]> Cypherpunks repositories - gostls13.git/log
gostls13.git
4 years agogo/doc: avoid panic on references to functions with no body
Norman B. Lancaster [Tue, 30 Mar 2021 17:30:45 +0000 (17:30 +0000)]
go/doc: avoid panic on references to functions with no body

This change guards a call to ast.Inspect with a nil check on the first
argument. This avoids a panic when inspecting a reference to a function
with a nil body. This can only happen when a function body is defined outside Go.

Fixes #42706

Change-Id: I91bc607b24b6224920c24cfd07e76ce7737a98d4
GitHub-Last-Rev: 08072b9ce5c1fd4ee77eba6f1acc0a84e838ad7b
GitHub-Pull-Request: golang/go#43011
Reviewed-on: https://go-review.googlesource.com/c/go/+/275516
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Trust: Daniel Martí <mvdan@mvdan.cc>
Trust: Emmanuel Odeke <emmanuel@orijtech.com>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Go Bot <gobot@golang.org>

4 years agoreflect: cache IsVariadic calls in Call
Ignacio Hagopian [Tue, 30 Mar 2021 17:18:06 +0000 (17:18 +0000)]
reflect: cache IsVariadic calls in Call

These calls are cacheable, so do that to avoid doing extra work.

This opportunity was discovered while taking a look at a CPU profile
while investigating #7818.

I added a BenchmarkCallMethod, which is similar to BechmarkCall but
for a method receiver.

Benchmark results, including the new BenchmarkCallMethod:

name                       old time/op    new time/op    delta
Call-16                      22.0ns ±19%    20.2ns ±17%  -8.08%  (p=0.000 n=40+40)
CallMethod-16                 100ns ± 3%      91ns ± 2%  -9.13%  (p=0.000 n=40+39)
CallArgCopy/size=128-16      15.7ns ± 1%    14.3ns ± 4%  -8.98%  (p=0.000 n=38+37)
CallArgCopy/size=256-16      15.9ns ± 3%    15.0ns ± 5%  -6.12%  (p=0.000 n=39+39)
CallArgCopy/size=1024-16     18.8ns ± 6%    17.1ns ± 6%  -9.03%  (p=0.000 n=38+38)
CallArgCopy/size=4096-16     26.6ns ± 3%    25.2ns ± 4%  -5.19%  (p=0.000 n=39+40)
CallArgCopy/size=65536-16     379ns ± 3%     371ns ± 5%  -2.11%  (p=0.000 n=39+40)

name                       old alloc/op   new alloc/op   delta
Call-16                       0.00B          0.00B         ~     (all equal)
CallMethod-16                 0.00B          0.00B         ~     (all equal)

name                       old allocs/op  new allocs/op  delta
Call-16                        0.00           0.00         ~     (all equal)
CallMethod-16                  0.00           0.00         ~     (all equal)

name                       old speed      new speed      delta
CallArgCopy/size=128-16    8.13GB/s ± 1%  8.92GB/s ± 4%  +9.77%  (p=0.000 n=38+38)
CallArgCopy/size=256-16    16.1GB/s ± 3%  17.1GB/s ± 5%  +6.56%  (p=0.000 n=39+39)
CallArgCopy/size=1024-16   54.6GB/s ± 6%  60.1GB/s ± 5%  +9.93%  (p=0.000 n=38+38)
CallArgCopy/size=4096-16    154GB/s ± 5%   163GB/s ± 4%  +5.63%  (p=0.000 n=40+40)
CallArgCopy/size=65536-16   173GB/s ± 3%   177GB/s ± 5%  +2.18%  (p=0.000 n=39+40)

Updates #7818.

Change-Id: I94f88811ea9faf3dc2543984a13b360b5db66a4b
GitHub-Last-Rev: 9bbaa1854aa32cade905eddb05737e6224c372a1
GitHub-Pull-Request: golang/go#43475
Reviewed-on: https://go-review.googlesource.com/c/go/+/281252
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Keith Randall <khr@golang.org>
Trust: Daniel Martí <mvdan@mvdan.cc>
Trust: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Go Bot <gobot@golang.org>

4 years agocmd/go: fix documentation on how to create new go.mod file
Shulhan [Sun, 28 Mar 2021 16:04:33 +0000 (23:04 +0700)]
cmd/go: fix documentation on how to create new go.mod file

The correct command to create new go.mod file should be 'go mod init',
not 'go help init'.

Change-Id: I1150621987d989997f8b75e6a13fe96423a11cf3
Reviewed-on: https://go-review.googlesource.com/c/go/+/305289
Reviewed-by: Jay Conrod <jayconrod@google.com>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Trust: Jay Conrod <jayconrod@google.com>
Trust: Daniel Martí <mvdan@mvdan.cc>
Run-TryBot: Jay Conrod <jayconrod@google.com>
TryBot-Result: Go Bot <gobot@golang.org>

4 years agocmd/compile: emit writebarriers in specified ABI
David Chase [Mon, 29 Mar 2021 22:46:37 +0000 (18:46 -0400)]
cmd/compile: emit writebarriers in specified ABI

old code was always ABI0, new code tracks the default
this may cause some write barrier removals to fail to fire

Updates #40724.

Change-Id: I656bdd5511c5bd6ee6e021999e30d842a6b9f0a5
Reviewed-on: https://go-review.googlesource.com/c/go/+/305671
Trust: David Chase <drchase@google.com>
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
4 years agocmd/compile/internal/noder: limit the number of goroutine
hqpko [Tue, 30 Mar 2021 14:53:24 +0000 (14:53 +0000)]
cmd/compile/internal/noder: limit the number of goroutine

Change-Id: I7f05108fea351a4895dde6090bf86beb822f2c28
GitHub-Last-Rev: e4a2d3a11aa258f2ddf5e39ab4769883f94e3dab
GitHub-Pull-Request: golang/go#45289
Reviewed-on: https://go-review.googlesource.com/c/go/+/305869
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>

4 years agoruntime: non-strict InlTreeIndex lookup in Frames.Next
Michael Pratt [Thu, 11 Mar 2021 17:28:45 +0000 (12:28 -0500)]
runtime: non-strict InlTreeIndex lookup in Frames.Next

When using cgo, some of the frames can be provided by cgoTraceback, a
cgo-provided function to generate C tracebacks. Unlike Go tracebacks,
cgoTraceback has no particular guarantees that it produces valid
tracebacks.

If one of the (invalid) frames happens to put the PC in the alignment
region at the end of a function (filled with int 3's on amd64), then
Frames.Next will find a valid funcInfo for the PC, but pcdatavalue will
panic because PCDATA doesn't cover this PC.

Tolerate this case by doing a non-strict PCDATA lookup. We'll still show
a bogus frame, but at least avoid throwing.

Fixes #44971

Change-Id: I9eed728470d6f264179a7615bd19845c941db78c
Reviewed-on: https://go-review.googlesource.com/c/go/+/301369
Trust: Michael Pratt <mpratt@google.com>
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
4 years agocrypto/elliptic: fix some typos
chenjie [Tue, 30 Mar 2021 07:23:53 +0000 (07:23 +0000)]
crypto/elliptic: fix some typos

Change-Id: I1c2900d4e1c0b6108f13c4060d994d966f3e18f3
GitHub-Last-Rev: 3fde453686ab3fdeb77d2f73c09e4a679056adf3
GitHub-Pull-Request: golang/go#45295
Reviewed-on: https://go-review.googlesource.com/c/go/+/305773
Reviewed-by: Emmanuel Odeke <emmanuel@orijtech.com>
Reviewed-by: Ben Shi <powerman1st@163.com>
Run-TryBot: Emmanuel Odeke <emmanuel@orijtech.com>
TryBot-Result: Go Bot <gobot@golang.org>

4 years agocmd/compile: fix creation of named generic types (setting of t.nod)
Dan Scales [Wed, 10 Mar 2021 23:25:21 +0000 (15:25 -0800)]
cmd/compile: fix creation of named generic types (setting of t.nod)

The correct setting of t.nod is needed when exporting types. Make sure
we create instantiated named types correctly so t.nod is set.

New test file interfacearg.go that tests this (by instantiating a type
with an interface). Also has tests for various kinds of method
expressions.

Change-Id: Ia7fd9debd495336b73788af9e35d72331bb7d2b5
Reviewed-on: https://go-review.googlesource.com/c/go/+/305730
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Dan Scales <danscales@google.com>
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
4 years agoruntime: fix typos in comments
Lizzzcai [Mon, 29 Mar 2021 10:51:06 +0000 (18:51 +0800)]
runtime: fix typos in comments

Change-Id: Ia70e8bdc6d2cf1195d7a3b5d33f180ae2db73e29
Reviewed-on: https://go-review.googlesource.com/c/go/+/305369
Reviewed-by: Emmanuel Odeke <emmanuel@orijtech.com>
Reviewed-by: Ben Shi <powerman1st@163.com>
Run-TryBot: Emmanuel Odeke <emmanuel@orijtech.com>
TryBot-Result: Go Bot <gobot@golang.org>

4 years agocmd/compile: update default ABI choices for calls and bodyless fn stack maps
David Chase [Wed, 24 Mar 2021 18:20:12 +0000 (14:20 -0400)]
cmd/compile: update default ABI choices for calls and bodyless fn stack maps

After recent discussion about bodyless functions, their wrappers,
their stack maps, nosplit, and callbacks, I was inspired to go and
be sure that more defaults were sensible.  This may not be all --
currently rtcall is "ABIDefault" which I think is correct, but I
am not 100% certain.

Updates #40724.

Change-Id: I95b14ee0e5952fa53e7fea9f6f5192358aa24f23
Reviewed-on: https://go-review.googlesource.com/c/go/+/304549
Trust: David Chase <drchase@google.com>
Run-TryBot: David Chase <drchase@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
4 years agocmd/compile: fix various small bugs related to type lists
Dan Scales [Mon, 29 Mar 2021 15:28:01 +0000 (08:28 -0700)]
cmd/compile: fix various small bugs related to type lists

Fix various small bugs related to delaying transformations due to type
params. Most of these relate to the need to delay a transformation when
an argument of an expression or statement has a type parameter that has
a structural constraint. The structural constraint implies the operation
should work, but the transformation can't happen until the actual value
of the type parameter is known.

 - delay transformations for send statements and return statements if
   any args/values have type params.

 - similarly, delay transformation of a call where the function arg has
   type parameters. This is mainly important for the case where the
   function arg is a pure type parameter, but has a structural
   constraint that requires it to be a function. Move the setting of
   n.Use to transformCall(), since we may not know how many return
   values there are until then, if the function arg is a type parameter.

 - set the type of unary expressions from the type2 type (as we do with
   most other expressions), since that works better with expressions
   with type params.

 - deal with these delayed transformations in subster.node() and convert
   the CALL checks to a switch statement.

 - make sure ir.CurFunc is set properly during stenciling, including
   closures (needed for transforming return statements during
   stenciling).

New test file typelist.go with tests for these cases.

Change-Id: I1b82f949d8cec47d906429209e846f4ebc8ec85e
Reviewed-on: https://go-review.googlesource.com/c/go/+/305729
Trust: Dan Scales <danscales@google.com>
Trust: Robert Griesemer <gri@golang.org>
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
4 years agoruntime: init plan9 hashkey by time
Meng Zhuo [Mon, 29 Mar 2021 07:10:50 +0000 (15:10 +0800)]
runtime: init plan9 hashkey by time

Maphash requires non-zero integer for initial hashkey

Fixes #45090

Change-Id: Ie567f648c19e81cddc8e72a1c64809fbf52df188
Reviewed-on: https://go-review.googlesource.com/c/go/+/303969
Trust: Meng Zhuo <mzh@golangcn.org>
Run-TryBot: Meng Zhuo <mzh@golangcn.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Richard Miller <millerresearch@gmail.com>
Reviewed-by: Keith Randall <khr@golang.org>
4 years agocmd/compile: wrap defer/go call with results
Cherry Zhang [Mon, 29 Mar 2021 18:17:18 +0000 (14:17 -0400)]
cmd/compile: wrap defer/go call with results

CL 298669 implemented wrapping for defer/go calls so the function
being called with defer or go statement has no arguments. This
simplifies the compiler and the runtime, especially with the
new ABI.

