Reuse block head or preceding instruction's line number for
register allocator's spill, fill, copy, rematerialization
instructionsl; and also for phi, and for no-src-pos
instructions. Assembler creates same line number tables
for copy-predecessor-line and for no-src-pos,
but copy-predecessor produces better-looking assembly
language output with -S and with GOSSAFUNC, and does not
require changes to tests of existing assembly language.
Split "copyInto" into two cases, one for register allocation,
one for otherwise. This caused the test score line change
count to increase by one, which may reflect legitimately
useful information preserved. Without any special treatment
for copyInto, the change count increases by 21 more, from
51 to 72 (i.e., quite a lot).
There is a test; using two naive "scores" for line number
churn, the old numbering is 2x or 4x worse.
Fixes#18902.
Change-Id: I0a0a69659d30ee4e5d10116a0dd2b8c5df8457b1
Reviewed-on: https://go-review.googlesource.com/36207
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
If there were more unused imports than
the maximum default number of errors to report,
the set of reported imports was non-deterministic.
Fix by accumulating and sorting them prior to output.
Fixes#20298
Change-Id: Ib3d5a15fd7dc40009523fcdc1b93ddc62a1b05f2
Reviewed-on: https://go-review.googlesource.com/42954
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
If we've already complained about a type T,
don't complain again about further expressions
involving it.
Fixes#20245 and hopefully all of its ilk.
Change-Id: Ic0abe8235d52e8a7ac40e3615aea8f3a54fd7cec
Reviewed-on: https://go-review.googlesource.com/42690
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
Compile:
package p
var f = func(...A)
Before this CL:
x.go:3:13: type %!v(PANIC=runtime error: invalid memory address or nil pointer dereference) is not an expression
x.go:3:17: undefined: A
After this CL:
x.go:3:13: type func(...<T>) is not an expression
x.go:3:17: undefined: A
Found with go-fuzz.
Fixes#20233
Change-Id: Ibb232b3954c4091071440eba48b44c4022a8083f
Reviewed-on: https://go-review.googlesource.com/42610
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Because the hint parameter is supposed to be treated
purely as a hint, if it doesn't meet the requirements
we disregard it and continue as if there was no hint
at all.
Fixes#19926
Change-Id: I86e7f99472fad6b99ba4e2fd33e4a9e55d55115e
Reviewed-on: https://go-review.googlesource.com/40854
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Without this, T can sneak through to the backend
with its width unknown.
Fixes#20174
Change-Id: I9b21e0e2641f75e360cc5e45dcb4eefe8255b675
Reviewed-on: https://go-review.googlesource.com/42175
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Updates #18089.
Test for that issue; it was inadvertently fixed
by CL 34988. Ensure that we don't regress on the fix.
Change-Id: Icb85fc20dbb0a47f028f088281319b552b16759d
Reviewed-on: https://go-review.googlesource.com/42173
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
The code in #20162 contains an embedded interface.
It didn't get dowidth'd by the frontend,
and during DWARF generation, ngotype asked
for a string description of it,
which triggered a request for the number of fields
in the interface, which triggered a dowidth,
which is disallowed in the backend.
The other changes in this CL are to support the test.
Fixes#20162
Change-Id: I4d0be5bd949c361d4cdc89a8ed28b10977e40cf9
Reviewed-on: https://go-review.googlesource.com/42131
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
dowidth is fundamentally unsafe to call from the back end;
it will cause data races.
Replace all calls to dowidth in the backend with
assertions that the width has been calculated.
Then fix all the cases in which that was not so,
including the cases from #20145.
Fixes#20145.
Change-Id: Idba3d19d75638851a30ec2ebcdb703c19da3e92b
Reviewed-on: https://go-review.googlesource.com/41970
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
When a constant doesn't fit in a single instruction, use two
paired instructions instead of the constant pool. For example
ADD $0xaa00bb, R0, R1
Used to rewrite to:
MOV ?(IP), R11
ADD R11, R0, R1
Instead, do:
ADD $0xaa0000, R0, R1
ADD $0xbb, R1, R1
Same number of instructions.
Good:
4 less bytes (no constant pool entry)
One less load.
Bad:
Critical path is one instruction longer.
It's probably worth it to avoid the loads, they are expensive.
Dave Cheney got us some performance numbers: https://perf.golang.org/search?q=upload:20170426.1
TL;DR mean 1.37% improvement.
Change-Id: Ib206836161fdc94a3962db6f9caa635c87d57cf1
Reviewed-on: https://go-review.googlesource.com/41612
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
This change adds line position tests for several yyerror calls in the
typechecker that are currently not tested in any way.
Untested yyerror calls were found by replacing them with
yerrorl(src.NoXPos, ...)
(thus destroying position information in the error), and then running
the test suite. No failures means no test coverage for the relevant
yyerror call.
