OpenBSD's __threxit syscall takes a pointer to a 32-bit value that will be
zeroed immediately before the thread exits. Make use of this instead of
zeroing freeWait from the exitThread assembly and using hacks like switching
to a static stack, so this works on 386.
Change-Id: I3ec5ead82b6496404834d148f713794d5d9da723
Reviewed-on: https://go-review.googlesource.com/105055
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Originally, scalar values were directly stored within interface values
as long as they fit into a pointer-sized slot of memory. And since
interface method calls always pass the full pointer-sized value as the
receiver argument, value-narrowing wrappers were necessary to adapt to
the calling convention for methods with smaller receiver types.
However, for precise garbage collection, we now only store actual
pointers within interface values, so these wrappers are no longer
necessary.
Passes toolstash-check.
Change-Id: I5303bfeb8d0f11db619b5a5d06b37ac898588670
Reviewed-on: https://go-review.googlesource.com/104875
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Mostly replacing C-Style loops with range expressions, but also other
simplifications like the introduction of writeBool and unindenting some
code.
Passes toolstash -cmp on std cmd.
Change-Id: I799bccd4e5d411428dcf122b8588a564a9217e7c
Reviewed-on: https://go-review.googlesource.com/104936
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
Reviewed-by: Ilya Tocar <ilya.tocar@intel.com>
VEX constants were used when instructions were added by hand.
Now all VEX-encoded instructions are auto-generated by x86avxgen,
so there is no need for those anymore.
Change-Id: Ida63e5e23a8b819b15f61ac98980dec45a21617c
Reviewed-on: https://go-review.googlesource.com/104775
Run-TryBot: Iskander Sharipov <iskander.sharipov@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Fixes golint receiver name complaints.
We can't go with "a" name as it sometimes is used for obj.Addr args.
Change-Id: I66556f4e3dc42cfaaa4db3ed7772fa6756ea9a9b
Reviewed-on: https://go-review.googlesource.com/104796
Run-TryBot: Iskander Sharipov <iskander.sharipov@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
While at it, also simplify a couple of switches.
Doesn't pass toolstash -cmp on std cmd, because orderBlock(&n2.Nbody) is
moved further down to the n3 loop.
Change-Id: I20a2a6c21eb9a183a59572e0fca401a5041fc40a
Reviewed-on: https://go-review.googlesource.com/104416
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Since it's been reliably failing on one of the linux-arm builders
(arm5spacemonkey) for a long time.
Updates #24221.
Change-Id: I8fccc7e16631de497ccc2c285e510a110a93ad95
Reviewed-on: https://go-review.googlesource.com/104535
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Inl, Inldcl, and InlCost are only applicable to functions with bodies
that can be inlined, so pull them out into a separate Inline type to
make understanding them easier.
A side benefit is that we can check if a function can be inlined by
just checking if n.Func.Inl is non-nil, which simplifies handling of
empty function bodies.
While here, remove some unnecessary Curfn twiddling, and make imported
functions use Inl.Dcl instead of Func.Dcl for consistency for local
functions.
Passes toolstash-check.
Change-Id: Ifd4a80349d85d9e8e4484952b38ec4a63182e81f
Reviewed-on: https://go-review.googlesource.com/104756
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
Noticed that we can simply use a []byte slice while investigating
a separate issue. Did the obvious simplification.
Change-Id: I921ebbb42135b5f1a10109236ceb9ae6e94ae7e2
Reviewed-on: https://go-review.googlesource.com/104757
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Stores to auto tmp variables can be hoisted to places
where the line numbers make debugging look "jumpy".
Turning those instructions into ones with is_stmt = 0 in
the DWARF (accomplished by marking ssa nodes with NotStmt)
makes debugging look better while still attributing the
instructions with the correct line number.
The same is true for certain register allocator spills and
reloads.
Change-Id: I97a394eb522d4911cc40b4bf5bf76d3d7221f6c0
Reviewed-on: https://go-review.googlesource.com/98415
Run-TryBot: David Chase <drchase@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
To improve debugging, instructions should be annotated with
DWARF is_stmt. The DWARF default before was is_stmt=1, and
to remove "jumpy" stepping the optimizer was tagging
instructions with a no-position position, which interferes
with the accuracy of profiling information. This allows
that to be corrected, and also allows more "jumpy" positions
to be annotated with is_stmt=0 (these changes were not made
for 1.10 because of worries about further messing up
profiling).