Currently, it does not wrap functions that has no arguments but
only results. For defer/go calls, the results are not used. But
the runtime needs to allocate stack space for the callee to store
the results. Wrapping functions with results makes the runtime
simpler.

TODO: maybe not wrap if all results are in registers.

Updates #40724.

Change-Id: I74d2f4db1cbf9979afbcd846facb30d11d72ab23
Reviewed-on: https://go-review.googlesource.com/c/go/+/305550
Trust: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
4 years agocmd/compile: be sure to wrap defer/go calls with arguments
Cherry Zhang [Thu, 25 Mar 2021 17:24:16 +0000 (13:24 -0400)]
cmd/compile: be sure to wrap defer/go calls with arguments

CL 298669 implemented wrapping for defer/go calls so the function
being called with defer or go statement has no arguments. This
simplifies the compiler and the runtime, especially with the
new ABI.

If the called function does not have any argument, we don't need
to wrap. But the code missed the cases of method receiver, as
well as some apparent argumentless builtin calls which may later
be rewritten to having arguments (e.g. recover). This CL makes
sure to wrap those cases. Also add a check to ensure that go and
defer calls are indeed argumentless.

Handle "defer recover()" specially, as recover() is lowered to
runtime.gorecover(FP) where FP is the frame's FP. FP needs to be
evaluated before wrapping.

Updates #40724.

Change-Id: I2758b6c69ab6aa02dd588441a457fe28ddd0d5a5
Reviewed-on: https://go-review.googlesource.com/c/go/+/304771
Trust: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
4 years agocmd/compile: check deferred nil interface call before wrapping it
Cherry Zhang [Mon, 29 Mar 2021 17:44:08 +0000 (13:44 -0400)]
cmd/compile: check deferred nil interface call before wrapping it

Currently, for "defer i.M()" if i is nil it panics at the point of
defer statement, not when deferred function is called. We need to
do the nil check before wrapping it.

Updates #40724.

Change-Id: I62c669264668991f71999e2cf4610a9066247f9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/305549
Trust: Cherry Zhang <cherryyz@google.com>
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
4 years agoruntime: add GC testing helpers for regabi signature fuzzer
Austin Clements [Wed, 24 Mar 2021 14:45:20 +0000 (10:45 -0400)]
runtime: add GC testing helpers for regabi signature fuzzer

This CL adds a set of helper functions for testing GC interactions.
These are intended for use in the regabi signature fuzzer, but are
generally useful for GC tests, so we make them generally available to
runtime tests.

These provide:

1. An easy way to force stack movement, for testing stack copying.

2. A simple and robust way to check the reachability of a set of
pointers.

3. A way to check what general category of memory a pointer points to,
mostly so tests can make sure they're testing what they mean to.

For #40724, but generally useful.

Change-Id: I15d33ccb3f5a792c0472a19c2cc9a8b4a9356a66
Reviewed-on: https://go-review.googlesource.com/c/go/+/305330
Trust: Austin Clements <austin@google.com>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Michael Pratt <mpratt@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
4 years agoruntime: abstract specials list iteration
Austin Clements [Mon, 22 Mar 2021 19:00:22 +0000 (15:00 -0400)]
runtime: abstract specials list iteration

The specials processing loop in mspan.sweep is about to get more
complicated and I'm too allergic to list manipulation to open code
more of it there.

Change-Id: I767a0889739da85fb2878fc06a5c55b73bf2ba7d
Reviewed-on: https://go-review.googlesource.com/c/go/+/305551
Trust: Austin Clements <austin@google.com>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Michael Pratt <mpratt@google.com>
4 years agocmd/internal/obj/ppc64: remove bogus MOVBU optab entry
Paul E. Murphy [Wed, 24 Mar 2021 20:50:54 +0000 (15:50 -0500)]
cmd/internal/obj/ppc64: remove bogus MOVBU optab entry

This was missed in https://golang.org/cl/303329 . It is another
impossible usage of MOVBU as a load like "MOVBU 0(rX), rY, rZ" or
"MOVBU rX(rB), rY, rZ".

Change-Id: Ib3dd984b6424907498ed65b798649f0b990d50a7
Reviewed-on: https://go-review.googlesource.com/c/go/+/304471
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
4 years agogo/types: remove use of ioutil (cleanup)
Robert Griesemer [Mon, 29 Mar 2021 20:31:14 +0000 (13:31 -0700)]
go/types: remove use of ioutil (cleanup)

Change-Id: I0f9437953cb994c6802efee92167702daacf000a
Reviewed-on: https://go-review.googlesource.com/c/go/+/305575
Trust: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
4 years agoruntime: fix some typos
徐志伟 [Mon, 29 Mar 2021 18:22:15 +0000 (18:22 +0000)]
runtime: fix some typos

Change-Id: I31f2081eb7c30a9583f479f9194e636fe721b9b3
GitHub-Last-Rev: d09f5fbdc5785dc3963b22ad75309740e0de258e
GitHub-Pull-Request: golang/go#45278
Reviewed-on: https://go-review.googlesource.com/c/go/+/305231
Reviewed-by: Emmanuel Odeke <emmanuel@orijtech.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Emmanuel Odeke <emmanuel@orijtech.com>
TryBot-Result: Go Bot <gobot@golang.org>

4 years agocmd/compile: restructure ABI wrapper generation, export ABI
Austin Clements [Thu, 25 Mar 2021 23:50:08 +0000 (19:50 -0400)]
cmd/compile: restructure ABI wrapper generation, export ABI

This CL restructures how we track function ABIs and generate ABI
wrappers in the compiler and adds import/export of ABIs across package
boundaries.

Currently, we start by tracking definition and referencing ABIs in two
global maps and eventually move some of this information into the
LSyms for functions. This complicates a lot of the existing code for
handling wrappers and makes it particularly hard to export ABI
information across packages. This change is built around instead
recording this information on the ir.Func.

First, this change replaces the global ABI def/ref maps with a type,
which makes the data flow and lifetime of this information clear in
gc.Main. These are populated during flag parsing.

Then, early in the front-end, we loop over all ir.Funcs to 1. attach
ABI def/ref information to the ir.Funcs and 2. create new ir.Funcs for
ABI wrappers. Step 1 is slightly subtle because the information is
keyed by linker symbol names, so we can't simply look things up in the
compiler's regular symbol table.

By generating ABI wrappers early in the front-end, we decouple this
step from LSym creation, which makes LSym creation much simpler (like
it was before ABI wrappers). In particular, LSyms for wrappers are now
created at the same time as all other functions instead of by
makeABIWrapper, which means we're back to the simpler, old situation
where InitLSym was the only thing responsible for constructing
function LSyms. Hence, we can restore the check that InitLSym is
called exactly once per function.

Attaching the ABI information to the ir.Func has several follow-on
benefits:

1. It's now easy to include in the export info. This enables direct
cross-package cross-ABI calls, which are important for the performance
of calling various hot assembly functions (e.g., internal/bytealg.*).
This was really the point of this whole change.

2. Since all Funcs, including wrappers, now record their definition
ABI, callTargetLSym no longer needs to distinguish wrappers from
non-wrappers, so it's now nearly trivial (it would be completely
trivial except that it has to work around a handful of cases where
ir.Name.Func is nil).

The simplification of callTargetLSym has one desirable but potentially
surprising side-effect: the compiler will now generate direct calls to
the definition ABI even when ABI wrappers are turned off. This is
almost completely unnoticeable except that cmd/internal/obj/wasm looks
for the call from runtime.deferreturn (defined in Go) to
runtime.jmpdefer (defined in assembly) to compile is specially. That
now looks like a direct call to ABI0 rather than going through the
ABIInternal alias.

While we're in here, we also set up the structures to support more
than just ABI0 and ABIInternal and add various additional consistency
checks all around.

Performance-wise, this reduces the overhead induced by wrappers from
1.24% geomean (on Sweet) to 0.52% geomean, and reduces the number of
benchmarks impacts >2% from 5 to 3. It has no impact on compiler speed.

Impact of wrappers before this change:

name                                old time/op  new time/op  delta
BiogoIgor                            15.8s ± 2%   15.8s ± 1%    ~     (p=0.863 n=25+25)
BiogoKrishna                         18.3s ± 6%   18.1s ± 7%  -1.39%  (p=0.015 n=25+25)
BleveIndexBatch100                   5.88s ± 3%   6.04s ± 6%  +2.72%  (p=0.000 n=25+25)
BleveQuery                           6.42s ± 1%   6.76s ± 1%  +5.31%  (p=0.000 n=24+24)
CompileTemplate                      245ms ± 3%   250ms ± 6%    ~     (p=0.068 n=22+25)
CompileUnicode                      93.6ms ± 2%  93.9ms ± 5%    ~     (p=0.958 n=22+25)
CompileGoTypes                       1.60s ± 2%   1.59s ± 2%    ~     (p=0.115 n=24+24)
CompileCompiler                      104ms ± 4%   104ms ± 3%    ~     (p=0.453 n=22+25)
CompileSSA                           11.0s ± 2%   11.0s ± 1%    ~     (p=0.789 n=24+25)
CompileFlate                         153ms ± 2%   153ms ± 1%    ~     (p=0.055 n=21+20)
CompileGoParser                      229ms ± 2%   230ms ± 2%    ~     (p=0.305 n=21+22)
CompileReflect                       585ms ± 5%   582ms ± 3%    ~     (p=0.365 n=25+25)
CompileTar                           211ms ± 1%   211ms ± 3%    ~     (p=0.592 n=20+22)
CompileXML                           282ms ± 3%   281ms ± 2%    ~     (p=0.937 n=22+23)
CompileStdCmd                        13.7s ± 3%   13.6s ± 2%    ~     (p=0.700 n=25+25)
FoglemanFauxGLRenderRotateBoat       8.67s ± 1%   8.78s ± 1%  +1.30%  (p=0.000 n=25+25)
FoglemanPathTraceRenderGopherIter1   20.5s ± 2%   20.9s ± 2%  +1.85%  (p=0.000 n=25+25)
GopherLuaKNucleotide                 30.1s ± 2%   31.1s ± 2%  +3.38%  (p=0.000 n=25+25)
MarkdownRenderXHTML                  246ms ± 5%   250ms ± 1%  +1.42%  (p=0.002 n=25+23)
Tile38WithinCircle100kmRequest       828µs ± 6%   885µs ± 6%  +6.85%  (p=0.000 n=23+25)
Tile38IntersectsCircle100kmRequest  1.04ms ± 5%  1.10ms ± 7%  +5.63%  (p=0.000 n=25+25)
Tile38KNearestLimit100Request        974µs ± 4%   972µs ± 4%    ~     (p=0.356 n=25+24)
[Geo mean]                           588ms        595ms       +1.24%

(https://perf.golang.org/search?q=upload:20210328.5)

