See #15604. This was a bug in a CL that has since been
rolled back. Adding a test to challenge the next attempter.
Change-Id: Ic43be254ea6eaab0071018cdc61d9b1c21f19cbf
Reviewed-on: https://go-review.googlesource.com/23000
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
In regalloc, a sparse map is preallocated for later use by
spill-in-loop sinking. However, variables (spills) are added
during register allocation before spill sinking, and a map
query involving any of these new variables will index out of
bounds in the map.
To fix:
1) fix the queries to use s.orig[v.ID].ID instead, to ensure
proper indexing. Note that s.orig will be nil for values
that are not eligible for spilling (like memory and flags).
2) add a test.
Fixes#15585.
Change-Id: I8f2caa93b132a0f2a9161d2178320d5550583075
Reviewed-on: https://go-review.googlesource.com/22911
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
This flag is experimental and the semantics may change
even after Go 1.7 is released. There are no changes to code
not using the flag.
The first part is for reading by future compiles.
The second part is for reading by the final link step.
Splitting the file this way allows distributed build systems
to ship the compile-input part only to compile steps and
the linker-input part only to linker steps.
The first part is basically just the export data,
and the second part is basically everything else.
The overall files still have the same broad structure,
so that existing tools will work with both halves.
It's just that various pieces are empty in the two halves.
This also copies the two bits of data the linker needed from
export data into the object header proper, so that the linker
doesn't need any export data at all. That eliminates a TODO
that was left for switching to the binary export data.
(Now the linker doesn't need to know about the switch.)
The default is still to write out a combined output file.
Nothing changes unless you pass -linkobj to the compiler.
There is no support in the go command for -linkobj,
since the go command doesn't copy objects around.
The expectation is that other build systems (like bazel, say)
might take advantage of this.
The header adjustment and the option for the split output
was intended as part of the zip archives, but the zip archives
have been cut from Go 1.7. Doing this to the current archives
both unblocks one step in the switch to binary export data
and enables alternate build systems to experiment with the
new flag using the Go 1.7 release.
Change-Id: I8b6eab25b8a22b0a266ba0ac6d31e594f3d117f3
Reviewed-on: https://go-review.googlesource.com/22500
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
The problem was fixed by the rollback in CL 22930.
This CL just adds a test to prevent regressions.
Fixes#15602
Change-Id: I37453f6e18ca43081266fe7f154c6d63fbaffd9b
Reviewed-on: https://go-review.googlesource.com/22931
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
The new export format keeps track of all types that are exported.
If a type is seen that was exported before, only a reference to
that type is emitted. The importer maintains a list of all the
seen types and uses that list to resolve type references.
The existing compiler infrastructure's invariants assumes that
only named types are referred to before they are fully set up.
Referring to unnamed incomplete types causes problems. One of
the issues was #15548.
Added a new internal flag 'trackAllTypes' to enable/disable
this type tracking. With this change only named types are
tracked.
Verified that this fix also addresses #15548, even w/o the
prior fix for that issue (in fact that prior fix is turned
off if trackAllTypes is disabled because it's not needed).
The test for #15548 covers also this change.
For #15548.
Change-Id: Id0b3ff983629703d025a442823f99649fd728a56
Reviewed-on: https://go-review.googlesource.com/22839
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
The boolean destination in an OAS2DOTTYPE expression craps out during
compilation when trying to assign to a map entry because, unlike slice entries,
map entries are not directly addressable in memory. The solution is to
properly order the boolean destination node so that map entries are set
via autotmp variables.
Fixes#14678
Change-Id: If344e8f232b5bdac1b53c0f0d21eeb43ab17d3de
Reviewed-on: https://go-review.googlesource.com/22833
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Accidentally checked in the version of file c.go that doesn't
exhibit the bug - hence the test was not testing the bug fix.
Double-checked that this version exposes the bug w/o the fix.
Change-Id: Ie4dc455229d1ac802a80164b5d549c2ad4d971f5
Reviewed-on: https://go-review.googlesource.com/22837
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Minux Ma <minux@golang.org>
external linking is now supported.
