -0.64% (p=0.000 n=28+28)
DecodeStringsSlice-8 31.9µs ± 1% 31.8µs ± 1% -0.34% (p=0.001 n=30+30)
DecodeBytesSlice-8 8.88µs ± 1% 8.84µs ± 1% -0.48% (p=0.000 n=30+30)
DecodeInterfaceSlice-8 64.1µs ± 1% 64.2µs ± 1% ~ (p=0.173 n=30+28)
DecodeMap-8 74.3µs ± 0% 74.2µs ± 0% ~ (p=0.131 n=29+30)
Fixes#42958
Change-Id: Ie048b8976fb403d8bcc72ac6bde4b33e133e2a47
Reviewed-on: https://go-review.googlesource.com/c/go/+/349931
Trust: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
They're only used in a single place.
Instead of calculating the end every time,
calculate it in the linker.
It'd be nice to recalculate baseaddr-vaddr,
but that generates relocations that are too large.
While we're here, remove some pointless uintptr -> uintptr conversions.
Change-Id: I91758f9bff11b365bc3a63fee172dbdc3d90b966
Reviewed-on: https://go-review.googlesource.com/c/go/+/354089
Trust: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Tests that require instrumentation must be in
a test script which only runs for architectures
where coverage is supported.
Change-Id: Ia56b1cef1071f848c9dce8c87597207937fdb6c9
Reviewed-on: https://go-review.googlesource.com/c/go/+/354372
Trust: Katie Hockman <katie@golang.org>
Run-TryBot: Katie Hockman <katie@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
This is a clean port of CL 354329 from types2 to go/types.
For #48819.
Change-Id: I9efdcdbfa6432f3cee64d924a4c67ecc6793cf86
Reviewed-on: https://go-review.googlesource.com/c/go/+/354349
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
This CL reverses the change in CL 284254 (which was ported
to types2) which originated in CL 240901 to address a crash
in a test created by a fuzzer (that crash appears to be
avoided in other ways, now).
This exposed another bug in typeset.go where we don't look
for the underlying type when testing if a type is an interface
or not. Fixed that as well.
Adjusted a test case that now doesn't report an error anymore
(which is good).
Fixes#48819.
Change-Id: I611d68e053d6b8a2f7176d0cd5a44da2df28ad21
Reviewed-on: https://go-review.googlesource.com/c/go/+/354329
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
The type set notation has been accepted a while ago.
We're not going back to supporting the original
type list notation. Remove support for it in the
parser and type checker.
Change-Id: I860651f80b89fa43a3a5a2a02cf823ec0dae583c
Reviewed-on: https://go-review.googlesource.com/c/go/+/354131
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
By making off an int64 at the beginning,
the code gets a lot simpler. Cleanup only.
Change-Id: I4a2519f953e2f71081a4ff3032f8fd6da06c7e24
Reviewed-on: https://go-review.googlesource.com/c/go/+/354138
Trust: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
With the removal of SSE2 runtime detection made in
golang.org/cl/344350 we can remove this mechanism as there
are no required features anymore.
For making sure CPUs running a go program support all
the minimal hardware requirements the go runtime should
do feature checks early in the runtime initialization
before it is likely any compiler emitted but unsupported
instructions are used. This is already the case for e.g.
checking MMX support on 386 arch targets.
Change-Id: If7b1cb6f43233841e917d37a18314d06a334a734
Reviewed-on: https://go-review.googlesource.com/c/go/+/354209
Trust: Martin Möhrmann <martin@golang.org>
Run-TryBot: Martin Möhrmann <martin@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Tobias Klauser <tobias.klauser@gmail.com>
Reviewed-by: Keith Randall <khr@golang.org>
As we use relative addressing for text symbols in functab, it is
important that the offsets we computed stay unchanged by the
external linker, i.e. all symbols in Textp should not be removed
by the external linker. Most of them are actually referenced (our
deadcode pass ensures that), except go.buildid which is generated
late and not used by the program. Keep it alive.
Should fix AIX builder.
Change-Id: Ibc4a8951be997b9d8d870d75c54754977d9b8333
Reviewed-on: https://go-review.googlesource.com/c/go/+/354369
Trust: Cherry Mui <cherryyz@google.com>
Trust: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Cherry Mui <cherryyz@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org>
In GOPATH mode, source files may import other packages using relative
(“local”) paths. In module mode, relative imports are never allowed:
import paths must always be fully specified.
When local imports are allowed, we pass a local-import prefix to the
compiler using the '-D' flag. That could theoretically change the
compiler's output, so it must be included in the cache key even when
-trimpath is set. (TODO: when -trimpath is set, the local-import
prefix ought to be trimmed anyway, so it still shouldn't matter.)