And after this change:

name                                old time/op  new time/op  delta
BiogoIgor                            15.9s ± 1%   15.8s ± 1%  -0.48%  (p=0.008 n=22+25)
BiogoKrishna                         18.4s ± 6%   17.8s ± 6%  -3.55%  (p=0.008 n=25+25)
BleveIndexBatch100                   5.86s ± 3%   5.97s ± 4%  +1.88%  (p=0.001 n=25+25)
BleveQuery                           6.42s ± 1%   6.75s ± 1%  +5.14%  (p=0.000 n=25+25)
CompileTemplate                      246ms ± 5%   245ms ± 2%    ~     (p=0.472 n=23+23)
CompileUnicode                      93.7ms ± 3%  93.5ms ± 2%    ~     (p=0.813 n=22+23)
CompileGoTypes                       1.60s ± 2%   1.60s ± 2%    ~     (p=0.108 n=25+23)
CompileCompiler                      104ms ± 3%   104ms ± 2%    ~     (p=0.845 n=23+23)
CompileSSA                           11.0s ± 2%   11.0s ± 2%    ~     (p=0.525 n=25+25)
CompileFlate                         152ms ± 1%   153ms ± 2%    ~     (p=0.408 n=22+22)
CompileGoParser                      230ms ± 1%   230ms ± 1%    ~     (p=0.363 n=21+23)
CompileReflect                       582ms ± 3%   584ms ± 4%    ~     (p=0.658 n=25+25)
CompileTar                           212ms ± 2%   211ms ± 2%    ~     (p=0.315 n=23+24)
CompileXML                           282ms ± 1%   282ms ± 1%    ~     (p=0.991 n=23+22)
CompileStdCmd                        13.6s ± 2%   13.6s ± 2%    ~     (p=0.699 n=25+24)
FoglemanFauxGLRenderRotateBoat       8.66s ± 1%   8.69s ± 1%  +0.28%  (p=0.002 n=25+24)
FoglemanPathTraceRenderGopherIter1   20.5s ± 3%   20.5s ± 2%    ~     (p=0.407 n=25+25)
GopherLuaKNucleotide                 30.1s ± 2%   31.2s ± 2%  +3.82%  (p=0.000 n=25+25)
MarkdownRenderXHTML                  246ms ± 3%   245ms ± 1%    ~     (p=0.478 n=23+22)
Tile38WithinCircle100kmRequest       820µs ± 4%   856µs ± 5%  +4.39%  (p=0.000 n=24+25)
Tile38IntersectsCircle100kmRequest  1.05ms ± 6%  1.07ms ± 6%  +1.91%  (p=0.014 n=25+25)
Tile38KNearestLimit100Request        970µs ± 4%   970µs ± 3%    ~     (p=0.819 n=22+24)
[Geo mean]                           588ms        591ms       +0.52%

(https://perf.golang.org/search?q=upload:20210328.6)

For #40724.

Change-Id: I1c374e32d4bbc88efed062a1b360017d3642140d
Reviewed-on: https://go-review.googlesource.com/c/go/+/305274
Trust: Austin Clements <austin@google.com>
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>

4 years agocmd/compile: eliminate -abiwraplimit
Austin Clements [Wed, 24 Mar 2021 17:55:31 +0000 (13:55 -0400)]
cmd/compile: eliminate -abiwraplimit

We haven't needed this debugging flag in a while and it's going to
complicate a change to how to generate wrappers. Eliminate it in favor
of just using the objabi.Experiment.RegabiWrappers global toggle.

Updates #40724.

Change-Id: Ieda660ea7a0167ae4e881b396ef556d7c962fe4c
Reviewed-on: https://go-review.googlesource.com/c/go/+/305273
Trust: Austin Clements <austin@google.com>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
4 years agocmd/compile: assert that function values reference ABIInternal
Austin Clements [Wed, 24 Mar 2021 14:32:13 +0000 (10:32 -0400)]
cmd/compile: assert that function values reference ABIInternal

Function values must always point to the ABIInternal entry point of a
function. It wasn't entirely obvious to me we were getting this right,
so this CL adds checks for this.

Updates #40724.

Change-Id: Idd854e996d63d9151c28ec5c9251b690453b1024
Reviewed-on: https://go-review.googlesource.com/c/go/+/305272
Trust: Austin Clements <austin@google.com>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
4 years agocmd/compile: set ir.Name.Func in more cases
Austin Clements [Sat, 27 Mar 2021 14:14:30 +0000 (10:14 -0400)]
cmd/compile: set ir.Name.Func in more cases

ir.Name.Func is non-nil for *almost* all function names. This CL fixes
a few more major cases that leave it nil, though there are still a few
cases left: interface method values, and algorithms generated by
eqFor, hashfor, and hashmem.

We'll need this for mapping from ir.Names to function ABIs shortly.
The remaining cases would be nice to fix, but they're all guaranteed
to be ABIInternal, so we can at least work around them.

For #40724.

Change-Id: Ifcfa781c78899ccea0bf155d80f8cfc27f30351e
Reviewed-on: https://go-review.googlesource.com/c/go/+/305271
Trust: Austin Clements <austin@google.com>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
4 years agocmd/compile: track funcsyms by ir.Name instead of types.Sym
Austin Clements [Wed, 24 Mar 2021 18:28:18 +0000 (14:28 -0400)]
cmd/compile: track funcsyms by ir.Name instead of types.Sym

This is a cleanup to bring funcsym tracking a little closer to the
ir.Func. (I thought I needed this for a later change. That turned out
not to be the case, but it's a nice cleanup.)

Change-Id: I53e692f5d7ba4be56d42d8e0aefc06284cea0661
Reviewed-on: https://go-review.googlesource.com/c/go/+/305270
Trust: Austin Clements <austin@google.com>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
4 years agocmd/compile: update a few stale comments
Austin Clements [Fri, 19 Mar 2021 18:55:23 +0000 (14:55 -0400)]
cmd/compile: update a few stale comments

CL 64811 removed dcopy. Update the comment in types.Sym.

The Russquake moved iexport.go. Update the path to it.

WRAPPER is now also used by ABI wrappers, so update the comment since
it's now more general than method wrappers.

Change-Id: Ie0df61dcef7168f6720838cd5c9a66adf546a44f
Reviewed-on: https://go-review.googlesource.com/c/go/+/305269
Trust: Austin Clements <austin@google.com>
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
4 years agogo/types: remove outdated comment
Robert Griesemer [Mon, 29 Mar 2021 18:32:50 +0000 (11:32 -0700)]
go/types: remove outdated comment

dev.typeparams is not used anymore for active development.

Change-Id: Ic773cbc70e3532375d75b6c6caa31f55f7c733b0
Reviewed-on: https://go-review.googlesource.com/c/go/+/305569
Trust: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
4 years agocmd/compile: remove typechecker calls in varDecl()
Dan Scales [Thu, 25 Mar 2021 19:23:44 +0000 (12:23 -0700)]
cmd/compile: remove typechecker calls in varDecl()

We can now use transformAssign.

The only remaining typechecker calls in the noder2 pass are for
CompLitExpr nodes (OCOMPLIT).

Change-Id: I25671c79cc30749767bb16f84e9f151b943eccd1
Reviewed-on: https://go-review.googlesource.com/c/go/+/305509
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Dan Scales <danscales@google.com>

4 years agocmd/compile/internal/types2: remove 'strict' argument from several methods
Robert Griesemer [Thu, 25 Mar 2021 19:31:25 +0000 (12:31 -0700)]
cmd/compile/internal/types2: remove 'strict' argument from several methods

The value is always 'false'. Brings the code closer in line with go/types.
Follow-up on https://golang.org/cl/304129.

Change-Id: I8bea550033f3187b44e9a54258e0cf642c11c369
Reviewed-on: https://go-review.googlesource.com/c/go/+/304849
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
4 years agocrypto/rsa: fix salt length calculation with PSSSaltLengthAuto
Himanshu Kishna Srivastava [Tue, 16 Mar 2021 17:07:46 +0000 (22:37 +0530)]
crypto/rsa: fix salt length calculation with PSSSaltLengthAuto

When PSSSaltLength is set, the maximum salt length must equal:

    (modulus_key_size - 1 + 7)/8 - hash_length - 2
and for example, with a 4096 bit modulus key, and a SHA-1 hash,
it should be:

     (4096 -1 + 7)/8 - 20 - 2 = 490
Previously we'd encounter this error:

     crypto/rsa: key size too small for PSS signature

Fixes #42741

Change-Id: I18bb82c41c511d564b3f4c443f4b3a38ab010ac5
Reviewed-on: https://go-review.googlesource.com/c/go/+/302230
Reviewed-by: Emmanuel Odeke <emmanuel@orijtech.com>
Reviewed-by: Filippo Valsorda <filippo@golang.org>
Trust: Emmanuel Odeke <emmanuel@orijtech.com>
Run-TryBot: Emmanuel Odeke <emmanuel@orijtech.com>
TryBot-Result: Go Bot <gobot@golang.org>

4 years agocmd/link/internal/ld: use linkerFlagSupported to check -Qunused-arguments
Tobias Klauser [Fri, 26 Mar 2021 10:00:40 +0000 (11:00 +0100)]
cmd/link/internal/ld: use linkerFlagSupported to check -Qunused-arguments

Rather than checking the linker name or its path for the string "clang",
use linkerFlagSupported to determine whether the -Qunused-arguments flag
may be passed to the linker.

Fixes #45241

Change-Id: I4c1e4d4ecba4cf5823e8f39cfda5d20404ebf513
Reviewed-on: https://go-review.googlesource.com/c/go/+/304692
Trust: Tobias Klauser <tobias.klauser@gmail.com>
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
4 years agomath: simplify comparison in FMA when swapping p and z
smasher164 [Thu, 6 Aug 2020 22:37:10 +0000 (18:37 -0400)]
math: simplify comparison in FMA when swapping p and z

Discovered by Junchen Li on CL 246858, the comparison before p and z are
swapped can be simplified from

    pe < ze || (pe == ze && (pm1 < zm1 || (pm1 == zm1 && pm2 < zm2)))

to

    pe < ze || pe == ze && pm1 < zm1

because zm2 is initialized to 0 before the branch.

Change-Id: Iee92d570038df2b0f8941ef6e422a022654ab2d6
Reviewed-on: https://go-review.googlesource.com/c/go/+/247241
Run-TryBot: Akhil Indurti <aindurti@gmail.com>
Run-TryBot: Emmanuel Odeke <emmanuel@orijtech.com>
Trust: Emmanuel Odeke <emmanuel@orijtech.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
4 years agoruntime: fix some typos
qcrao [Sun, 28 Mar 2021 12:17:35 +0000 (12:17 +0000)]
runtime: fix some typos

Change-Id: I18b9508904f19d5aa68355c937c30b5fdf35442c
Reviewed-on: https://go-review.googlesource.com/c/go/+/305249
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Emmanuel Odeke <emmanuel@orijtech.com>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>

4 years agocmd/compile/internal/ssa: strengthen phiopt pass
fanzha02 [Tue, 28 Jul 2020 02:29:06 +0000 (10:29 +0800)]
cmd/compile/internal/ssa: strengthen phiopt pass

The current phiopt pass just transforms the following code
  x := false
  if b { x = true}
into
  x = b

But we find code in runtime.atoi like this:
  neg := false
  if s[0] == '-' {
    neg = true
    s = s[1:]
  }

The current phiopt pass does not covert it into code like:
  neg := s[0] == '-'
  if neg { s = s[1:] }

Therefore, this patch strengthens the phiopt pass so that the
boolean Phi value "neg" can be replaced with a copy of control
value "s[0] == '-'", thereby using "cmp+cset" instead of a branch.

But in some cases even replacing the boolean Phis cannot eliminate
this branch. In the following case, this patch replaces "d" with a
copy of "a<0", but the regalloc pass will insert the "Load {c}"
value into an empty block to split the live ranges, which causes
the branch to not be eliminated.

