The newly added routines are exact copies of the generic routines,
except for the function names and that growWork_fastX calls evacuate_fastX.
Actual optimization will happen in subsequent CLs.
This is intended to ease reviewing.
Change-Id: I52ef7dd40b2bdfc9cba2496544c0604e6e71cf7f
Reviewed-on: https://go-review.googlesource.com/59130
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
In TimeoutHandler, use a request whose context has been configured with
the handler's timeout
Fixes#20712
Change-Id: Ie670148f85fdad46841ff29232042309e15665ae
Reviewed-on: https://go-review.googlesource.com/46412
Run-TryBot: Tom Bergan <tombergan@google.com>
Reviewed-by: Tom Bergan <tombergan@google.com>
Just to make it clearer which regexps are positive and which
regexps are negative.
Change-Id: Ia190e89be28048fcae2491506f552afad90a5f85
Reviewed-on: https://go-review.googlesource.com/59490
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Martin Möhrmann <moehrmann@google.com>
Reviewed-by: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
We can rematerialize only ops that have SP or SB as their only argument.
There are some ADDQconst(SP) that can be rematerialized, but are spilled/filled instead,
so mark addconst as rematerializeable. This shaves ~1kb from go tool.
Change-Id: Ib4cf4fe5f2ec9d3d7e5f0f77f1193eba66ca2f08
Reviewed-on: https://go-review.googlesource.com/54393
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
When running multiple iOS builds on the same host, GOIOS_DEVICE_ID
is used to distinguish the devices. To improve support,
- Only restart the particular device when invoking iostest.bash
with the -restart flag.
- Make the exec wrapper lock file per-device.
For the iOS builder.
Change-Id: Id6f222981f25036399a43c3202a393dba89d87cb
Reviewed-on: https://go-review.googlesource.com/57970
Run-TryBot: Elias Naur <elias.naur@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
The SSA compiler currently generates MOVOstore instructions
to optimize 16 bytes moves on AMD64 architecture.
However, we can't use the MOVOstore instruction on Plan 9,
because floating point operations are not allowed in the
note handler.
We rely on the useSSE flag to disable the use of the
MOVOstore instruction on Plan 9 and replace it by two
MOVQstore instructions.
Fixes#21625
Change-Id: Idfefcceadccafe1752b059b5fe113ce566c0e71c
Reviewed-on: https://go-review.googlesource.com/59171
Run-TryBot: David du Colombier <0intro@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ilya Tocar <ilya.tocar@intel.com>
Current documentation lacks simple examples for functions Regexp.Expand
and Regexp.ExpandString whose usage is unclear from description alone.
This commit adds examples that demonstrate usage in practical way.
Fixes#21649
Change-Id: I7b2c06c8ab747f69a6578f0595bf0f3c742ac479
Reviewed-on: https://go-review.googlesource.com/59470
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Most of these are return values that were part of a receiving parameter,
so they're still accessible.
A few others are not, but those have never had a use.
Found with github.com/mvdan/unparam, after Kevin Burke's suggestion that
the tool should also warn about unused result parameters.
Change-Id: Id8b5ed89912a99db22027703a88bd94d0b292b8b
Reviewed-on: https://go-review.googlesource.com/55910
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
* use bool instead of int if it's adequate.
* remove blank lines.
Change-Id: Ic4a5644a33ed9fc7ce388ef8ba15f1732446fcfc
Reviewed-on: https://go-review.googlesource.com/59375
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
The PKCS #1 v2.2 document has been moved to an EMC
website with a new URL. This CL updates the reference to the document to
the new URL.
The new URL is referenced under https://en.wikipedia.org/wiki/PKCS_1Fixes#21642
Change-Id: Ib8738b0c4c3bb9ec427bebea20c4aacd607ba0db
Reviewed-on: https://go-review.googlesource.com/59351
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
This has been supported since Go 1 and there's even a test for it.
The documentation was missing.
Fixes#21409.