The is_stmt values are placed in a pc-encoded table and
passed through a symbol derived from the name of the
function and processed in the linker alongside its
processing of each function's pc/line tables.
The only change in binary size is in the .debug_line tables
measured with "objdump -h --section=.debug_line go1.test"
For go1.test, these are 2614 bytes larger,
or 0.72% of the size of .debug_line,
or 0.025% of the file size.
This will increase in proportion to how much the is_stmt
flag is used (toggled).
Change-Id: Ic1f1aeccff44591ad0494d29e1a0202a3c506a7a
Reviewed-on: https://go-review.googlesource.com/93664
Run-TryBot: David Chase <drchase@google.com>
Reviewed-by: Heschi Kreinick <heschi@google.com>
Add IsStmt information to src.lico so that suitable lines
for breakpoints (or not) can be noted, eventually for
communication to the debugger via the linker and DWARF.
The expectation is that the front end will apply statement
boundary marks because it has best information about the
input, and the optimizer will attempt to preserve these.
The exact method for placing these marks is still TBD;
ideally stopping "at" line N in unoptimized code will occur
at a point where none of the side effects of N have occurred
and all of the inputs for line N can still be observed.
The optimizer will work with the same markings supplied
for unoptimized code.
It is a goal that non-optimizing compilation should conserve
statement marks.
The optimizer will also use the not-a-statement annotation
to indicate instructions that have a line number (for
profiling purposes) but should not be the target of
debugger step, next, or breakpoints. Because instructions
marked as statements are sometimes removed, a third value
indicating that a position (instruction) can serve as a
statement if the optimizer removes the current instruction
marked as a statement for the same line. The optimizer
should attempt to conserve statement marks, but it is not
a bug if some are lost.
Includes changes to html output for GOSSAFUNC to indicate
not-default is-a-statement with bold and not-a-statement
with strikethrough.
Change-Id: Ia22c9a682f276e2ca2a4ef7a85d4b6ebf9c62b7f
Reviewed-on: https://go-review.googlesource.com/93663
Run-TryBot: David Chase <drchase@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
The go/printer (and thus gofmt) uses a heuristic to determine
whether to break alignment between elements of an expression
list which is spread across multiple lines. The heuristic only
kicked in if the entry sizes (character length) was above a
certain threshold (20) and the ratio between the previous and
current entry size was above a certain value (4).
This heuristic worked reasonably most of the time, but also
led to unfortunate breaks in many cases where a single entry
was suddenly much smaller (or larger) then the previous one.
The behavior of gofmt was sufficiently mysterious in some of
these situations that many issues were filed against it.
The simplest solution to address this problem is to remove
the heuristic altogether and have a programmer introduce
empty lines to force different alignments if it improves
readability. The problem with that approach is that the
places where it really matters, very long tables with many
(hundreds, or more) entries, may be machine-generated and
not "post-processed" by a human (e.g., unicode/utf8/tables.go).
If a single one of those entries is overlong, the result
would be that the alignment would force all comments or
values in key:value pairs to be adjusted to that overlong
value, making the table hard to read (e.g., that entry may
not even be visible on screen and all other entries seem
spaced out too wide).
Instead, we opted for a slightly improved heuristic that
behaves much better for "normal", human-written code.
1) The threshold is increased from 20 to 40. This disables
the heuristic for many common cases yet even if the alignment
is not "ideal", 40 is not that many characters per line with
todays screens, making it very likely that the entire line
remains "visible" in an editor.
2) Changed the heuristic to not simply look at the size ratio
between current and previous line, but instead considering the
geometric mean of the sizes of the previous (aligned) lines.
This emphasizes the "overall picture" of the previous lines,
rather than a single one (which might be an outlier).
3) Changed the ratio from 4 to 2.5. Now that we ignore sizes
below 40, a ratio of 4 would mean that a new entry would have
to be 4 times bigger (160) or smaller (10) before alignment
would be broken. A ratio of 2.5 seems more sensible.