For example:
  func test(a, b, c int) (bool, int) {
    d := false
    if (a<0) {
      if (b<0) {
        c = c+1
      }
      d = true
    }
    return d, c
  }

The optimized assembly code:
  MOVD "".a(FP), R0
  TBZ $63, R0, 48
  MOVD "".c+16(FP), R1
  ADD $1, R1, R2
  MOVD "".b+8(FP), R3
  CMP ZR, R3
  CSEL LT, R2, R1, R1
  CMP ZR, R0
  CSET LT, R0
  MOVB R0, "".~r3+24(FP)
  MOVD R1, "".~r4+32(FP)
  RET (R30)
  MOVD "".c+16(FP), R1
  JMP 28

The benchmark:

name          old time/op            new time/op            delta
pkg:cmd/compile/internal/ssa goos:linux goarch:arm64
PhioptPass  117783.250000ns +- 1%  117219.111111ns +- 1%   ~  (p=0.074  n=8+9)

Statistical data from compilecmp tool:

compilecmp local/master -> HEAD
local/master (a826f7dc45): debug/dwarf: support DW_FORM_rnglistx aka formRnglistx
HEAD (e57e003c10): cmd/compile/internal/ssa: strengthen phiopt pass

benchstat -geomean  /tmp/2516644532 /tmp/1075915815
completed 50 of 50, estimated time remaining 0s (ETA 7:10PM)
name                      old time/op       new time/op       delta
Template                        554ms _ 3%        553ms _ 3%    ~     (p=0.986 n=49+48)
Unicode                         252ms _ 4%        249ms _ 4%  -1.33%  (p=0.002 n=47+49)
GoTypes                         3.16s _ 3%        3.18s _ 3%  +0.77%  (p=0.022 n=44+48)
Compiler                        257ms _ 4%        258ms _ 4%    ~     (p=0.121 n=50+49)
SSA                             24.2s _ 4%        24.2s _ 5%    ~     (p=0.694 n=49+50)
Flate                           338ms _ 4%        338ms _ 4%    ~     (p=0.592 n=43+46)
GoParser                        506ms _ 3%        507ms _ 3%    ~     (p=0.942 n=49+50)
Reflect                         1.37s _ 4%        1.37s _ 5%    ~     (p=0.408 n=50+50)
Tar                             486ms _ 3%        487ms _ 4%    ~     (p=0.911 n=47+50)
XML                             619ms _ 2%        619ms _ 3%    ~     (p=0.368 n=46+48)
LinkCompiler                    1.29s _31%        1.32s _23%    ~     (p=0.306 n=49+44)
ExternalLinkCompiler            3.39s _10%        3.36s _ 6%    ~     (p=0.311 n=48+46)
LinkWithoutDebugCompiler        846ms _37%        793ms _24%  -6.29%  (p=0.040 n=50+49)
[Geo mean]                      974ms             971ms       -0.36%

name                      old user-time/op  new user-time/op  delta
Template                        910ms _12%        893ms _13%    ~     (p=0.098 n=49+49)
Unicode                         495ms _28%        492ms _18%    ~     (p=0.562 n=50+46)
GoTypes                         4.42s _15%        4.39s _13%    ~     (p=0.684 n=49+50)
Compiler                        419ms _22%        422ms _16%    ~     (p=0.579 n=48+50)
SSA                             36.5s _ 7%        36.6s _ 8%    ~     (p=0.465 n=50+47)
Flate                           521ms _21%        523ms _16%    ~     (p=0.889 n=50+47)
GoParser                        810ms _12%        792ms _15%    ~     (p=0.149 n=50+50)
Reflect                         1.98s _13%        2.02s _13%    ~     (p=0.144 n=47+50)
Tar                             826ms _15%        806ms _19%    ~     (p=0.115 n=49+49)
XML                             988ms _14%       1003ms _14%    ~     (p=0.179 n=50+50)
LinkCompiler                    1.79s _ 8%        1.84s _11%  +2.81%  (p=0.001 n=49+49)
ExternalLinkCompiler            3.69s _ 4%        3.71s _ 3%    ~     (p=0.261 n=50+50)
LinkWithoutDebugCompiler        838ms _10%        827ms _11%    ~     (p=0.323 n=50+48)
[Geo mean]                      1.44s             1.44s       -0.05%

name                      old alloc/op      new alloc/op      delta
Template                       39.0MB _ 1%       39.0MB _ 1%    ~     (p=0.445 n=50+49)
Unicode                        28.5MB _ 0%       28.5MB _ 0%    ~     (p=0.460 n=50+50)
GoTypes                         169MB _ 1%        169MB _ 1%    ~     (p=0.092 n=48+50)
Compiler                       23.4MB _ 1%       23.4MB _ 1%  -0.19%  (p=0.032 n=50+49)
SSA                            1.54GB _ 0%       1.55GB _ 1%  +0.14%  (p=0.001 n=50+50)
Flate                          23.8MB _ 1%       23.8MB _ 2%    ~     (p=0.702 n=49+49)
GoParser                       35.4MB _ 1%       35.4MB _ 1%    ~     (p=0.786 n=50+50)
Reflect                        85.3MB _ 1%       85.3MB _ 1%    ~     (p=0.298 n=50+50)
Tar                            34.6MB _ 2%       34.6MB _ 2%    ~     (p=0.683 n=50+50)
XML                            44.5MB _ 3%       44.0MB _ 2%  -1.05%  (p=0.000 n=50+46)
LinkCompiler                    136MB _ 0%        136MB _ 0%  +0.01%  (p=0.005 n=50+50)
ExternalLinkCompiler            128MB _ 0%        128MB _ 0%    ~     (p=0.179 n=50+50)
LinkWithoutDebugCompiler       84.3MB _ 0%       84.3MB _ 0%  +0.01%  (p=0.006 n=50+50)
[Geo mean]                     70.7MB            70.6MB       -0.07%

name                      old allocs/op     new allocs/op     delta
Template                         410k _ 0%         410k _ 0%    ~     (p=0.606 n=48+49)
Unicode                          310k _ 0%         310k _ 0%    ~     (p=0.674 n=50+50)
GoTypes                         1.81M _ 0%        1.81M _ 0%    ~     (p=0.674 n=50+50)
Compiler                         202k _ 0%         202k _ 0%  +0.02%  (p=0.046 n=50+50)
SSA                             16.3M _ 0%        16.3M _ 0%  +0.10%  (p=0.000 n=50+50)
Flate                            244k _ 0%         244k _ 0%    ~     (p=0.834 n=49+50)
GoParser                         380k _ 0%         380k _ 0%    ~     (p=0.410 n=50+50)
Reflect                         1.08M _ 0%        1.08M _ 0%    ~     (p=0.782 n=48+50)
Tar                              368k _ 0%         368k _ 0%    ~     (p=0.585 n=50+49)
XML                              453k _ 0%         453k _ 0%  -0.01%  (p=0.025 n=49+49)
LinkCompiler                     713k _ 0%         713k _ 0%  +0.01%  (p=0.044 n=50+50)
ExternalLinkCompiler             794k _ 0%         794k _ 0%  +0.01%  (p=0.000 n=50+49)
LinkWithoutDebugCompiler         251k _ 0%         251k _ 0%    ~     (p=0.092 n=47+50)
[Geo mean]                       615k              615k       +0.01%

name                      old maxRSS/op     new maxRSS/op     delta
Template                        37.0M _ 4%        37.2M _ 3%    ~     (p=0.062 n=48+48)
Unicode                         36.9M _ 5%        37.3M _ 4%  +1.10%  (p=0.021 n=50+47)
GoTypes                         94.3M _ 3%        94.9M _ 4%  +0.69%  (p=0.022 n=45+46)
Compiler                        33.4M _ 3%        33.4M _ 5%    ~     (p=0.964 n=49+50)
SSA                              741M _ 3%         738M _ 3%    ~     (p=0.164 n=50+50)
Flate                           28.5M _ 6%        28.8M _ 4%  +1.07%  (p=0.009 n=50+49)
GoParser                        35.0M _ 3%        35.3M _ 4%  +0.83%  (p=0.010 n=50+48)
Reflect                         57.2M _ 6%        57.1M _ 4%    ~     (p=0.815 n=50+49)
Tar                             34.9M _ 3%        35.0M _ 3%    ~     (p=0.134 n=49+48)
XML                             39.5M _ 5%        40.0M _ 3%  +1.35%  (p=0.001 n=50+48)
LinkCompiler                     220M _ 2%         220M _ 2%    ~     (p=0.547 n=49+48)
ExternalLinkCompiler             235M _ 2%         236M _ 2%    ~     (p=0.538 n=47+44)
LinkWithoutDebugCompiler         179M _ 1%         179M _ 1%    ~     (p=0.775 n=50+50)
[Geo mean]                      74.9M             75.2M       +0.43%

name                      old text-bytes    new text-bytes    delta
HelloSize                       784kB _ 0%        784kB _ 0%  +0.01%  (p=0.000 n=50+50)

name                      old data-bytes    new data-bytes    delta
HelloSize                      13.1kB _ 0%       13.1kB _ 0%    ~     (all equal)

name                      old bss-bytes     new bss-bytes     delta
HelloSize                       206kB _ 0%        206kB _ 0%    ~     (all equal)

name                      old exe-bytes     new exe-bytes     delta
HelloSize                      1.28MB _ 0%       1.28MB _ 0%  +0.00%  (p=0.000 n=50+50)

file      before    after     _       %
addr2line 4006300   4004484   -1816   -0.045%
api       5029956   5029324   -632    -0.013%
asm       4936311   4939423   +3112   +0.063%
buildid   2595059   2595291   +232    +0.009%
cgo       4401029   4397333   -3696   -0.084%
compile   22246677  22246863  +186    +0.001%
cover     4443825   4443065   -760    -0.017%
dist      3366078   3365838   -240    -0.007%
doc       3776391   3776615   +224    +0.006%
fix       3218800   3218648   -152    -0.005%
link      6365321   6365345   +24     +0.000%
nm        3923625   3923857   +232    +0.006%
objdump   4295569   4295041   -528    -0.012%
pack      2390745   2389217   -1528   -0.064%
pprof     12870094  12866942  -3152   -0.024%
test2json 2587265   2587073   -192    -0.007%
trace     9612629   9613981   +1352   +0.014%
vet       6791008   6792072   +1064   +0.016%
total     106856682 106850412 -6270   -0.006%

Update #37608

Change-Id: Ic6206b22fd1faf570be9fd3c2511aa6c4ce38cdb
Reviewed-on: https://go-review.googlesource.com/c/go/+/252937
Trust: fannie zhang <Fannie.Zhang@arm.com>
Run-TryBot: fannie zhang <Fannie.Zhang@arm.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
4 years agoruntime: overwrite existing keys for mapassign_faststr variant
Cuong Manh Le [Fri, 26 Mar 2021 16:29:25 +0000 (23:29 +0700)]
runtime: overwrite existing keys for mapassign_faststr variant

Fixes #45045

Change-Id: Ifcc7bd31591870446ce3e5127489a0b887d413f1
Reviewed-on: https://go-review.googlesource.com/c/go/+/305089
Trust: Cuong Manh Le <cuong.manhle.vn@gmail.com>
Run-TryBot: Cuong Manh Le <cuong.manhle.vn@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
4 years agotime: add Time.Unix{Milli,Micro} and to-Time helpers UnixMicro, UnixMilli
Conrad Irwin [Wed, 17 Feb 2021 22:13:01 +0000 (14:13 -0800)]
time: add Time.Unix{Milli,Micro} and to-Time helpers UnixMicro, UnixMilli

Adds helper functions for users working with other systems which
represent time in milliseconds or microseconds since the Unix epoch.