Change-Id: I5813488f6a98c1b4506c239e968d43344b91be12
Reviewed-on: https://go-review.googlesource.com/59412
Reviewed-by: Ian Lance Taylor <iant@golang.org>
The IsMetaPackage function was made exported when it was moved from
cmd/go to cmd/go/internal/load in CL 36196. Its documentation wasn't
updated accordingly. This change fixes that, resolving a golint issue.
Updates #18653.
Change-Id: Icf89461000754d0f09e6617b11c838e4c050d5a6
Reviewed-on: https://go-review.googlesource.com/59430
Reviewed-by: Dave Cheney <dave@cheney.net>
One example of a heavily-used zero-size value is encoding/binary.BigEndian.
Change-Id: I8e873c447e154ab2ca61b7315df774693891270c
Reviewed-on: https://go-review.googlesource.com/59330
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Went mainly for the ones that make no sense, such as the ones
mid-sentence or after commas.
Change-Id: Ie245d2c19cc7428a06295635cf6a9482ade25ff0
Reviewed-on: https://go-review.googlesource.com/57293
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Many aspects of the package is woefully undocumented.
With the recent flurry of improvements, the package is now at feature
parity with the GNU and TAR tools. Thoroughly all of the public API
and perform some minor stylistic cleanup in some code segments.
Change-Id: Ic892fd72c587f30dfe91d1b25b88c9c8048cc389
Reviewed-on: https://go-review.googlesource.com/59210
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
The PAX specification says the following:
<<<
'g' represents global extended header records for the following files in the archive.
The format of these extended header records shall be as described in pax Extended Header.
Each value shall affect all subsequent files that do not override that value
in their own extended header record and until another global extended header record
is reached that provides another value for the same field.
>>>
This CL adds support for parsing and composing global PAX records,
but intentionally does not provide support for automatically
persisting the global state across files.
Changes made:
* When Reader encounters a TypeXGlobalRecord header, it parses the
PAX records and returns them to the user ad-verbatim. Reader does not
store them in its state, ensuring it has no effect on future Next calls.
* When Writer receives a TypeXGlobalRecord header, it writes the
PAX records to the archive ad-verbatim. It does not store them in
its state, ensuring it has no effect on future WriteHeader calls.
* The restriction regarding empty record values is lifted since this
value is used to represent deletion in global headers.
Why provide raw support only:
* Some archives in the wild have a global header section (often empty)
and it is the user's responsibility to manually read and discard it's body.
The logic added here allows users to more easily skip over these sections.
* For users that do care about global headers, having access to the raw
records allows them to implement the functionality of global headers themselves
and manually persist the global state across files.
* We can still upgrade to a full implementation in the future.
Why we don't provide full support:
* Even though the PAX specification describes their operation in detail,
both the GNU and BSD tar tools (which are the most common implementations)
do not have a consistent interpretation of many details.
* Global headers were a controversial feature in PAX, by admission of the
specification itself:
<<<
The concept of a global extended header (typeflag g) was controversial.
The typeflag g global headers should not be used with interchange media that
could suffer partial data loss in transporting the archive.
>>>
* Having state persist from entry-to-entry complicates the implementation
for a feature that is not widely used and not well supported.
Change-Id: I1d904cacc2623ddcaa91525a5470b7dbe226c7e8
Reviewed-on: https://go-review.googlesource.com/59190
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
Otherwise, if there are any parallel tests, it will hang and panic with
"all goroutines are asleep - deadlock!".
Do not use flag.Uint to handle the error for us because we also want to
error on N==0, and because it would make setting the default to
GOMAXPROCS(0) more difficult, since it's an int.
Check for it right after flag.Parse, and mimic flag errors by printing
the usage and returning exit code 2.
Fixes#20542.
Change-Id: I0c9d4587f83d406a8f5e42ed74e40be46d639ffb
Reviewed-on: https://go-review.googlesource.com/54150
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
This addresses the case of a -timeout panic, but not the more
general case of a signal arriving. See CL 48370 and CL 44352
for recent difficulties in that area.