Applied updated gofmt to all of src and misc. Also tested
against several former issues that complained about this
and verified that the output for the given examples is
satisfactory (added respective test cases).
Some of the files changed because they were not gofmt-ed
in the first place.
For #644.
For #7335.
For #10392.
(and probably more related issues)
Fixes#22852.
Change-Id: I5e48b3d3b157a5cf2d649833b7297b33f43a6f6e
The trace viewer interprets the slice as a non-terminating
time interval which is quite opposit to what trace records indicate
(i.e., almostly immediately terminating time interval).
As observed in the issue #24663 this can result in quite misleading
visualization of the trace.
Work around the trace viewer's issue by setting a small value
(0.0001usec) as the duration if the time interval is not positive.
Change-Id: I1c2aac135c194d0717f5c01a98ca60ffb14ef45c
Reviewed-on: https://go-review.googlesource.com/104716
Reviewed-by: Heschi Kreinick <heschi@google.com>
This mode is similar to the default traceview mode where the execution
trace is presented in P-oriented way. Each row represents a P, and each
slice represents the time interval of a goroutine's execution on the P.
The difference is that, in this mode, only the execution of goroutines
involved in the specified task is highlighted, and other goroutine
execution or events are greyed out. So, users can focus on how a task is
executed while considering other affecting conditions such as other
goroutines, network events, or process scheduling.
Example: https://user-images.githubusercontent.com/4999471/38116793-a6f995f0-337f-11e8-8de9-88eec2f2c497.png
Here, for a while the program remained idle after the first burst of
activity related to the task because all other goroutines were also
being blocked or waiting for events, or no incoming network traffic
(indicated by the lack of any network activity). This is a bit hard to
discover when the usual task-oriented view (/trace?taskid=<taskid>)
mode.
Also, it simplifies the traceview generation mode logic.
/trace ---> 0
/trace?goid ---> modeGoroutineOriented
/trace?taskid ---> modeGoroutineOriented|modeTaskOriented
/trace?focustask ---> modeTaskOriented
Change-Id: Idcc0ae31b708ddfd19766f4e26ee7efdafecd3a5
Reviewed-on: https://go-review.googlesource.com/103555
Run-TryBot: Hyang-Ah Hana Kim <hyangah@gmail.com>
Reviewed-by: Heschi Kreinick <heschi@google.com>
Currently, the runtime falls back to asking for any address the OS can
offer for the heap when it runs out of hint addresses. However, the
race detector assumes the heap lives in [0x00c000000000,
0x00e000000000), and will fail in a non-obvious way if we go outside
this region.
Fix this by actively throwing a useful error if we run out of heap
hints in race mode.
This problem is currently being triggered by TestArenaCollision, which
intentionally triggers this fallback behavior. Fix the test to look
for the new panic message in race mode.
Fixes#24670.
Updates #24133.
Change-Id: I57de6d17a3495dc1f1f84afc382cd18a6efc2bf7
Reviewed-on: https://go-review.googlesource.com/104717
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
On darwin, only writable symbol is exported
(cmd/link/internal/ld/macho.go:/machoShouldExport).
For plugin to work correctly, global variables, including
runtime.framepointer_enabled which is set by the linker, need
to be exported when dynamic linking. Put it in DATA so it is
exported. Also in Go it is defined as a var, which is not
read-only.
While here, do the same for runtime.goarm.
Fixes#24653.
Change-Id: I9d1b7d5a648be17103d20b97be65a901cb69f5a2
Reviewed-on: https://go-review.googlesource.com/104715
Run-TryBot: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Variables can be declared and shadowing is supported, but modifying
existing variables via assignments was not available.
This meant that modifying a variable from a nested block was not
possible:
{{ $v := "init" }}
{{ if true }}
{{ $v := "changed" }}
{{ end }}
v: {{ $v }} {{/* "init" */}}
Introduce the "=" assignment token, such that one can now do:
{{ $v := "init" }}
{{ if true }}
{{ $v = "changed" }}
{{ end }}
v: {{ $v }} {{/* "changed" */}}
To avoid confusion, rename PipeNode.Decl to PipeNode.Vars, as the
variables may not always be declared after this change. Also change a
few other names to better reflect the added ambiguity of variables in
pipelines.