Fixes #44196

Change-Id: Ibc4490b52ddec94ebd0c692cb7b52a33e4536759
Reviewed-on: https://go-review.googlesource.com/c/go/+/293349
Reviewed-by: Emmanuel Odeke <emmanuel@orijtech.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Emmanuel Odeke <emmanuel@orijtech.com>
TryBot-Result: Go Bot <gobot@golang.org>

4 years agonet: clear completed Buffers to permit earlier collection
Baokun Lee [Tue, 23 Mar 2021 03:18:50 +0000 (11:18 +0800)]
net: clear completed Buffers to permit earlier collection

Fixes #45163

Change-Id: Ie034145e3818930bb19371d73ec6960cbdc55aa7
Reviewed-on: https://go-review.googlesource.com/c/go/+/303829
Run-TryBot: Baokun Lee <bk@golangcn.org>
Trust: Baokun Lee <bk@golangcn.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Emmanuel Odeke <emmanuel@orijtech.com>
4 years agocmd/compile: fix long RMW bit operations on AMD64
Pat Gavlin [Fri, 26 Mar 2021 17:48:42 +0000 (17:48 +0000)]
cmd/compile: fix long RMW bit operations on AMD64

Under certain circumstances, the existing rules for bit operations can
produce code that writes beyond its intended bounds. For example,
consider the following code:

    func repro(b []byte, addr, bit int32) {
    _ = b[3]
    v := uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 | uint32(b[3])<<24 | 1<<(bit&31)
    b[0] = byte(v)
    b[1] = byte(v >> 8)
    b[2] = byte(v >> 16)
    b[3] = byte(v >> 24)
    }

Roughly speaking:

1. The expression `1 << (bit & 31)` is rewritten into `(SHLL 1 bit)`
2. The expression `uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 |
   uint32(b[3])<<24` is rewritten into `(MOVLload &b[0])`
3. The statements `b[0] = byte(v) ... b[3] = byte(v >> 24)` are
   rewritten into `(MOVLstore &b[0], v)`
4. `(ORL (SHLL 1, bit) (MOVLload &b[0]))` is rewritten into
   `(BTSL (MOVLload &b[0]) bit)`. This is a valid transformation because
   the destination is a register: in this case, the bit offset is masked
   by the number of bits in the destination register. This is identical
   to the masking performed by `SHL`.
5. `(MOVLstore &b[0] (BTSL (MOVLload &b[0]) bit))` is rewritten into
   `(BTSLmodify &b[0] bit)`. This is an invalid transformation because
   the destination is memory: in this case, the bit offset is not
   masked, and the chosen instruction may write outside its intended
   32-bit location.

These changes fix the invalid rewrite performed in step (5) by
explicitly maksing the bit offset operand to `BT(S|R|C)(L|Q)modify`. In
the example above, the adjusted rules produce
`(BTSLmodify &b[0] (ANDLconst [31] bit))` in step (5).

These changes also add several new rules to rewrite bit sets, toggles,
and clears that are rooted at `(OR|XOR|AND)(L|Q)modify` operators into
appropriate `BT(S|R|C)(L|Q)modify` operators. These rules catch cases
where `MOV(L|Q)store ((OR|XOR|AND)(L|Q) ...)` is rewritten to
`(OR|XOR|AND)(L|Q)modify` before the `(OR|XOR|AND)(L|Q) ...` can be
rewritten to `BT(S|R|C)(L|Q) ...`.

Overall, compilecmp reports small improvements in code size on
darwin/amd64 when the changes to the compiler itself are exlcuded:

file                               before   after    Δ       %
runtime.s                          536464   536412   -52     -0.010%
bytes.s                            32629    32593    -36     -0.110%
strings.s                          44565    44529    -36     -0.081%
os/signal.s                        7967     7959     -8      -0.100%
cmd/vendor/golang.org/x/sys/unix.s 81686    81678    -8      -0.010%
math/big.s                         188235   188253   +18     +0.010%
cmd/link/internal/loader.s         89295    89056    -239    -0.268%
cmd/link/internal/ld.s             633551   633232   -319    -0.050%
cmd/link/internal/arm.s            18934    18928    -6      -0.032%
cmd/link/internal/arm64.s          31814    31801    -13     -0.041%
cmd/link/internal/riscv64.s        7347     7345     -2      -0.027%
cmd/compile/internal/ssa.s         4029173  4033066  +3893   +0.097%
total                              21298280 21301472 +3192   +0.015%

Change-Id: I2e560548b515865129e1724e150e30540e9d29ce
GitHub-Last-Rev: 9a42bd29a55b3917651aecab6932074df96535ae
GitHub-Pull-Request: golang/go#45242
Reviewed-on: https://go-review.googlesource.com/c/go/+/304869
Reviewed-by: Keith Randall <khr@golang.org>
Trust: Josh Bleecher Snyder <josharian@gmail.com>

4 years agocmd/vendor, cmd/pprof: use golang.org/x/term directly
Tobias Klauser [Thu, 25 Mar 2021 20:26:18 +0000 (21:26 +0100)]
cmd/vendor, cmd/pprof: use golang.org/x/term directly

The cmd/pprof package currently uses golang.org/x/crypto/ssh/terminal
which - as of CL 258003 - is merely a wrapper around golang.org/x/term.

Thus, drop the dependency on golang.org/x/crypto/ssh/terminal and use
golang.org/x/term directly.

Change-Id: Ib15f1f110c338b9dba4a91a873171948ae6298a4
Reviewed-on: https://go-review.googlesource.com/c/go/+/304691
Trust: Tobias Klauser <tobias.klauser@gmail.com>
Run-TryBot: Bryan C. Mills <bcmills@google.com>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Dmitri Shuralyov <dmitshur@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>

4 years agocmd/compile: add arm64 rules to optimize go codes to constant 0
fanzha02 [Tue, 23 Mar 2021 07:35:30 +0000 (15:35 +0800)]
cmd/compile: add arm64 rules to optimize go codes to constant 0

Optimize the following codes to constant 0.

  function shift (x uint32) uint64 {
    return uint64(x) >> 32
  }

Change-Id: Ida6b39d713cc119ad5a2f01fd54bfd252cf2c975
Reviewed-on: https://go-review.googlesource.com/c/go/+/303830
Trust: fannie zhang <Fannie.Zhang@arm.com>
Run-TryBot: fannie zhang <Fannie.Zhang@arm.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
4 years agocmd/compile: add transform functions for OXDOT and builtins
Dan Scales [Wed, 24 Mar 2021 21:50:02 +0000 (14:50 -0700)]
cmd/compile: add transform functions for OXDOT and builtins

Pull out the tranformation part of the typechecking functions for:
 - selector expressions (OXDOT)
 - calls to builtin functions (which go through the typechecker loop
   twice, once for the call and once for each different kind of
   builtin).

Some of the transformation functions create new nodes that should have
the same type as the original node. For consistency, now each of the
transformation functions requires that the node passed in has its type
and typecheck flag set. If the transformation function replaces or adds
new nodes, it will set the type and typecheck flag for those new nodes.

As usual, passes all the gotests, even with -G=3 enabled.

Change-Id: Ic48b0ce5f58425f4a358afa78315bfc7c28066c4
Reviewed-on: https://go-review.googlesource.com/c/go/+/304729
Trust: Dan Scales <danscales@google.com>
Trust: Robert Griesemer <gri@golang.org>
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
4 years agoio/fs: implement FileInfoToDirEntry
Michel Levieux [Thu, 18 Feb 2021 14:53:46 +0000 (15:53 +0100)]
io/fs: implement FileInfoToDirEntry

Implements FileInfoToDirEntry which converts an fs.FileInfo to fs.DirEntry.

Fixes #42387.

Change-Id: Ie723b6ed583c6c5ecf22bbe64e3b6496f5114254
Reviewed-on: https://go-review.googlesource.com/c/go/+/293649
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Emmanuel Odeke <emmanuel@orijtech.com>
4 years agocmd/compile: mark R16, R17 clobbered for non-standard calls on ARM64
Cherry Zhang [Fri, 28 Jun 2019 13:30:36 +0000 (09:30 -0400)]
cmd/compile: mark R16, R17 clobbered for non-standard calls on ARM64

On ARM64, (external) linker generated trampoline may clobber R16
and R17. In CL 183842 we change Duff's devices not to use those
registers. However, this is not enough. The register allocator
also needs to know that these registers may be clobbered in any
calls that don't follow the standard Go calling convention. This
include Duff's devices and the write barrier.

Fixes #32773, second attempt.

Change-Id: Ia52a891d9bbb8515c927617dd53aee5af5bd9aa4
Reviewed-on: https://go-review.googlesource.com/c/go/+/184437
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Meng Zhuo <mzh@golangcn.org>
Reviewed-by: Keith Randall <khr@golang.org>
Trust: Meng Zhuo <mzh@golangcn.org>

4 years agocmd/compile/internal/ssa: unnecessary loop break
cui [Sun, 10 Jan 2021 06:52:16 +0000 (06:52 +0000)]
cmd/compile/internal/ssa: unnecessary loop break

Change-Id: I32860a36b4acf5412c20bac2e8ebbb3965b796fe
GitHub-Last-Rev: c0076390168b2425a1e2c30ea2e9ac32fb4c3365
GitHub-Pull-Request: golang/go#43617
Reviewed-on: https://go-review.googlesource.com/c/go/+/282832
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Keith Randall <khr@golang.org>
Trust: Cherry Zhang <cherryyz@google.com>

4 years agocmd/cover: use golang.org/x/tools/cover directly
Tobias Klauser [Thu, 25 Mar 2021 19:48:27 +0000 (20:48 +0100)]
cmd/cover: use golang.org/x/tools/cover directly

As suggested by Bryan in CL 249759, remove the forwarding aliases in
cmd/cover and use the symbols from golang.org/x/tools directly.

cmd/cover is not an importable package, so it is fine to remove these
exported symbols.

Change-Id: I887c5e9349f2dbe4c90be57f708412b844e18081
Reviewed-on: https://go-review.googlesource.com/c/go/+/304690
Trust: Tobias Klauser <tobias.klauser@gmail.com>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
4 years agonet/http/httptest: wait for user ConnState hooks
Damien Neil [Thu, 25 Mar 2021 18:11:03 +0000 (11:11 -0700)]
net/http/httptest: wait for user ConnState hooks

Ensure that user ConnState callbacks have completed before returning
from (*httptest.Server).Close.

Fixes: #37510
Fixes: #37505
Fixes: #45237
Change-Id: I8fe7baa089fbe4f3836bf6ae9767c7b1270d1331
Reviewed-on: https://go-review.googlesource.com/c/go/+/304829
Trust: Damien Neil <dneil@google.com>
Run-TryBot: Damien Neil <dneil@google.com>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
4 years agonet: only perform IPv4 map check for AF_INET6 sockets
Josh Rickmar [Thu, 25 Mar 2021 14:37:48 +0000 (14:37 +0000)]
net: only perform IPv4 map check for AF_INET6 sockets

This change avoids executing syscalls testing if IPv4 address mapping
is possible unless the socket being opened belongs to the AF_INET6
family.

In a pledged OpenBSD process, this test is only allowed when the
"inet" pledge is granted; however this check was also being performed
for AF_UNIX sockets (separately permitted under the "unix" pledge),
and would cause the process to be killed by the kernel.  By avoiding
the IPv4 address mapping check until the socket is checked to be
AF_INET6, a pledged OpenBSD process using AF_UNIX sockets without the
"inet" pledge won't be killed for this misbehavior.

The OpenBSD kernel is not currently ready to support using UNIX domain
sockets with only the "unix" pledge (and without "inet"), but this is
one change necessary to support this.

Change-Id: If6962a7ad999b71bcfc9fd8e10d9c4067fa3f338
GitHub-Last-Rev: 3c5541b334fad54606a3d14dfe1b63b28b33a0d1
GitHub-Pull-Request: golang/go#45155
Reviewed-on: https://go-review.googlesource.com/c/go/+/303276
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Tobias Klauser <tobias.klauser@gmail.com>
Reviewed-by: Aaron Bieber <deftly@gmail.com>
Trust: Ian Lance Taylor <iant@golang.org>

4 years agocmd/compile/internal/types2: review of importer_test.go
Robert Griesemer [Tue, 23 Mar 2021 22:07:32 +0000 (15:07 -0700)]
cmd/compile/internal/types2: review of importer_test.go

This is a small helper file that provides a default importer
for the type checker tests. There is no go/types equivalent.