Change-Id: I13e90c39dad86e60781adecdbe8e6bc9e522f740
Reviewed-on: https://go-review.googlesource.com/19811
Reviewed-by: Minux Ma <minux@golang.org>
This is necessary to avoid disrupting the go1 suite and gives
us a place to put other tests of basic compiler function and
correctness.
Change-Id: I36933819ff2bfe6a2121fff2be9a98efd2123d9a
Reviewed-on: https://go-review.googlesource.com/22597
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
The underlying issues have been fixed.
All the individual fixes have their own tests,
but it's still useful to have a plain source test.
Fixes#15084
Change-Id: I06c485a7d0716201bd57d1f3be53668dddd7ec14
Reviewed-on: https://go-review.googlesource.com/22426
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
This is a follow-up to CLs 19769 and 19770.
Change-Id: Ia9b71055613b80df4ce62b34fcc4f479f04f72fd
Reviewed-on: https://go-review.googlesource.com/22399
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
func f(x, y, z *int) {
a := []*int{x,y,z}
...
}
We used to use:
var tmp [3]*int
a := tmp[:]
a[0] = x
a[1] = y
a[2] = z
Now we do:
var tmp [3]*int
tmp[0] = x
tmp[1] = y
tmp[2] = z
a := tmp[:]
Doesn't sound like a big deal, but the compiler has trouble
eliminating write barriers when using the former method because it
doesn't know that the slice points to the stack. In the latter
method, the compiler knows the array is on the stack and as a result
doesn't emit any write barriers.
This turns out to be extremely common when building ... args, like
for calls fmt.Printf.
Makes go binaries ~1% smaller.
Doesn't have a measurable effect on the go1 fmt benchmarks,
unfortunately.
Fixes#14263
Update #6853
Change-Id: I9074a2788ec9e561a75f3b71c119b69f304d6ba2
Reviewed-on: https://go-review.googlesource.com/22395
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
See discussion at [1]. True value must have a fixed non-zero
representation meaning that a && b can be implemented as a & b.
[1] https://groups.google.com/forum/#!topic/golang-dev/xV0vPuFP9Vg
This change helps with m := a && b, but it's more common to see
if a && b { do something } which is not handled.
Change-Id: Ib6f9ff898a0a8c05d12466e2464e4fe781035394
Reviewed-on: https://go-review.googlesource.com/22313
Run-TryBot: Alexandru Moșoi <alexandru@mosoi.ro>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
They are guaranteed to be non-nil, no point in inserting
nil checks for them.
Fixes#15390
Change-Id: I3b9a0f2319affc2139dcc446d0a56c6785ae5a86
Reviewed-on: https://go-review.googlesource.com/22291
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
func f(a, b bool) bool {
return a || b
}
is now a single instructions (excluding loading and unloading the arguments):
v10 = ORB <bool> v11 v12 : AX
Change-Id: Iff63399410cb46909f4318ea1c3f45a029f4aa5e
Reviewed-on: https://go-review.googlesource.com/21872
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
The result of ODOTPTR, as well as a bunch of other ops,
should be the type of the result, not always a pointer type.
This fixes an amd64p32 bug where we were incorrectly truncating
a 64-bit slice index to 32 bits, and then barfing on a weird
load-64-bits-but-then-truncate-to-32-bits op that doesn't exist.
Fixes#15252
Change-Id: Ie62f4315fffd79f233e5449324ccc0879f5ac343
Reviewed-on: https://go-review.googlesource.com/22094
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Print numGC followed by numGC1, rather than printing numGC twice.
Change-Id: I8e7144b6a11d4ae9be0d82d88b86fed04b906e2f
Reviewed-on: https://go-review.googlesource.com/22087
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Map keys are currently validated in multiple locations but share
a common validation routine. The problem is that early validations
should be lenient enough to allow for forward types while the final
validations should not. The final validations should fail on forward
types since they've already settled.
This change also separates the key type checking from the creation
of the map via typMap. Instead of the mapqueue being populated in
copytype() by checking the map line number, it's populated in the
same block that validates the key type. This isolates key validation
logic while type checking.