For #19683
Change-Id: Iedb3d2f02141b332e9bfa76dbf5ae930ad2fddc3
Reviewed-on: https://go-review.googlesource.com/41477
Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
doubleselect.go defines a flag to control the number of iterations,
but never called flag.Parse so it was unusable.
Change-Id: Ib5d0c7119e7f7c9a808dcc02d0d9cc6ba5bbc16e
Reviewed-on: https://go-review.googlesource.com/41299
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
At VARKILLs, zero a variable if it is ambiguously live.
After the VARKILL anything this variable references
might be collected. If it were to become live again later,
the GC will see references to already-collected objects.
We don't know a variable is ambiguously live until very
late in compilation (after lowering, register allocation, ...),
so it is hard to generate the code in an arch-independent way.
We also have to be careful not to clobber any registers.
Fortunately, this almost never happens so performance is ~irrelevant.
There are only 2 instances where this triggers in the stdlib.
Fixes#20029
Change-Id: Ia9585a91d7b823fad4a9d141d954464cc7af31f4
Reviewed-on: https://go-review.googlesource.com/41076
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
Follow-up on https://go-review.googlesource.com/#/c/39998/
which dropped this information.
The reported blocks are the innermost blocks containing a
label jumped to from outside, not the outermost block as
reported originally by cmd/compile.
We could report the outermost block with a slighly more
involved algorithm (need to track containing blocks for
all unresolved forward gotos), but since gccgo also reports
the innermost blocks, the current approach seems good enough.
Change-Id: Ic0235b8fafe8d5f99dc9872b58e90e8d9e72c5db
Reviewed-on: https://go-review.googlesource.com/40980
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Instead of a separate check control flow pass (checkcfg.go)
operating on nodes, perform this check at parse time on the
new syntax tree. Permits this check to be done concurrently,
and doesn't depend on the specifics of the symbol's dclstack
implementation anymore. The remaining dclstack uses will be
removed in a follow-up change.
- added CheckBranches Mode flag (so we can turn off the check
if we only care about syntactic correctness, e.g. for tests)
- adjusted test/goto.go error messages: the new branches
checker only reports if a goto jumps into a block, but not
which block (we may want to improve this again, eventually)
- also, the new branches checker reports one variable that
is being jumped over by a goto, but it may not be the first
one declared (this is fine either way)
- the new branches checker reports additional errors for
fixedbugs/issue14006.go (not crucial to avoid those errors)
- the new branches checker now correctly reports only
variable declarations being jumped over, rather than
all declarations (issue 8042). Added respective tests.
Fixes#8042.
Change-Id: I53b6e1bda189748e1e1fb5b765a8a64337c27d40
Reviewed-on: https://go-review.googlesource.com/39998
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Now only cmd/asm and cmd/compile depend on cmd/internal/obj. Changing
the assembler backends no longer requires reinstalling cmd/link or
cmd/addr2line.
There's also now one canonical definition of the object file format in
cmd/internal/objabi/doc.go, with a warning to update all three
implementations.
objabi is still something of a grab bag of unrelated code (e.g., flag
and environment variable handling probably belong in a separate "tool"
package), but this is still progress.
Fixes#15165.
Fixes#20026.
Change-Id: Ic4b92fac7d0d35438e0d20c9579aad4085c5534c
Reviewed-on: https://go-review.googlesource.com/40972
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
This rewrites runtime.Caller in terms of stackExpander, which already
handles inlined frames and partially skipped frames. This also has the
effect of making runtime.Caller understand cgo frames if there is a cgo
symbolizer.
Updates #19348.
Change-Id: Icdf4df921aab5aa394d4d92e3becc4dd169c9a6e
Reviewed-on: https://go-review.googlesource.com/40270
Run-TryBot: David Lazar <lazard@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
The fixedbugs/issue12536.go file was erroneously deleted just before
committing the patch that fixed the issue (CL 14400).
That's an easy test and there's a small reproducer in the issue, add
it back.
Updates #12536
Change-Id: Ib7b0cd245588299e9a5469e1d75805fd0261ce1a
Reviewed-on: https://go-review.googlesource.com/40712
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Also adjust truncfltlit to make it more similar to trunccmplxlit, and
make it report an error for bad Etypes.
Fixes#19947
Change-Id: I6684523e989c2293b8a8e85bd2bfb9c399c5ea36
Reviewed-on: https://go-review.googlesource.com/40453
Reviewed-by: Robert Griesemer <gri@golang.org>
When casting an ideal to complex{64,128}, for example during the
evaluation of
var a = complex64(0) / 1e-50
we want the compiler to report a division-by-zero error if a divisor
would be zero after the cast.