Modifying the text/template/parse package in a backwards incompatible
manner is acceptable, given that the package godoc clearly states that
it isn't intended for general use. It's the equivalent of an internal
package, back when internal packages didn't exist yet.
To make the changes to the parse package sit well with the cmd/api test,
update except.txt with the changes that we aren't worried about.
Fixes#10608.
Change-Id: I1f83a4297ee093fd45f9993cebb78fc9a9e81295
Reviewed-on: https://go-review.googlesource.com/84480
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Rob Pike <r@golang.org>
If A's external test package imports B, which imports A, and A's
internal test code adds something to A that invalidates anything in A's
export data, then we need to build B against the test-augmented version
of A before using it to build A's external test package.
https://golang.org/cl/92215 taught 'go test' to do this rebuilding
properly, but 'go vet' was not taught the same trick when it learned to
vet test packages in https://golang.org/cl/87636. This commit moves the
necessary logic into the load.TestPackagesFor function so it can be
shared by 'go test' and 'go vet'.
Fixes#23701.
Change-Id: I1086d447eca02933af53de693384eac99a08d9bd
Reviewed-on: https://go-review.googlesource.com/104315
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
And delete them from asm_test.
Change-Id: Id533130470da9176a401cb94972f626f43a62148
Reviewed-on: https://go-review.googlesource.com/103656
Run-TryBot: Alberto Donizetti <alb.donizetti@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Giovanni Bajo <rasky@develer.com>
Add the following helpers in lookup_windows.go:
1) lookupGroupName() is used to obtain the SID of a group based
on name.
2) listGroupsForUsernameAndDomain() uses NetUserGetLocalGroups()
as a WINAPI backend to obtain the list of local groups for this
user.
3) lookupUserPrimaryGroup() is now used to populate the User.Gid
field when looking up a user by name.
Implement listGroups(), lookupGroupId(), lookupGroup() and no longer
return unimplemented errors.
Do not skip Windows User.Gid tests in user_test.go.
Change-Id: I81fd41b406da51f9a4cb24e50d392a333df81141
GitHub-Last-Rev: d1448fd55d
GitHub-Pull-Request: golang/go#24222
Reviewed-on: https://go-review.googlesource.com/98137
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Make the StackCopyNoCache test easier to read.
Add a StackCopyPtr test that actually has some pointers
that need adjusting.
Change-Id: I5b07c26f40cb485c9de97ed63fac89a9e6f36650
Reviewed-on: https://go-review.googlesource.com/104195
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
The logic in addBranchRestrictions didn't allow to correctly
model OpIs(Slice)Bound for signed domain, and it was also partly
implemented within addRestrictions.
Thanks to the previous changes, it is now possible to handle
the negative conditions correctly, so that we can learn
both signed/LT + unsigned/LT on the positive side, and
signed/GE + unsigned/GE on the negative side (but only if
the index can be proved to be non-negative).
This is able to prove ~50 more slice accesses in std+cmd.
Change-Id: I9858080dc03b16f85993a55983dbc4b00f8491b0
Reviewed-on: https://go-review.googlesource.com/104037
Run-TryBot: Giovanni Bajo <rasky@develer.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
addRestrictions was taking a branch parameter, binding its logic
to that of addBranchRestrictions. Since we will need to use it
for updating the facts table for induction variables, refactor it
to remove the branch parameter.
Passes toolstash -cmp.
Change-Id: Iaaec350a8becd1919d03d8574ffd1bbbd906d068
Reviewed-on: https://go-review.googlesource.com/104036
Run-TryBot: Giovanni Bajo <rasky@develer.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
That "else" was needed due to gc DCE limitations.
Now it's not the case and we can avoid go lint complaints.
(See #23521 and https://golang.org/cl/91056.)
There is inlining test for bigEndianWord, so if test
is passing, no performance regression should occur.