Fixes#14988
Change-Id: Ia47cf6213585d6c63b3a35249104c0439feae658
Reviewed-on: https://go-review.googlesource.com/21830
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Removes 49 more bound checks in make.bash. For example:
var a[100]int
for i := 0; i < 50; i++ {
use a[i+25]
}
Change-Id: I85e0130ee5d07f0ece9b17044bba1a2047414ce7
Reviewed-on: https://go-review.googlesource.com/21379
Reviewed-by: David Chase <drchase@google.com>
Run-TryBot: Alexandru Moșoi <alexandru@mosoi.ro>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Be more careful about inserting instrumentation in racewalk.
If the node being instrumented is an OAS, and it has a non-
empty Ninit, then append instrumentation to the Ninit list
rather than letting it be inserted before the OAS (and the
compilation of its init list). This deals with the case that
the Ninit list defines a variable used in the RHS of the OAS.
Fixes#15091.
Change-Id: Iac91696d9104d07f0bf1bd3499bbf56b2e1ef073
Reviewed-on: https://go-review.googlesource.com/21771
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: David Chase <drchase@google.com>
Fails for the same reason as ppc64 and mips64 (incomplete
optimization).
Change-Id: Ieb4d997fc27d4f2b756e63dd7f588abe10c0213a
Reviewed-on: https://go-review.googlesource.com/20963
Reviewed-by: Bill O'Farrell <billotosyr@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Make sure the results of unsigned constant-folded
shifts are sign-extended into the AuxInt field.
Fixes#15175
Change-Id: I3490d1bc3d9b2e1578ed30964645508577894f58
Reviewed-on: https://go-review.googlesource.com/21586
Reviewed-by: Alexandru Moșoi <alexandru@mosoi.ro>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Fold the comparison when the SHR result is small.
Useful for:
- murmur mix like hashing where higher bits are desirable, i.e. hash = uint32(i * C) >> 18
- integer log2 via DeBruijn sequence: http://graphics.stanford.edu/~seander/bithacks.html#IntegerLogDeBruijn
Change-Id: If70ae18cb86f4cc83ab6213f88ced03cc4986156
Reviewed-on: https://go-review.googlesource.com/21514
Run-TryBot: Alexandru Moșoi <alexandru@mosoi.ro>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
Test goprint.go sometimes failed on a slow builder (plan9_arm)
because of timing dependency. Instead of sleeping for a fixed
time to allow the child goroutine to finish, wait explicitly for
child termination by calling runtime.NumGoroutine until the
returned value is 1.
Fixes#15097
Change-Id: Ib3ef5ec3c8277083c774542f48bcd4ff2f79efde
Reviewed-on: https://go-review.googlesource.com/21603
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
No point in doing anything for x=x assignments.
In addition, skipping these assignments prevents generating:
VARDEF x
COPY x -> x
which is bad because x is incorrectly considered
dead before the vardef.
Fixes#14904
Change-Id: I6817055ec20bcc34a9648617e0439505ee355f82
Reviewed-on: https://go-review.googlesource.com/21470
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Dave Cheney <dave@cheney.net>
Missed a case for closure calls (OCALLFUNC && indirect) in
esc.go:esccall.
Cleanup to runtime code for windows to more thoroughly hide
a technical escape. Also made code pickier about failing
to late non-optional kernel32.dll.
Fixes#14409.
Change-Id: Ie75486a2c8626c4583224e02e4872c2875f7bca5
Reviewed-on: https://go-review.googlesource.com/20102
Run-TryBot: David Chase <drchase@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Since BCE happens over several passes (opt, loopbce, prove)
it's easy to regress especially with rewriting.
The pass is only activated with special debug flag.
Change-Id: I46205982e7a2751156db8e875d69af6138068f59
Reviewed-on: https://go-review.googlesource.com/21510
Run-TryBot: Alexandru Moșoi <alexandru@mosoi.ro>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
Identify this assignment case and instead of the more general error
prog.go:6: cannot assign to students["sally"].age
produce
prog.go:6: cannot directly assign to struct field students["sally"].age in map
that explains why the assignment is not possible.
Fixes#13779.
Change-Id: I90c10b445f907834fc1735aa66e44a0f447aa74f
Reviewed-on: https://go-review.googlesource.com/21462
Reviewed-by: David Chase <drchase@google.com>
Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Change-Id: I91873aaebf79bdf1c00d38aacc1a1fb8d79656a7
Reviewed-on: https://go-review.googlesource.com/21433
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Handle this case:
if 0 <= i && i < len(a) {
use a[i]
}
Shaves about 5k from pkg/tools/linux_amd64/*.