We already do this for floats; for example
var b = float32(0) / 1e-50
generates a 'division by zero' error at compile time (because
float32(1e-50) is zero, and the cast is done before performing the
division).
There's no such check in the path for complex{64,128} expressions, and
no cast is performed before the division in the evaluation of
var a = complex64(0) / 1e-50
which compiles just fine.
This patch changes the convlit1 function so that complex ideals
components (real and imag) are correctly truncated to float{32,64}
when doing an ideal -> complex{64, 128} cast.
Fixes#11674
Change-Id: Ic5f8ee3c8cfe4c3bb0621481792c96511723d151
Reviewed-on: https://go-review.googlesource.com/37891
Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
Continues outside of a loop are not allowed. Most of these possibilities
were tested in label1.go, but one was missing - a plain continue in a
switch/select but no enclosing loop.
This used to error with a "continue not in loop" in 1.8, but recently
was broken by c03e75e5. In particular, innerloop does not only account
for loops, but also for switches and selects. Swap it by bools that
track whether breaks and continues should be allowed.
While at it, improve the wording of errors for breaks that are not where
they should be. Change "loop" by "loop, switch, or select" since they
can be used in any of those.
And add tests to make sure this isn't broken again. Use a separate func
since I couldn't get the compiler to crash on f() itself, possibly due
to the recursive call on itself.
Fixes#19934.
Change-Id: I8f09c6c2107fd95cac50efc2a8cb03cbc128c35e
Reviewed-on: https://go-review.googlesource.com/40357
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
This avoids false positives
like those found in #19880.
Fixes#19880
Change-Id: I583c16cc3c71e7462a72500db9ea2547c468f8c1
Reviewed-on: https://go-review.googlesource.com/40255
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Given code such as
type T struct {
_ string
}
func f() {
var x = T{"space"}
// ...
}
the compiler rewrote the 'var x' line as
var x T
x._ = "space"
The compiler then rejected the assignment to
a blank field, thus rejecting valid code.
It also failed to catch a number of invalid assignments.
And there were insufficient checks for validity
when emitting static data, leading to ICEs.
To fix, check earlier for explicit blanks field names,
explicitly handle legit blanks in sinit,
and don't try to emit static data for nodes
for which typechecking has failed.
Fixes#19482
Change-Id: I594476171d15e6e8ecc6a1749e3859157fe2c929
Reviewed-on: https://go-review.googlesource.com/38006
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Currently we expand comparison with small constant strings into len check
and a sequence of byte comparisons. Generate 16/32/64-bit comparisons,
instead of bytewise on 386 and amd64. Also increase limits on what is
considered small constant string.
Shaves ~30kb (0.5%) from go executable.
This also updates test/prove.go to keep test case valid.
Change-Id: I99ae8871a1d00c96363c6d03d0b890782fa7e1d9
Reviewed-on: https://go-review.googlesource.com/38776
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Dupok symbols may be defined in multiple packages. Its associated
package is chosen sort of arbitrarily (the first containing package
that the linker loads). Canonicalize its package to the package
with which it will be laid down in text, which is the first package
in dependency order that defines the symbol. So later passes (for
example, trampoline insertion pass) know that the dupok symbol
is laid down along with the package.
Fixes#19764.
Change-Id: I7cbc7474ff3016d5069c8b7be04af934abab8bc3
Reviewed-on: https://go-review.googlesource.com/39150
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
Reviewed-by: David Chase <drchase@google.com>
Now that we no longer generate dead code,
it is possible to follow block predecessors
into infinite loops with no variable definitions,
causing an infinite loop during phi insertion.
To fix that, check explicitly whether the predecessor
is dead in lookupVarOutgoing, and if so, bail.
The loop in lookupVarOutgoing is very hot code,
so I am wary of adding anything to it.
However, a long, CPU-only benchmarking run shows no
performance impact at all.
Fixes#19783
Change-Id: I8ef8d267e0b20a29b5cb0fecd7084f76c6f98e47
Reviewed-on: https://go-review.googlesource.com/38913
Reviewed-by: David Chase <drchase@google.com>
The uintptr-typed Data field in reflect.SliceHeader and
reflect.StringHeader needs special treatment because it is
really a pointer. Add the special treatment in walk for
bug #19168 to escape analysis.
Includes extra debugging that was helpful.
Fixes#19743.
Change-Id: I6dab5002f0d436c3b2a7cdc0156e4fc48a43d6fe
Reviewed-on: https://go-review.googlesource.com/38738
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Previously, an inlined call to wg.Done() in package main would have the
following incorrect symbol name:
main.(*sync.WaitGroup).Done
This change modifies methodname to return the correct symbol name:
sync.(*WaitGroup).Done
This fix was suggested by @mdempsky.
Fixes#19467.