"-timeout" here means flag usage to distinguish from the
default timeout termination which uses signals.
Fixes#19394
Change-Id: I5452d5422c0c080e940cbcc8c6606049975268c6
Reviewed-on: https://go-review.googlesource.com/48491
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
This CL adds the following new publicly visible API:
type Header struct { ...; PAXRecords map[string]string }
The new Header.PAXRecords field is a map of all PAX extended header records.
We suggest (but do not enforce) that users use VENDOR-prefixed keys
according to the following in the PAX specification:
<<<
The standard developers have reserved keyword name space for vendor extensions.
It is suggested that the format to be used is:
VENDOR.keyword
where VENDOR is the name of the vendor or organization in all uppercase letters.
>>>
When reading, the Header.PAXRecords is populated with all PAX records
encountered so far, including basic ones (e.g., "path", "mtime", etc).
When writing, the fields of Header will be merged into PAXRecords,
overwriting any records that may conflict.
Since PAXRecords is a more expressive feature than Xattrs and
is entirely a superset of Xattrs, we mark Xattrs as deprecated,
and steer users towards the new PAXRecords API.
The issue has a discussion about adding a Header.SetPAXRecord method
to help validate records and keep the Header fields in sync.
However, we do not include that in this CL since that helper
method can always be added in the future.
There is no support for global records.
Fixes#14472
Change-Id: If285a52749acc733476cf75a2c7ad15bc1542071
Reviewed-on: https://go-review.googlesource.com/58390
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Add this early in the cycle so that we can start regression testing
of the master toolchain.
Change-Id: Ida3ccad6e9642648f489babd12877fc8a5eca07a
Reviewed-on: https://go-review.googlesource.com/59151
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
If Less(a, b) returns true when a is less than b, the correct way to
check if a is greater than b is to use Less(b, a). It is wrong to use
!Less(a, b) because that checks if a is greater than *or equal to* b.
1. The decreasingDistance function in Example_sortKeys makes this
mistake. Fix it.
2. The documentation of multiSorter.Less says it loops along the less
functions until it finds a comparison "that is either Less or
!Less". This is nonsense, because (Less(a, b) or !Less(a, b)) is
always true. Fix the documentation to say that it finds a
comparison "that discriminates between the two items (one is less
than the other)". The implementation already does this correctly.
Change-Id: If52b79f68e4fdb0d1095edf29bdecdf154a61b8d
Reviewed-on: https://go-review.googlesource.com/57752
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
The current code treats the type of SIMD&FP register as C_REG incorrectly.
The fix code converts C_REG type into C_FREG type.
Uncomment fcsels/fcseld test cases.
Fixes#21582
Change-Id: I754c51f72a0418bd352cbc0f7740f14cc599c72d
Reviewed-on: https://go-review.googlesource.com/58350
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Fix two small but serious bugs in the DWARF location list code that
should have been caught by the automated tests I didn't write.
After emitting debug information for a user variable, mark it as done
so that it doesn't get emitted again. Otherwise it would be written once
per slot it was decomposed into.
Correct a merge error in CL 44350: the location list abbreviations need
to have DW_AT_decl_line too, otherwise the resulting DWARF is gibberish.
Change-Id: I6ab4b8b32b7870981dac80eadf0ebfc4015ccb01
Reviewed-on: https://go-review.googlesource.com/59070
Run-TryBot: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
Change-Id: Ic3ce2f3c055f2636ec8fc9cec8592e596b18dc05
Reviewed-on: https://go-review.googlesource.com/54771
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Sometimes (often for calls) we generate code like this:
MOVQ (addr),AX
MOVQ 8(addr),BX
MOVQ AX,(otheraddr)
MOVQ BX,8(otheraddr)
Replace it with
MOVUPS (addr),X0
MOVUPS X0,(otheraddr)
For completeness do the same for 8,16,32-bit loads/stores too.