The actual change is removing the "// UNREVIEWED" marker.

Change-Id: Ic1f9858bdd9b818d9ddad754e072d9d14d8fb9b3
Reviewed-on: https://go-review.googlesource.com/c/go/+/304252
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
4 years agocmd/compile/internal/types2: review of examples test
Robert Griesemer [Tue, 23 Mar 2021 21:54:19 +0000 (14:54 -0700)]
cmd/compile/internal/types2: review of examples test

The only changes between (equivalent, and reviewed) go/types/examples directory
are in examples/types.go2. The go/types/examples/types.go2 file should be updated
accordingly.

$ f=examples/types.go2; diff $f $HOME/goroot/src/go/types/$f
1d0
< // UNREVIEWED
109c108
< var _ (T /* ERROR cannot use generic type T */ )[ /* ERROR unexpected \[ */ int]
---
> var _ (T /* ERROR cannot use generic type T */ )[ /* ERROR expected ';' */ int]
147a147,154
> // We accept parenthesized embedded struct fields so we can distinguish between
> // a named field with a parenthesized type foo (T) and an embedded parameterized
> // type (foo(T)), similarly to interface embedding.
> // They still need to be valid embedded types after the parentheses are stripped
> // (i.e., in contrast to interfaces, we cannot embed a struct literal). The name
> // of the embedded field is derived as before, after stripping parentheses.
> // (7/14/2020: See comment above. We probably will revert this generalized ability
> // if we go with [] for type parameters.)
149,152c156,158
<  ( /* ERROR cannot parenthesize */ int8)
<  ( /* ERROR cannot parenthesize */ *int16)
<  *( /* ERROR cannot parenthesize */ int32)
<  List[int]
---
>  int8
>  *int16
>  *List[int]
155,156c161
<  * /* ERROR int16 redeclared */ int16
<  List /* ERROR List redeclared */ [int]
---
>  * /* ERROR List redeclared */ List[int]
280a286
>

The actual changes are removing the "// UNREVIEWED" markers.

Change-Id: I8a80fa11f3c84f9a403c690b537973a53e1adc2c
Reviewed-on: https://go-review.googlesource.com/c/go/+/304250
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
4 years agocmd/compile/internal/types2: review of example_test.go
Robert Griesemer [Tue, 23 Mar 2021 21:30:57 +0000 (14:30 -0700)]
cmd/compile/internal/types2: review of example_test.go

The changes between (equivalent, and reviewed) go/types/example_test.go
and example_test.go can be seen by comparing patchset 1 and 2. The actual
changes are removing the "// UNREVIEWED" marker.

The primary differences to go/types/example_test.go are:
- use of syntax instead of go/ast package
- no ExampleMethodSet test (types2 doesn't have MethodSet)
- some code in ExampleInfo is disabled due to less precise
  position information provided by the syntax tree

Change-Id: I035284357acc8ecb7849022b5a9d873ae2235987
Reviewed-on: https://go-review.googlesource.com/c/go/+/304249
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
4 years agocmd/compile/internal/types2: review of api_test.go
Robert Griesemer [Tue, 23 Mar 2021 19:44:17 +0000 (12:44 -0700)]
cmd/compile/internal/types2: review of api_test.go

The changes between (equivalent, and reviewed) go/types/api_test.go
and api_test.go can be seen by comparing patchset 1 and 2. The actual
changes are removing the "// UNREVIEWED" marker, the addition of the
TestConvertibleTo and TestAssignableTo tests, and adjustments to test
prefixes (genericPkg, brokenPkg to be in line with go/types).

There are several differences to go/types/api_test.go:
- use of syntax rather than go/ast package
- use of the parseSrc helper function
- TestTypesInfo test entries reflect different handling of untyped nil
- TestInferredInfo is (for go1.17) in another file controlled by a build
  constraint in go/types
- TestSelection test is currently skipped (types2 position information
  is not accurate enough)
- TestScopeLookupParent doesn't have access to a scanner and instead
  relies on syntax.CommentsDo.
- Broken packages are assumed to contain generic code for the tests.

Change-Id: Ic14e6fb9d6bef5416df39e465b5994de76f84097
Reviewed-on: https://go-review.googlesource.com/c/go/+/304131
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
4 years agocmd/compile/internal/types2: review of lookup.go
Robert Griesemer [Tue, 23 Mar 2021 19:19:50 +0000 (12:19 -0700)]
cmd/compile/internal/types2: review of lookup.go

The changes between (equivalent, and reviewed) go/types/lookup.go
and lookup.go can be seen by comparing patchset 1 and 2. The actual
changes are removing the "// UNREVIEWED" marker.

Note: The function ptrRecv in types2/lookup.go is found in
      methodset.go in go/types (methodset.go doesn't exist
      in types2).

Change-Id: I48cfd3df0947becb4c3b5e55b89263917bcfbf16
Reviewed-on: https://go-review.googlesource.com/c/go/+/304129
Reviewed-by: Robert Findley <rfindley@google.com>
Trust: Robert Griesemer <gri@golang.org>

4 years agocmd/compile/internal/types2: review of check.go
Robert Griesemer [Tue, 23 Mar 2021 19:11:26 +0000 (12:11 -0700)]
cmd/compile/internal/types2: review of check.go

The changes between (equivalent, and reviewed) go/types/check.go
and check.go can be seen by comparing patchset 1 and 2. The actual
changes are removing the "// UNREVIEWED" marker.

The primary differences to go/types/check.go are:
- use of syntax instead of go/ast package
- tracing is controlled via flag not the "trace" constant

Change-Id: I1c9998afb3e0b7e29f5b169d3a4054cf22841490
Reviewed-on: https://go-review.googlesource.com/c/go/+/304109
Reviewed-by: Robert Findley <rfindley@google.com>
Trust: Robert Griesemer <gri@golang.org>

4 years agocmd/go: add -benchtime to cacheable test flags
Johan Jansson [Wed, 24 Mar 2021 15:14:29 +0000 (17:14 +0200)]
cmd/go: add -benchtime to cacheable test flags

Add -benchtime to the list of flags that allow caching test results.

If -benchtime is set without -bench, no benchmarks are run. The cache
does not need to be invalidated in this case.

If -benchtime is set with -bench, benchmarks are run. The cache is
invalidated due to the -bench flag in this case.

Fixes #44555

Change-Id: I2eb5c9f389a587d150fb984590d145251d0fa2dc
Reviewed-on: https://go-review.googlesource.com/c/go/+/304689
Reviewed-by: Jay Conrod <jayconrod@google.com>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
Trust: Jay Conrod <jayconrod@google.com>
Run-TryBot: Jay Conrod <jayconrod@google.com>
TryBot-Result: Go Bot <gobot@golang.org>

4 years agocmd/link: make symbol data writable before toc fixup
Paul E. Murphy [Tue, 23 Mar 2021 20:40:54 +0000 (15:40 -0500)]
cmd/link: make symbol data writable before toc fixup

On ppc64le, we need to insert a load to restore the toc
pointer in R2 after calling into plt stubs. Sometimes the
symbol data is loaded into readonly memory. This is the
case when linking with the race detector code.

Likewise, add extra checks to ensure we can, and are
replacing a nop.

Change-Id: Iea9d9ee7a5ba0f4ce285f4d0422823de1c037cb7
Reviewed-on: https://go-review.googlesource.com/c/go/+/304430
Run-TryBot: Paul Murphy <murp@ibm.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Lynn Boger <laboger@linux.vnet.ibm.com>

4 years agodatabase/sql: remove unnecessary types in composite literals
Elias506 [Thu, 25 Mar 2021 14:46:15 +0000 (14:46 +0000)]
database/sql: remove unnecessary types in composite literals

Change-Id: I30c576f826c82cbc62ce28ea7f4886702bd6605d
GitHub-Last-Rev: 2fead200dba13ec74d81e842d8bae97d1e67fcc4
GitHub-Pull-Request: golang/go#42618
Reviewed-on: https://go-review.googlesource.com/c/go/+/270000
Reviewed-by: Daniel Theophanes <kardianos@gmail.com>
Trust: Cherry Zhang <cherryyz@google.com>

4 years agocmd/compile: fix defer desugar keepalive arg handling buglet
Than McIntosh [Thu, 25 Mar 2021 12:30:19 +0000 (08:30 -0400)]
cmd/compile: fix defer desugar keepalive arg handling buglet

Fix a bug in the go/defer desugar handling of keepalive arguments. The
go/defer wrapping code has special handling for calls whose arguments
are pointers that have been cast to "uintptr", so as to insure that
call "keepalive" machinery for such calls continues to work. This
patch fixes a bug in the special case code to insure that it doesn't
kick in for other situations where you have an unsafe.Pointer ->
uintptr argument (outside the keepalive context).

Fixes make.bat on windows with GOEXPERIMENT=regabidefer in effect.

Change-Id: I9db89c4c73f0db1235901a4fae57f62f88c94ac3
Reviewed-on: https://go-review.googlesource.com/c/go/+/304457
Trust: Than McIntosh <thanm@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: David Chase <drchase@google.com>
4 years agocmd/go: emit error when listing with -f and -json
Ayan George [Tue, 2 Mar 2021 22:44:22 +0000 (17:44 -0500)]
cmd/go: emit error when listing with -f and -json

Fixes #44738
Change-Id: Ie57ddcbe87408c9644313ec2a9ea347b4d6de76b
Reviewed-on: https://go-review.googlesource.com/c/go/+/298029
Reviewed-by: Bryan C. Mills <bcmills@google.com>
Trust: Jay Conrod <jayconrod@google.com>

4 years agopath/filepath: make Rel handle Windows UNC share
Aman Gupta [Fri, 4 Sep 2020 23:07:12 +0000 (16:07 -0700)]
path/filepath: make Rel handle Windows UNC share

Fixes #41230

Change-Id: Iea15e4ae6d56328333fd22de5d78dfcad78ef1bc
Reviewed-on: https://go-review.googlesource.com/c/go/+/253197
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
Trust: Alex Brainman <alex.brainman@gmail.com>
Trust: Emmanuel Odeke <emmanuel@orijtech.com>
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org>

4 years agocmd/compile/internal/types2: review of stdlib_test.go
Robert Griesemer [Tue, 23 Mar 2021 19:30:18 +0000 (12:30 -0700)]
cmd/compile/internal/types2: review of stdlib_test.go

The changes between (equivalent, and reviewed) go/types/stdlib_test.go
and stdlib_test.go can be seen by comparing patchset 1 and 2. The actual
changes are removing the "// UNREVIEWED" marker, using the os package
instead of ioutil, and some comment adjustments. Also, bug251.go passes
because of recent changes.

The primary difference is in the firstComment function which
doesn't have access to a scanner and instead uses the syntax
package's CommentsDu function.

Change-Id: I946ffadc97e87c692f76f369a1b16cceee528477
Reviewed-on: https://go-review.googlesource.com/c/go/+/304130
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
4 years agocmd/compile/internal/types2: review of assignments.go
Robert Griesemer [Tue, 23 Mar 2021 19:03:51 +0000 (12:03 -0700)]
cmd/compile/internal/types2: review of assignments.go

The changes between (equivalent, and reviewed) go/types/assignments.go
and assignments.go can be seen by comparing patchset 1 and 2. The actual
changes are removing the "// UNREVIEWED" marker.

The primary differences to go/types/assignments.go are:
- use of syntax instead of go/ast package
- no reporting of error codes (for now)
- different handling of nil values (we can't use Typ[UntypedNil]
  to represent an untyped nil because types2 gives such nil values
  context-dependent types)

Change-Id: I5d8a58f43ca8ed2daa060c46842a6ebc11b3cb35
Reviewed-on: https://go-review.googlesource.com/c/go/+/304051
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
4 years agocmd/compile/internal/types2: review of api.go
Robert Griesemer [Tue, 23 Mar 2021 18:55:16 +0000 (11:55 -0700)]
cmd/compile/internal/types2: review of api.go

The changes between (equivalent, and reviewed) go/types/api.go
and api.go can be seen by comparing patchset 1 and 2. The actual
changes are removing the "// UNREVIEWED" marker.