Change-Id: Id84d63f361e5e51a52293904ff042966c83c16e9
Reviewed-on: https://go-review.googlesource.com/104555
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
Floating point test instructions allow special cases (NaN, ±∞ and
a few other useful properties) to be checked directly.
This CL adds the following instructions to the assembler:
* LTEBR - load and test (float32)
* LTDBR - load and test (float64)
* TCEB - test data class (float32)
* TCDB - test data class (float64)
Note that I have only added immediate versions of the 'test data
class' instructions for now as that's the only case I think the
compiler will use.
Change-Id: I3398aab2b3a758bf909bd158042234030c8af582
Reviewed-on: https://go-review.googlesource.com/104457
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Performance optimization for the internals of the Regexp type. This adds
no features and has no user-visible impact beyond performance. Copy now
shares the cache, so memory usage for programs that use Copy a lot
should go down; Copy has effectively become a no-op.
The before v. after benchmark results show a lot of noise from run to
run, but there's a clear improvement to the Shared case and no detriment
to the Copied case.
BenchmarkMatchParallelShared-4 361 77.9 -78.42%
BenchmarkMatchParallelCopied-4 70.3 72.2 +2.70%
Macro benchmarks show that the lock contention in Regexp is gone, and my
server is now able to scale linearly 2.5x times more than before (and I
only stopped there because I ran out of CPU in my test machine).
Fixes#24411
Change-Id: Ib33abff2802f27599f5d09084775e95b54e3e1d7
Reviewed-on: https://go-review.googlesource.com/101715
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Provide the fixed size from the key pair.
Change-Id: I365c8d0f7d915229ef089e46458d4c83273fc648
Reviewed-on: https://go-review.googlesource.com/103876
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
When making a shallow copy of a node, various methods were used,
including calling nod(OXXX, nil, nil) and then overwriting it, or
"n1 := *n" and then using &n1.
Add a copy method instead, simplifying all of those and making them
consistent.
Passes toolstash -cmp on std cmd.
Change-Id: I3f3fc88bad708edc712bf6d87214cda4ddc43b01
Reviewed-on: https://go-review.googlesource.com/72710
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
prove used a complex logic when trying to prove branch conditions:
tryPushBranch() was sometimes leaving a checkpoint on the factsTable,
sometimes not, and the caller was supposed to check the return value
to know what to do.
Since we're going to make the prove descend logic a little bit more
complex by adding also induction variables, simplify the tryPushBranch
logic, by removing any factsTable checkpoint handling from it.
Passes toolstash -cmp.
Change-Id: Idfb1703df8a455f612f93158328b36c461560781
Reviewed-on: https://go-review.googlesource.com/104035
Run-TryBot: Giovanni Bajo <rasky@develer.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Previously, n.Pos was reassigned to lineno when declare was called,
which might not match where the identifier actually appeared in the
source. This caused a loss of position precision for function
parameters (which were all declared at the last parameter's position),
and required some clumsy workarounds in bimport.go.
This CL changes declare to leave n.Pos alone and also fixes a few
places where n.Pos was not being set correctly.
Change-Id: Ibe5b5fd30609c684367207df701f9a1bfa82867f
Reviewed-on: https://go-review.googlesource.com/104275
Reviewed-by: Robert Griesemer <gri@golang.org>
Don't report errors if we don't have a correct type switch
guard; instead ignore it and leave it to the type-checker
to report the error. This leads to better error messages
concentrating on the type switch guard rather than errors
around (confusing) syntactic details.
Also clean up some code setting up AssertExpr (they never
have a nil Type field) and remove some incorrect TODOs.
Fixes#24470.
Change-Id: I69512f36e0417e3b5ea9c8856768e04b19d654a8
Reviewed-on: https://go-review.googlesource.com/103615
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
s/Thearch/thearch/
This reduces the amount of exported global variables,
which in turn could make it easier to refactor them later.
Also updated somewhat vague comment about ld.Thearch.
There is no need for Thearch to be exported as Archinit is
called by ld.Main.
Updates #22095
Change-Id: I266b291f6eac0165f70c51964738206e066cea08
Reviewed-on: https://go-review.googlesource.com/103878
Run-TryBot: Iskander Sharipov <iskander.sharipov@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