Shaves 1% from code sections of go tool.
/localdisk/itocar/golang/bin/go 10293917
go_old 10334877 [40960 bytes]
read-only data = 682 bytes (0.040769%)
global text (code) = 38961 bytes (1.036503%)
Total difference 39643 bytes (0.674628%)
Updates #6853
Change-Id: I1f0d2f60273a63a079b58927cd1c4e3429d2e7ae
Reviewed-on: https://go-review.googlesource.com/57130
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Throughout the spec we use the notion of a constant x being
representable by a value of type T. While intuitively clear,
at least for floating-point and complex constants types, the
concept was not well-defined. In the section on Conversions
there was an extra rule for floating-point types only and it
missed the case of floating-point values overflowing to an
infinity after rounding.
Since the concept is important to Go, and a compiler most
certainly will have a function to test "representability",
it seems warranted to define the term explicitly in the spec.
This change introduces a new entry "Representability" under
the section on "Properties of types and values", and defines
the term explicitly, together with examples.
The phrase used is "representable by" rather than "representable as"
because the former use is prevalent in the spec.
Additionally, it clarifies that a floating-point constant
that overflows to an infinity after rounding is never
representable by a value of a floating-point type, even though
infinities are valid values of IEEE floating point types.
This is required because there are not infinite value constants
in the language (like there is also no -0.0) and representability
also matters for constant conversions. This is not a language
change, and type-checkers have been following this rule before.
The change also introduces links throughout the spec to the new
section as appropriate and removes duplicate text and examples
elsewhere (Constants and Conversions sections), leading to
simplifications in the relevant paragraphs.
Fixes#15389.
Change-Id: I8be0e071552df0f18998ef4c5ef521f64ffe8c44
Reviewed-on: https://go-review.googlesource.com/57530
Reviewed-by: Rob Pike <r@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Prior to this change, we use typedmemmove to write the key
value to its new location in mapassign_fast32 and mapassign_fast64.
(The use of typedmemmove was a last-minute fix in the 1.9 cycle;
see #21297 and CL 53414.)
This is significantly less inefficient than direct assignment or
calling writebarrierptr directly.
Fortunately, there aren't many cases to consider.
On systems with 32 bit pointers:
* A 32 bit AMEM value either is a single pointer or has no pointers.
* A 64 bit AMEM value may contain a pointer at the beginning,
a pointer at 32 bits, or two pointers.
On systems with 64 bit pointers:
* A 32 bit AMEM value contains no pointers.
* A 64 bit AMEM value either is a single pointer or has no pointers.
All combinations except the 32 bit pointers / 64 bit AMEM value are
cheap and easy to handle, and the problematic case is likely rare.
The most popular map keys appear to be ints and pointers.
So we handle them exhaustively. The sys.PtrSize checks are constant branches
and are eliminated by the compiler.
An alternative fix would be to return a pointer to the key,
and have the calling code do the assignment, at which point the compiler
would have full type information.
Initial tests suggest that the performance difference between these
strategies is negligible, and this fix is considerably simpler,
and has much less impact on binary size.
Fixes#21321
Change-Id: Ib03200e89e2324dd3c76d041131447df66f22bfe
Reviewed-on: https://go-review.googlesource.com/59110
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Implement int reg <-> fp reg moves on amd64.
If we see a load to int reg followed by an int->fp move, then we can just
load to the fp reg instead. Same for stores.
math.Abs is now:
MOVQ "".x+8(SP), AX
SHLQ $1, AX
SHRQ $1, AX
MOVQ AX, "".~r1+16(SP)
math.Copysign is now:
MOVQ "".x+8(SP), AX
SHLQ $1, AX
SHRQ $1, AX
MOVQ "".y+16(SP), CX
SHRQ $63, CX
SHLQ $63, CX
ORQ CX, AX
MOVQ AX, "".~r2+24(SP)
math.Float64bits is now:
MOVSD "".x+8(SP), X0
MOVSD X0, "".~r1+16(SP)
(it would be nicer to use a non-SSE reg for this, nothing is perfect)
And due to the fix for #21440, the inlined version of these improve as well.