Change-Id: I6675ff49aa306b0d241b074c5738e448204cd981
Reviewed-on: https://go-review.googlesource.com/21431
Run-TryBot: Alexandru Moșoi <alexandru@mosoi.ro>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
The issue was seen when inlining an exported function that contained
a fallthrough statement.
Fixes#15071
Change-Id: I1e8215ad49d57673dba7e8f8bd2ed8ad290dc452
Reviewed-on: https://go-review.googlesource.com/21452
Reviewed-by: Dave Cheney <dave@cheney.net>
There are 5293 loop in the main go repository.
A survey of the top most common for loops:
18 for __k__ := 0; i < len(sa.Addr); i++ {
19 for __k__ := 0; ; i++ {
19 for __k__ := 0; i < 16; i++ {
25 for __k__ := 0; i < length; i++ {
30 for __k__ := 0; i < 8; i++ {
49 for __k__ := 0; i < len(s); i++ {
67 for __k__ := 0; i < n; i++ {
376 for __k__ := range __slice__ {
685 for __k__, __v__ := range __slice__ {
2074 for __, __v__ := range __slice__ {
The algorithm to find induction variables handles all cases
with an upper limit. It currently doesn't find related induction
variables such as c * ind or c + ind.
842 out of 22954 bound checks are removed for src/make.bash.
1957 out of 42952 bounds checks are removed for src/all.bash.
Things to do in follow-up CLs:
* Find the associated pointer for `for _, v := range a {}`
* Drop the NilChecks on the pointer.
* Replace the implicit induction variable by a loop over the pointer
Generated garbage can be reduced if we share the sdom between passes.
% benchstat old.txt new.txt
name old time/op new time/op delta
Template 337ms ± 3% 333ms ± 3% ~ (p=0.258 n=9+9)
GoTypes 1.11s ± 2% 1.10s ± 2% ~ (p=0.912 n=10+10)
Compiler 5.25s ± 1% 5.29s ± 2% ~ (p=0.077 n=9+9)
MakeBash 33.5s ± 1% 34.1s ± 2% +1.85% (p=0.011 n=9+9)
name old alloc/op new alloc/op delta
Template 63.6MB ± 0% 63.9MB ± 0% +0.52% (p=0.000 n=10+9)
GoTypes 218MB ± 0% 219MB ± 0% +0.59% (p=0.000 n=10+9)
Compiler 978MB ± 0% 985MB ± 0% +0.69% (p=0.000 n=10+10)
name old allocs/op new allocs/op delta
Template 582k ± 0% 583k ± 0% +0.10% (p=0.000 n=10+10)
GoTypes 1.78M ± 0% 1.78M ± 0% +0.12% (p=0.000 n=10+10)
Compiler 7.68M ± 0% 7.69M ± 0% +0.05% (p=0.000 n=10+10)
name old text-bytes new text-bytes delta
HelloSize 581k ± 0% 581k ± 0% -0.08% (p=0.000 n=10+10)
CmdGoSize 6.40M ± 0% 6.39M ± 0% -0.08% (p=0.000 n=10+10)
name old data-bytes new data-bytes delta
HelloSize 3.66k ± 0% 3.66k ± 0% ~ (all samples are equal)
CmdGoSize 134k ± 0% 134k ± 0% ~ (all samples are equal)
name old bss-bytes new bss-bytes delta
HelloSize 126k ± 0% 126k ± 0% ~ (all samples are equal)
CmdGoSize 149k ± 0% 149k ± 0% ~ (all samples are equal)
name old exe-bytes new exe-bytes delta
HelloSize 947k ± 0% 946k ± 0% -0.01% (p=0.000 n=10+10)
CmdGoSize 9.92M ± 0% 9.91M ± 0% -0.06% (p=0.000 n=10+10)
Change-Id: Ie74bdff46fd602db41bb457333d3a762a0c3dc4d
Reviewed-on: https://go-review.googlesource.com/20517
Reviewed-by: David Chase <drchase@google.com>
Run-TryBot: Alexandru Moșoi <alexandru@mosoi.ro>
Added a debug flag "-d closure" to explain compilation of
closures (should this be done some other way? Should we
rewrite the "-m" flag to "-d escapes"?) Used this to
discover that cause was an OXXX node in the captured vars
list, and in turn noticed that OXXX nodes are explicitly
ignored in all other processing of captured variables.