Change-Id: I0117838679ac5353789299c618ff8c326712d94d
Reviewed-on: https://go-review.googlesource.com/37866
Reviewed-by: Austin Clements <austin@google.com>
The `skip` argument passed to runtime.Caller and runtime.Callers should
be interpreted as the number of logical calls to skip (rather than the
number of physical stack frames to skip). This changes runtime.Callers
to skip inlined calls in addition to physical stack frames.
The result value of runtime.Callers is a slice of program counters
([]uintptr) representing physical stack frames. If the `skip` parameter
to runtime.Callers skips part-way into a physical frame, there is no
convenient way to encode that in the resulting slice. To avoid changing
the API in an incompatible way, our solution is to store the number of
skipped logical calls of the first frame in the _second_ uintptr
returned by runtime.Callers. Since this number is a small integer, we
encode it as a valid PC value into a small symbol called:
runtime.skipPleaseUseCallersFrames
For example, if f() calls g(), g() calls `runtime.Callers(2, pcs)`, and
g() is inlined into f, then the frame for f will be partially skipped,
resulting in the following slice:
pcs = []uintptr{pc_in_f, runtime.skipPleaseUseCallersFrames+1, ...}
We store the skip PC in pcs[1] instead of pcs[0] so that `pcs[i:]` will
truncate the captured stack trace rather than grow it for all i.
Updates #19348.
Change-Id: I1c56f89ac48c29e6f52a5d085567c6d77d499cf1
Reviewed-on: https://go-review.googlesource.com/37854
Run-TryBot: David Lazar <lazard@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Previously, we could not run tests with -l=4 on NaCl since the buildrun
action is not supported on NaCl. This lets us run tests with build flags
on NaCl.
Change-Id: I103370c7b823b4ff46f47df97e802da0dc2bc7c3
Reviewed-on: https://go-review.googlesource.com/38170
Run-TryBot: David Lazar <lazard@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Gccgo crashed compiling a function that returned multiple zero-sized values.
Change-Id: I499112cc310e4a4f649962f4d2bc9fee95dee1b6
Reviewed-on: https://go-review.googlesource.com/38772
Reviewed-by: Than McIntosh <thanm@google.com>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
This was cherry-picked to 1.8 as CL 38587, but on master issue was fixed
by CL 37661. Add still relevant part (test) and close issue, since test passes.
Fixes#19201
Change-Id: I6415792e2c465dc6d9bd6583ba1e54b107bcf5cc
Reviewed-on: https://go-review.googlesource.com/37376
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
Prior to this CL, Type.Width != 0 was the mark
of a Type whose Width had been calculated.
As a result, dowidth always recalculated
the width of struct{}.
This, combined with the prohibition on calculating
the width of a FuncArgsStruct and the use of
struct{} as a function argument,
meant that there were circumstances in which
it was forbidden to call dowidth on a type.
This inhibits refactoring to call dowidth automatically,
rather than explicitly.
Instead add a helper method, Type.WidthCalculated,
and implement as Type.Align > 0.
Type.Width is not a good candidate for tracking
whether the width has been calculated;
0 is a value type width, and Width is subject to
too much magic value game-playing.
For good measure, add a test for #11354.
Change-Id: Ie9a9fb5d924e7a2010c1904ae5e38ed4a38eaeb2
Reviewed-on: https://go-review.googlesource.com/38468
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Ensure that we have a test for when the compiler
encounters a type switch on a non-interface value.
Change-Id: Icb222f986894d0190e1241ca65396b4950e7d14f
Reviewed-on: https://go-review.googlesource.com/38661
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Avoid construction of incorrect syntax trees in presence of errors.
For #19663.
Change-Id: I43025a3cf0fe02cae9a57e8bb9489b5f628c3fd7
Reviewed-on: https://go-review.googlesource.com/38604
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Simple phi insertion already had a heuristic to check
for dead blocks, namely having no predecessors.
When we stopped generating code for dead blocks,
we eliminated some values contained in more subtle
dead blocks, which confused phi insertion.
Compensate by beefing up the reachability check.
Fixes#19678
Change-Id: I0081e4a46f7ce2f69b131a34a0553874a0cb373e
Reviewed-on: https://go-review.googlesource.com/38602
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
CL 37499 allows inlining more functions by ignoring dead code.
However, that dead code can contain non-exportable constructs.
Teach the exporter not to export dead code.
Fixes#19679
Change-Id: Idb1d3794053514544b6f1035d29262aa6683e1e7
Reviewed-on: https://go-review.googlesource.com/38601
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
Almost never happens in practice.
The compiler will generate reasonable code anyway,
since assignments involving [0]T never do any work.
Fixes#19696Fixes#19671
Change-Id: I350d2e0c5bb326c4789c74a046ab0486b2cee49c
Reviewed-on: https://go-review.googlesource.com/38599
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