name old time/op new time/op delta
Abs 1.38ns ± 5% 0.89ns ±10% -35.54% (p=0.000 n=10+10)
Copysign 1.56ns ± 7% 1.35ns ± 6% -13.77% (p=0.000 n=9+10)
Fixes#13095
Change-Id: Ibd7f2792412a6668608780b0688a77062e1f1499
Reviewed-on: https://go-review.googlesource.com/58732
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Ilya Tocar <ilya.tocar@intel.com>
Make it dead simple to see visually what the function outputs in
various scenarios.
Change-Id: I8f6fcd72fa1515361481f0510412cde221e1d4e3
Reviewed-on: https://go-review.googlesource.com/51630
Run-TryBot: Kevin Burke <kev@inburke.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Reviewed-by: Hiroshi Ioka <hirochachacha@gmail.com>
CL 36428 changed the way nanotime works so on Darwin and Windows it
now depends on runtime.startNano, which is computed at runtime.init
time. Unfortunately, the `runtimeInitTime = nanotime()` initialization
happened *before* runtime.init, so on these platforms runtimeInitTime
is set incorrectly. The one (and only) consequence of this is that the
start time printed in gctrace lines is bogus:
gc 1 18446653480.186s 0%: 0.092+0.47+0.038 ms clock, 0.37+0.15/0.81/1.8+0.15 ms cpu, 4->4->1 MB, 5 MB goal, 8 P
To fix this, this commit moves the runtimeInitTime initialization to
shortly after runtime.init, at which point nanotime is safe to use.
This also requires changing the condition in newproc1 that currently
uses runtimeInitTime != 0 simply to detect whether or not the main M
has started. Since runtimeInitTime could genuinely be 0 now, this
introduces a separate flag to newproc1.
Fixes#21554.
Change-Id: Id874a4b912d3fa3d22f58d01b31ffb3548266d3b
Reviewed-on: https://go-review.googlesource.com/58690
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rick Hudson <rlh@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Generated with
github.com/catapult/tracing/bin/vulcanize_trace_viewer
catapult @ ab4d571fa
Renamed trace_viewer_lean.html to trace_viewer_full.html
to make it clear we are using the full version of trace viewer
(waiting for https://github.com/catapult-project/catapult/issues/2247
to be fixed).
Update #15302
Change-Id: Ice808bb27ab79a1dec9fc863e0c5a761027ebfbe
Reviewed-on: https://go-review.googlesource.com/58750
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Since golang.org/cl/31670, we've stopped using the 'embedded' function
for handling struct embeddings within package export data. Now the
only remaining use is for Go source files, which allows for some
substantial simplifications:
1. CenterDot never appears within Go source files, so that logic can
simply be removed.
2. The field name will always be declared in the local package.
Passes toolstash-check.
Change-Id: I59505f62824206dd5de0782918f98fbef6e93224
Reviewed-on: https://go-review.googlesource.com/58790
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
The enumeration of numeric types missed the complex types.
Clarify by removing the explicit enumeration and referring
to numeric types instead.
Fixes#21579.
Change-Id: If36c2421f8501eeec82a07f442ac2e16a35927ba
Reviewed-on: https://go-review.googlesource.com/58491
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
Discovered while debugging CL 53644.
No test case because these are purely internal conversions that should
never end up resulting in compiler warnings or even generated code.
Updates #19683.
Change-Id: I0d9333ef2c963fa22eb9b5335bb022bcc9b25708
Reviewed-on: https://go-review.googlesource.com/58190
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
The old wording seemed to imply that nil is a kind of type.
Slightly reworded for clarity.
Fixes#21580.
Change-Id: I29898bf0125a10cb8dbb5c7e63ec5399ebc590ca
Reviewed-on: https://go-review.googlesource.com/58490
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Rob Pike <r@golang.org>