Couldn't figure out a reproducer, did verify that this OXXX
was not caused by an unnamed return value (which is one use
of these). Verified lack of heap allocation by examining -S
output.
Assembly:
(runtime/mgc.go:1371) PCDATA $0, $2
(runtime/mgc.go:1371) CALL "".notewakeup(SB)
(runtime/mgc.go:1377) LEAQ "".gcBgMarkWorker.func1·f(SB), AX
(runtime/mgc.go:1404) MOVQ AX, (SP)
(runtime/mgc.go:1404) MOVQ "".autotmp_2242+88(SP), CX
(runtime/mgc.go:1404) MOVQ CX, 8(SP)
(runtime/mgc.go:1404) LEAQ go.string."GC worker (idle)"(SB), AX
(runtime/mgc.go:1404) MOVQ AX, 16(SP)
(runtime/mgc.go:1404) MOVQ $16, 24(SP)
(runtime/mgc.go:1404) MOVB $20, 32(SP)
(runtime/mgc.go:1404) MOVQ $0, 40(SP)
(runtime/mgc.go:1404) PCDATA $0, $2
(runtime/mgc.go:1404) CALL "".gopark(SB)
Added a check for compiling_runtime to ensure that this is
caught in the future. Added a test to test the check.
Verified that 1.5.3 did NOT reject the test case when
compiled with -+ flag, so this is not a recently added bug.
Cause of bug is two-part -- there was no leaking closure
detection ever, and instead it relied on capture-of-variables
to trigger compiling_runtime test, but closures improved in
1.5.3 so that mere capture of a value did not also capture
the variable, which thus allowed closures to escape, as well
as this case where the escape was spurious. In
fixedbugs/issue14999.go, compare messages for f and g;
1.5.3 would reject g, but not f. 1.4 rejects both because
1.4 heap-allocates parameter x for both.
Fixes#14999.
Change-Id: I40bcdd27056810628e96763a44f2acddd503aee1
Reviewed-on: https://go-review.googlesource.com/21322
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
The change in 20907 fixed varexpr but broke aliased. After that change,
a reference to a field in a struct would not be seen as aliasing itself.
Before that change, it would, but only because all fields in a struct
aliased everything.
This CL changes the compiler to consider all references to a field as
aliasing all other fields in that struct. This is imperfect--a
reference to one field does not alias another field--but is a simple fix
for the immediate problem. A better fix would require tracking the
specific fields as well.
Fixes#15042.
Change-Id: I5c95c0dd7b0699e53022fce9bae2e8f50d6d1d04
Reviewed-on: https://go-review.googlesource.com/21390
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Compound AUTO types weren't named previously. That was because live
variable analysis (plive.go) doesn't handle spilling to compound types.
It can't handle them because there is no valid place to put VARDEFs when
regalloc is spilling compound types.
compound types = multiword builtin types: complex, string, slice, and
interface.
Instead, we split named AUTOs into individual one-word variables. For
example, a string s gets split into a byte ptr s.ptr and an integer
s.len. Those two variables can be spilled to / restored from
independently. As a result, live variable analysis can handle them
because they are one-word objects.
This CL will change how AUTOs are described in DWARF information.
Consider the code:
func f(s string, i int) int {
x := s[i:i+5]
g()
return lookup(x)
}
The old compiler would spill x to two consecutive slots on the stack,
both named x (at offsets 0 and 8). The new compiler spills the pointer
of x to a slot named x.ptr. It doesn't spill x.len at all, as it is a
constant (5) and can be rematerialized for the call to lookup.
So compound objects may not be spilled in their entirety, and even if
they are they won't necessarily be contiguous. Such is the price of
optimization.
Re-enable live variable analysis tests. One test remains disabled, it
fails because of #14904.
Change-Id: I8ef2b5ab91e43a0d2136bfc231c05d100ec0b801
Reviewed-on: https://go-review.googlesource.com/21233
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