However, when local imports are disallowed, we should not pass the
local-import prefix and it should not affect cmd/go's cache key or the
final build ID of any artifact.
For #48557
Change-Id: I2d627d67d13e5da2cac6d411cd4e2d87e510876c
Reviewed-on: https://go-review.googlesource.com/c/go/+/352810
Trust: Bryan C. Mills <bcmills@google.com>
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Jay Conrod <jayconrod@google.com>
This change also fixes a bug with calculating the
total interesting count. When fuzzing with an empty
corpus, the fuzzing engine adds an starting corpus
value in that run in order to start fuzzing. That
meant that the interesting total count was off by one:
it would start at 1, even though the cache was empty.
Added some tests for this as well.
Fixes#48787
Change-Id: I47acf96f0a0797214ebb24a95366d8460bf303bc
Reviewed-on: https://go-review.googlesource.com/c/go/+/354150
Trust: Katie Hockman <katie@golang.org>
Run-TryBot: Katie Hockman <katie@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Jay Conrod <jayconrod@google.com>
Reviewed-by: Roland Shoemaker <roland@golang.org>
Many uses of Index/IndexByte/IndexRune/Split/SplitN
can be written more clearly using the new Cut functions.
Do that. Also rewrite to other functions if that's clearer.
For #46336.
Change-Id: I68d024716ace41a57a8bf74455c62279bde0f448
Reviewed-on: https://go-review.googlesource.com/c/go/+/351711
Trust: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Using Cut is a clearer way to write the vast majority (>70%)
of existing code that calls Index, IndexByte, IndexRune, and SplitN.
There is more discussion on https://golang.org/issue/46336.
Fixes#46336.
Change-Id: Ia418ed7c3706c65bf61e1b2c5baf534cb783e4d3
Reviewed-on: https://go-review.googlesource.com/c/go/+/351710
Trust: Russ Cox <rsc@golang.org>
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
This CL adds late expanded memequal(x, const, sz) inlining for 2, 4, 8
bytes size. This PoC is using the same method as CL 248404.
This optimization fires about 100 times in Go compiler (1675 occurrences
reduced to 1574, so -6%).
Also, added unit-tests to codegen/comparisions.go file.
Updates #37275
Change-Id: Ia52808d573cb706d1da8166c5746ede26f46c5da
Reviewed-on: https://go-review.googlesource.com/c/go/+/328291
Reviewed-by: Cherry Mui <cherryyz@google.com>
Run-TryBot: Cherry Mui <cherryyz@google.com>
Trust: David Chase <drchase@google.com>
This adds a rule to avoid the zero extension after an AtomicLoad8
or AtomicLoad32 since the atomic load has already filled it with
zeros. This eliminates an instruction in a high use block in findObject
and the AtomicLoad8 appears many times within runtime.
Change-Id: I7e684bf73d3812110bd371e05b1aa44fa235fc9b
Reviewed-on: https://go-review.googlesource.com/c/go/+/354029
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Trust: Lynn Boger <laboger@linux.vnet.ibm.com>
RFC 2616 permits multiple digits in the major and minor numbers of an
HTTP version:
https://datatracker.ietf.org/doc/html/rfc2616#section-3.1
RFC 7230 obsoletes 2616 and tightens the specification to permit only a
single digit in the major and minor number:
https://datatracker.ietf.org/doc/html/rfc7230#section-2.6
Use the stricter definition.
Also fix a bug which caused version numbers with a leading "+" to
be accepted (e.g., "HTTP/1.+1".)
Fixes#46587.
Change-Id: Ic5923bb858e5ac402cfde486fba2c075e221553d
Reviewed-on: https://go-review.googlesource.com/c/go/+/325874
Trust: Damien Neil <dneil@google.com>
Trust: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Damien Neil <dneil@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Change-Id: Ied57fb6e71d56618d46aeb36a37a709e08b4346e
Reviewed-on: https://go-review.googlesource.com/c/go/+/354136
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
The slowest thing that can happen in funcdata is a cache miss
on moduledata.gofunc. Move that memory load earlier.
Also, for better ergonomics when working on this code,
do more calculations as uintptrs.
name old time/op new time/op delta
StackCopyWithStkobj-8 10.5ms ± 5% 9.9ms ± 4% -6.03% (p=0.000 n=15+15)
Change-Id: I590f4449725983c7f8d274c4ac7ed384d9018d85
Reviewed-on: https://go-review.googlesource.com/c/go/+/354134
Trust: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
funcspdelta should be inlined: It is a tiny wrapper around another func.
The sanity check prevents that. Condition the sanity check on debugPcln.
While we're here, make the sanity check throw when it fails.