The primary differences to go/types/api.go are:
- use of syntax instead of go/ast package
- use of simpler Error type (for now)
- additional exported Config flags
- different handling of nil values (we can't use Typ[UntypedNil]
  to represent an untyped nil because types2 gives such nil values
  context-dependent types)

Change-Id: I7d46b29d460c656d7a36fe70108a370383266373
Reviewed-on: https://go-review.googlesource.com/c/go/+/304050
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
4 years agocmd/compile/internal/types2: review of expr.go
Robert Griesemer [Fri, 19 Mar 2021 00:15:05 +0000 (17:15 -0700)]
cmd/compile/internal/types2: review of expr.go

The changes between (equivalent, and reviewed) go/types/expr.go
and expr.go can be seen by comparing patchset 1 and 2. The actual
changes are removing the "// UNREVIEWED" marker.

The primary differences to go/types/expr.go are:
- use of package syntax rather than ast
- no reporting of error codes in errors
- implicit conversions of untyped nil lead to a typed nil
  (in go/types, nil remains untyped)

Change-Id: I1e235b20ebda597eb7ce597d1749f26431addde2
Reviewed-on: https://go-review.googlesource.com/c/go/+/303092
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
4 years agocmd/go/internal/load: use setLoadPackageDataError in loadImport
Bryan C. Mills [Tue, 23 Mar 2021 15:57:36 +0000 (11:57 -0400)]
cmd/go/internal/load: use setLoadPackageDataError in loadImport

This makes the error handling in loadImport somewhat more uniform,
with no discernable effect on reported errors.

Noticed in CL 303869.

Updates #36087
Updates #38034

This somewhat simplifies the code, with no discernable effect on

Change-Id: I30521f658f264d6f99d1844d6701269bbb372246
Reviewed-on: https://go-review.googlesource.com/c/go/+/304069
Trust: Bryan C. Mills <bcmills@google.com>
Run-TryBot: Bryan C. Mills <bcmills@google.com>
Reviewed-by: Michael Matloob <matloob@golang.org>
Reviewed-by: Jay Conrod <jayconrod@google.com>
4 years agocmd/go: attribute direct imports from indirect dependencies to the importing package
Bryan C. Mills [Tue, 23 Mar 2021 04:26:39 +0000 (00:26 -0400)]
cmd/go: attribute direct imports from indirect dependencies to the importing package

For #36460
Updates #40775

Change-Id: I833ee8ee733151aafcff508bf91d0e6e37c50032
Reviewed-on: https://go-review.googlesource.com/c/go/+/303869
Trust: Bryan C. Mills <bcmills@google.com>
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Jay Conrod <jayconrod@google.com>
4 years agocmd/go/internal/modload: replace the global buildList with structured requirements
Bryan C. Mills [Wed, 3 Mar 2021 15:27:19 +0000 (10:27 -0500)]
cmd/go/internal/modload: replace the global buildList with structured requirements

This is intended to be a pure-refactoring change, with very little
observable change in the behavior of the 'go' command. A few error
messages have prefixes changed (by virtue of being attached to
packages or modules instead of the build list overall), and
'go list -m' (without arguments) no longer loads the complete module
graph in order to provide the name of the (local) main module.

The previous modload.buildList variable contained a flattened build
list, from which the go.mod file was reconstructed using various
heuristics and metadata cobbled together from the original go.mod
file, the package loader (which was occasionally constructed without
actually loading packages, for the sole purpose of populating
otherwise-unrelated metadata!), and the updated build list.

This change replaces that variable with a new package-level variable,
named "requirements". The new variable is structured to match the
structure of the go.mod file: it explicitly specifies the roots of the
module graph, from which the complete module graph and complete build
list can be reconstructed (and cached) on demand. Similarly, the
"direct" markings on the go.mod requirements are now stored alongside
the requirements themselves, rather than side-channeled through the
loader.

The requirements are now plumbed explicitly though the modload
package, with accesses to the package-level variable occurring only
within top-level exported functions. The structured requirements are
logically immutable, so a new copy of the requirements is constructed
whenever the requirements are changed, substantially reducing implicit
communication-by-sharing in the package.

For #36460
Updates #40775

Change-Id: I97bb0381708f9d3e42af385b5c88a7038e1f0556
Reviewed-on: https://go-review.googlesource.com/c/go/+/293689
Trust: Bryan C. Mills <bcmills@google.com>
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Michael Matloob <matloob@golang.org>
4 years agotest: skip fixedbugs/issue36705 on Windows
Bryan C. Mills [Thu, 25 Mar 2021 01:51:44 +0000 (21:51 -0400)]
test: skip fixedbugs/issue36705 on Windows

This test is verifying that setting or unsetting an environment
variable in Go via the "os" package makes that change visible to the C
getenv function. The test has been failing on Windows since CL 304569;
it isn't clear to me whether it was running at all before that point.

On Windows the getenv and _putenv C functions are not thread-safe,
so Go's os.Setenv and os.Getenv use the SetEnvironmentVariable and
GetEnvironmentVariable system calls instead. That seems to work fine
in practice; however, changes via SetEnvironmentVariable are
empirically not visible to the C getenv function on certain versions
of Windows.

The MSDN getenv documentation¹ states that ‘getenv operates only on
the data structures accessible to the run-time library and not on the
environment “segment” created for the process by the operating system.
Therefore, programs that use the envp argument to main or wmain may
retrieve invalid information.’ That may be related to what we're
seeing here.

(https://github.com/curl/curl/issues/4774 describes this same behavior
observed in the curl project.)

¹https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/getenv-wgetenv?view=msvc-160#remarks

Updates #36705

Change-Id: I222792f75c650f32c5025b0fa3edab232ff66353
Reviewed-on: https://go-review.googlesource.com/c/go/+/304669
Trust: Bryan C. Mills <bcmills@google.com>
Run-TryBot: Bryan C. Mills <bcmills@google.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>

4 years agocrypto/x509: fix spelling error
Kevin Burke [Wed, 24 Mar 2021 23:02:24 +0000 (16:02 -0700)]
crypto/x509: fix spelling error

Change-Id: Ieb1900531f42acf2c8b98ac89fceb8b87c8e5d0c
Reviewed-on: https://go-review.googlesource.com/c/go/+/304609
Reviewed-by: Roland Shoemaker <roland@golang.org>
Trust: Kevin Burke <kev@inburke.com>

4 years agocmd/compile: create/use noder2 transform functions for more node types
Dan Scales [Tue, 23 Mar 2021 17:19:11 +0000 (10:19 -0700)]
cmd/compile:  create/use noder2 transform functions for more node types

Pull out the transformation part of the typechecking functions for:
 - assignment statements
 - return statements
 - send statements
 - select statements
 - type conversions
 - normal function/method calls
 - index operations

The transform functions are like the original typechecking functions,
but with all code removed related to:
  - Detecting compile-time errors (already done by types2)
  - Setting the actual type of existing nodes (already done based on
    info from types2)
  - Dealing with untyped constants

Moved all the transformation functions to a separate file, transform.go.

Continuing with the same pattern, we delay transforming a node if it has
any type params in its args, marking it with a typecheck flag of 3, and
do the actual transformation during stenciling.

Assignment statements are tricky, since their transformation must be
delayed if any of the left or right-hands-sides are delayed.

Still to do are:
 - selector expressions (OXDOT)
 - composite literal expressions (OCOMPLIT)
 - builtin function calls

Change-Id: Ie608cadbbc69b40db0067a5536cf707dd974aacc
Reviewed-on: https://go-review.googlesource.com/c/go/+/304049
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Dan Scales <danscales@google.com>
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
4 years agotesting: update permitted number of -race goroutines
Ian Lance Taylor [Tue, 23 Mar 2021 22:44:46 +0000 (15:44 -0700)]
testing: update permitted number of -race goroutines

The value 8128 appears to be correct as of 2021-03-23. The value is
determined by the value of kMaxTid currently at

https://github.com/llvm/llvm-project/blob/main/compiler-rt/lib/tsan/rtl/tsan_defs.h#L68

Fixes #45193

Change-Id: If4a30d7dbebd6775bede42f565dc8a741b8b036c
Reviewed-on: https://go-review.googlesource.com/c/go/+/304254
Trust: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
4 years agotest: only run bug513.go if cgo is enabled
Ian Lance Taylor [Wed, 24 Mar 2021 20:18:13 +0000 (13:18 -0700)]
test: only run bug513.go if cgo is enabled

Change-Id: I868eeb79edaba9e3afc1407ae18b89daf7e67037
Reviewed-on: https://go-review.googlesource.com/c/go/+/304570
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>

4 years agotest: recognize cgo build tag
Ian Lance Taylor [Wed, 24 Mar 2021 20:16:42 +0000 (13:16 -0700)]
test: recognize cgo build tag

This requires us to add a fake argument to issue36705.go so that the
test driver will build it with "go run" rather than "go tool compile".

Change-Id: Id08b97d898ee3e9d6c1fbb072a0a9317ed9faedd
Reviewed-on: https://go-review.googlesource.com/c/go/+/304569
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
4 years agocmd/internal/moddeps: fix false positive when $TMPDIR is symlinked
Dmitri Shuralyov [Wed, 24 Mar 2021 16:10:22 +0000 (12:10 -0400)]
cmd/internal/moddeps: fix false positive when $TMPDIR is symlinked

os.Getwd notes that if the current directory can be reached via
multiple paths (due to symbolic links), Getwd may return any one
of them. A way to ensure that the desired path is used is to set
the PWD environment variable pointing to it.

The go generate command has started to update the PWD environment
variable as of CL 287152, which was the missing link previously
resulting in mkwinsyscall misunderstanding whether it's inside
the std lib when symbolic links are involved (issue 44079).

Now all that's left is for us to also set the PWD environment
variable when invoking the go command in the test, so that it
too knows the intended working directory path to use.

Fixes #44080.
Updates #44079.
Updates #43862.

Change-Id: I65c9d19d0979f486800b9b328c9b45a1a3180e81
Reviewed-on: https://go-review.googlesource.com/c/go/+/304449
Trust: Dmitri Shuralyov <dmitshur@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
4 years agoruntime: bypass ABI wrapper when calling needm on Windows
Michael Anthony Knyszek [Thu, 18 Mar 2021 17:03:50 +0000 (17:03 +0000)]
runtime: bypass ABI wrapper when calling needm on Windows

On Windows, when calling into needm in cgocallback on a new thread that
is unknown to the Go runtime, we currently call through an ABI wrapper.
The ABI wrapper tries to restore the G register from TLS.

On other platforms, TLS is set up just enough that the wrapper will
simply load a nil g from TLS, but on Windows TLS isn't set up at all, so
there's nowhere for the wrapper to load from.

So, bypass the wrapper in the call to needm. needm takes no arguments
and returns no results so there are no special ABI considerations,
except that we must clear X15 which is used as a zero register in Go
code (a function normally performed by the ABI wrapper). needm is also
otherwise already special and carefully crafted to avoid doing anything
that would require a valid G or M, at least until it is able to create
one.

While we're here, this change simplifies setg so that it doesn't set up
TLS on Windows and instead provides an OS-specific osSetupTLS to do
that.

The result of this is that setg(nil) no longer clears the TLS space
pointer on Windows. There's exactly one place this is used (dropm) where
it doesn't matter anymore, and an empty TLS means that setg's wrapper
will crash on the return path. Another result is that the G slot in the
TLS will be properly cleared, however, which isn't true today.

For #40724.