Change-Id: Iec022b8463b13a8e5a6d8479e7ddcb68909d6fe0
Reviewed-on: https://go-review.googlesource.com/c/go/+/354133
Trust: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
This CL enables intrinsic support to emit the following prefetch
instructions for PPC64 platform that are already emitted on other
platforms
1. Prefetch - prefetches data from memory address to cache;
2. PrefetchStreamed - prefetches data from memory address, with a
hint that this data is being streamed.
Benchmarks picked from go/test/bench/garbage
Parameters tested with:
GOMAXPROCS=8
tree2 -heapsize=1000000000 -cpus=8
tree -n=18
parser
peano
Performance results with this change on POWER9
name old time/op new time/op delta
Tree2-8 75.3ms ± 2% 65.0ms ± 6% -13.61% (p=0.003 n=5+7)
Tree-8 576ms ± 2% 576ms ± 1% ~ (p=0.756 n=11+10)
Parser-8 3.60s ± 2% 3.59s ± 1% ~ (p=0.818 n=6+6)
Peano-8 84.8ms ± 1% 84.6ms ± 1% ~ (p=0.180 n=6+6)
Results on POWER8 and POWER10 are similar
Change-Id: If4ac95a85aaa7b2266014e1f8fb7cd7440cbf906
Reviewed-on: https://go-review.googlesource.com/c/go/+/353730
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Trust: Michael Knyszek <mknyszek@google.com>
In the last CL I missed the fact that except for the final value the
code already unwraps the argument.
For #31103
Change-Id: Ic9099aeb50c6b3322fc14a90ac8026c1d8cb1698
Reviewed-on: https://go-review.googlesource.com/c/go/+/354091
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
In the case of a nointerface method on an instantiated type, we still
have to call methodWrapper, because methodWrapper generates the actual
generic method on the type as well. Currently, we don't call
methodWrapper, so the method on the instantiated type never gets filled
in.
Adjusted the code to still call methodWrapper, but not use the result,
in the case of a nointerface method on an instantiated type.
Change-Id: I34bca58de2861aa772be04eb8dd7695c5b7f3a77
Reviewed-on: https://go-review.googlesource.com/c/go/+/353369
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Trust: Dan Scales <danscales@google.com>
Add a simple test with an exported generic function that does
recover/defer, to test that recover/defer are exported/imported
properly (and a generic function with recover/defer works fine).
Change-Id: Idc3af101cbb78fc96bf945f1f5eab2740dd8994b
Reviewed-on: https://go-review.googlesource.com/c/go/+/353883
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Trust: Dan Scales <danscales@google.com>
This test hasn't passed since CL 349997, but the failure was not
detected because the Go project's builders do not have a 'fossil'
binary installed (#48802).
For #43684
Change-Id: I25544574ab48f4f146ae3795e541179e78815758
Reviewed-on: https://go-review.googlesource.com/c/go/+/354149
Trust: Bryan C. Mills <bcmills@google.com>
Run-TryBot: Bryan C. Mills <bcmills@google.com>
Reviewed-by: Russ Cox <rsc@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
This reverts commit golang.org/cl/352953.
Reason for revert: unsafe.Slice is considerably slower.
Part of this is extra safety checks (good), but most of it
is the function call overhead. We should consider open-coding it (#48798).
Impact of this change:
name old time/op new time/op delta
StackCopyWithStkobj-8 12.1ms ± 5% 11.6ms ± 3% -4.03% (p=0.009 n=10+8)
Change-Id: Ib2448e3edac25afd8fb55ffbea073b8b11521bde
Reviewed-on: https://go-review.googlesource.com/c/go/+/354090
Trust: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
This is a clean port of CL 354070 from types2 to go/types.
Change-Id: I44de1b8e6c0177e2a33e7f36a82465dc520c35aa
Reviewed-on: https://go-review.googlesource.com/c/go/+/354092
Trust: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
For now, the constraint's underlying type set must be a single
type that is sliceable.
Change-Id: I08b6a2e88fe35e8238a95b3f40dc969689021a0f
Reviewed-on: https://go-review.googlesource.com/c/go/+/354070
Trust: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
In case of amd64 the compiler issues checks if extensions are
available on a platform. With GOAMD64 microarchitecture levels
provided, some of the checks could be eliminated.
Change-Id: If15c178bcae273b2ce7d3673415cb8849292e087
Reviewed-on: https://go-review.googlesource.com/c/go/+/352010
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
roundsd and FMA (vfmadd231sd).
Change-Id: I2d91332667e577bd9bb903ac58904f62b8454128
Reviewed-on: https://go-review.googlesource.com/c/go/+/354069
Trust: Keith Randall <khr@golang.org>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