Change-Id: I65c3d924a3b16abe667b06fd91d467d6d5da31d7
Reviewed-on: https://go-review.googlesource.com/c/go/+/303070
Trust: Michael Knyszek <mknyszek@google.com>
Run-TryBot: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
4 years agocmd/compile: disable shortcircuit optimization for intertwined phi values
Keith Randall [Tue, 23 Mar 2021 21:48:47 +0000 (14:48 -0700)]
cmd/compile: disable shortcircuit optimization for intertwined phi values

We need to be careful that when doing value graph surgery, we not
re-substitute a value that has already been substituted. That can lead
to confusing a previous iteration's value with the current iteration's
value.

The simple fix in this CL just aborts the optimization if it detects
intertwined phis (a phi which is the argument to another phi). It
might be possible to keep the optimization with a more complicated
CL, but:
  1) This CL is clearly safe to backport.
  2) There were no instances of this abort triggering in
     all.bash, prior to the test introduced in this CL.

Fixes #45175

Change-Id: I2411dca03948653c053291f6829a76bec0c32330
Reviewed-on: https://go-review.googlesource.com/c/go/+/304251
Trust: Keith Randall <khr@golang.org>
Trust: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
4 years agocmd/link: mangle function names with ABI on PE
Cherry Zhang [Wed, 24 Mar 2021 15:19:19 +0000 (11:19 -0400)]
cmd/link: mangle function names with ABI on PE

When ABI wrappers are used, we may end up with two functions
having the same name. On ELF we mangle the name with ABI. Do the
same for PE.

TODO: other platforms?

Change-Id: If89f214a6286bc28c062c1aa1bad78dc353a9231
Reviewed-on: https://go-review.googlesource.com/c/go/+/304432
Trust: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
4 years agonet/http: fix request cancellation race
Damien Neil [Sat, 20 Mar 2021 05:01:10 +0000 (22:01 -0700)]
net/http: fix request cancellation race

When a in-flight request is cancelled, (*Transport).cancelRequest is
called. The cancelRequest function looks up and invokes a cancel
function before returning. The function lookup happens with reqMu held,
but the cancel function is invoked after dropping the mutex.

If two calls to cancelRequest are made at the same time, it is possible
for one to return before the cancel function has been invoked.

This race causes flakiness in TestClientTimeoutCancel:
  - The test cancels a request while a read from the request body is
    pending.
  - One goroutine calls (*Transport).cancelRequest. This goroutine
    will eventually invoke the cancel function.
  - Another goroutine calls (*Transport).cancelRequest and closes the
    request body. The cancelRequest call returns without invoking
    the cancel function.
  - The read from the request body returns an error. The reader
    checks to see if the request has been canceled, but concludes
    that it has not (because the cancel function hasn't been invoked
    yet).

To avoid this race condition, call the cancel function with the
transport reqMu mutex held.

Calling the cancel function with the mutex held does not introduce any
deadlocks that I can see. The only non-noop request cancel functions
are:

A send to a buffered channel:
https://go.googlesource.com/go/+/refs/heads/master/src/net/http/transport.go#1362

The (*persistConn).cancelRequest function, which does not cancel any
other requests:
https://go.googlesource.com/go/+/refs/heads/master/src/net/http/transport.go#2526

Fixes #34658.

Change-Id: I1b83dce9b0b1d5cf7c7da7dbd03d0fc90c9f5038
Reviewed-on: https://go-review.googlesource.com/c/go/+/303489
Trust: Damien Neil <dneil@google.com>
Run-TryBot: Damien Neil <dneil@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
4 years agomath/big: don't require runtime.(*Frame).Next symbol present
Cherry Zhang [Wed, 24 Mar 2021 15:02:46 +0000 (11:02 -0400)]
math/big: don't require runtime.(*Frame).Next symbol present

I don't know why the test requires runtime.(*Frame).Next symbol
present in the binary under test. I assume it is just some
sanity check? With CL 268479 runtime.(*Frame).Next can be pruned
by the linker. Replace it with runtime.main which should always
be present.

May fix the longtest builders.

Change-Id: Id3104c058b2786057ff58be41b1d35aeac2f3073
Reviewed-on: https://go-review.googlesource.com/c/go/+/304431
Trust: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
4 years agocmd/link: separate elf addend size from reloc size
Paul E. Murphy [Tue, 23 Mar 2021 14:51:43 +0000 (09:51 -0500)]
cmd/link: separate elf addend size from reloc size

The size of the field may be smaller than the addend,
such is the case with R_PPC64_TOC16_HA/LO and similar
relocations.

Add an extra return value to ldelf.relSize to account for
addend size which may be larger than the relocated field,
and fix the related ppc64 relocations.

Such relocs can be seen in large PIC blobs such
as the ppc64le race detector included with golang.

Change-Id: I457186fea5d0ec5572b9bbf79bb7fa21a36cc1b0
Reviewed-on: https://go-review.googlesource.com/c/go/+/303990
Run-TryBot: Paul Murphy <murp@ibm.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Lynn Boger <laboger@linux.vnet.ibm.com>

4 years agocmd/compile, cmd/link: use weak reference in itab
Cherry Zhang [Sun, 8 Nov 2020 20:18:35 +0000 (15:18 -0500)]
cmd/compile, cmd/link: use weak reference in itab

When converting a type T to a non-empty interface I, we build the
itab which contains the code pointers of the methods. Currently,
this brings those methods live (if the itab is live), even if the
interface method is never used. This CL changes the itab to use
weak references, so the methods can be pruned if not otherwise
live.

Fixes #42421.

Change-Id: Iee5de2ba11d603c5a102a2ba60440d839a7f9702
Reviewed-on: https://go-review.googlesource.com/c/go/+/268479
Trust: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Than McIntosh <thanm@google.com>
4 years agocmd/internal/obj: remove bogus load/store optab entries from ppc64
Paul E. Murphy [Tue, 9 Mar 2021 22:55:04 +0000 (16:55 -0600)]
cmd/internal/obj: remove bogus load/store optab entries from ppc64

No valid operation should match those removed by this patch. They
kind of look as if they match X-form load/stores on ppc64, but the
second argument is always ignored when translating to machine code.

Similarly, it should be noted an X-form memory access encodes into
an Addr which is a classified as a ZOREG argument with a non-zero
index, and a register type Addr.

Change-Id: I1adbb020d1b2612b18949d0e7eda05dbb3e8a25c
Reviewed-on: https://go-review.googlesource.com/c/go/+/303329
Reviewed-by: Carlos Eduardo Seo <carlos.seo@linaro.org>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
Run-TryBot: Paul Murphy <murp@ibm.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Carlos Eduardo Seo <carlos.seo@linaro.org>

4 years agocmd/go: move psuedo-version and version sorting to x/mod
Jay Conrod [Tue, 23 Mar 2021 21:03:46 +0000 (17:03 -0400)]
cmd/go: move psuedo-version and version sorting to x/mod

Fixes #44969

Change-Id: I01e7b1cf73f0f506aa805bbfe4a9ccaed3d44efe
Reviewed-on: https://go-review.googlesource.com/c/go/+/304229
Trust: Jay Conrod <jayconrod@google.com>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
4 years agoos/exec: use testenv.SkipFlaky in TestExtraFilesFDShuffle
Tobias Klauser [Tue, 23 Mar 2021 10:59:34 +0000 (11:59 +0100)]
os/exec: use testenv.SkipFlaky in TestExtraFilesFDShuffle

Change-Id: Icc1980ea0f6363c667bf91371404a8df6ed6b8ab
Reviewed-on: https://go-review.googlesource.com/c/go/+/303950
Trust: Tobias Klauser <tobias.klauser@gmail.com>
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
Reviewed-by: Emmanuel Odeke <emmanuel@orijtech.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>

4 years agocmd/compile: remove more dead code and data structures
David Chase [Tue, 23 Mar 2021 19:51:29 +0000 (15:51 -0400)]
cmd/compile: remove more dead code and data structures

Remove more now-redundant code, methods, and types
associated with transition to register ABI.
Repaired some broken comments.

Tested on link-register architectures (arm64, ppc64le)

Updates #40724.

Change-Id: Ie8433f6d38ec4a1d9705f22dcb596f267d81f203
Reviewed-on: https://go-review.googlesource.com/c/go/+/304189
Trust: David Chase <drchase@google.com>
Run-TryBot: David Chase <drchase@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
4 years agocmd/compile: fix array case in types-for-register parameter
David Chase [Tue, 23 Mar 2021 23:55:17 +0000 (19:55 -0400)]
cmd/compile: fix array case in types-for-register parameter

Corrected typo/thinko.

We should keep the test for this, but it doesn't run yet because of reflection
as far as I know (but I am not testing w/ GOEXPERIMENT).

See https://github.com/golang/go/issues/44816#issuecomment-805297295

Updates #40724
Updates #44816

Change-Id: Ia12d0d4db00a8ec7174e72de460173876bd17874
Reviewed-on: https://go-review.googlesource.com/c/go/+/304233
Trust: David Chase <drchase@google.com>
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
4 years agocmd/compile: don't let -race override explicit -d=checkptr=0
Ian Lance Taylor [Tue, 23 Mar 2021 22:24:27 +0000 (15:24 -0700)]
cmd/compile: don't let -race override explicit -d=checkptr=0

Change-Id: Icfa204761045b72a8ea173fd55eddf1f0e58d819
Reviewed-on: https://go-review.googlesource.com/c/go/+/304253
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
4 years agocmd/compile: wrap/desugar defer calls for register abi
Than McIntosh [Wed, 24 Feb 2021 17:55:52 +0000 (12:55 -0500)]
cmd/compile: wrap/desugar defer calls for register abi

Adds code to the compiler's "order" phase to rewrite go and defer
statements to always be argument-less. E.g.

 defer f(x,y)       =>     x1, y1 := x, y
   defer func() { f(x1, y1) }

This transformation is not beneficial on its own, but it helps
simplify runtime defer handling for the new register ABI (when
invoking deferred functions on the panic path, the runtime doesn't
need to manage the complexity of determining which args to pass in
register vs memory).

This feature is currently enabled by default if GOEXPERIMENT=regabi or
GOEXPERIMENT=regabidefer is in effect.

Included in this CL are some workarounds in the runtime to insure that
"go" statement targets in the runtime are argument-less already (since
wrapping them can potentially introduce heap-allocated closures, which
are currently not allowed). The expectation is that these workarounds
will be temporary, and can go away once we either A) change the rules
about heap-allocated closures, or B) implement some other scheme for
handling go statements.

Change-Id: I01060d79a6b140c6f0838d6e6813f807ccdca319
Reviewed-on: https://go-review.googlesource.com/c/go/+/298669
Trust: Than McIntosh <thanm@google.com>
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: David Chase <drchase@google.com>
4 years agoos/exec: simplify TestContextCancel
Shuhei Takahashi [Sun, 21 Mar 2021 15:22:23 +0000 (00:22 +0900)]
os/exec: simplify TestContextCancel

TestContextCancel is a test that ensures a process is killed soon after
canceling the context, even if Wait is not called (#16222). The test
checks whether the process exited without calling Wait by writing some
data to its stdin.

Currently the test involves two goroutines writing to stdin and reading
from stdout. However the reading goroutine is not very necessary to
detect the process exit.

This patch simplifies the test by connecting the process stdout to
/dev/null.

For #42061

Change-Id: I0447a1c024ee5abb050c627ec3766b731b02181a
Reviewed-on: https://go-review.googlesource.com/c/go/+/303352
Trust: Emmanuel Odeke <emmanuel@orijtech.com>
Run-TryBot: Emmanuel Odeke <emmanuel@orijtech.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
4 years agoio: fix spelling in documentation for io.Discard
Simão Gomes Viana [Tue, 23 Mar 2021 21:50:41 +0000 (21:50 +0000)]
io: fix spelling in documentation for io.Discard

In the process of refactoring ioutil.Discard to io.Discard in
CL 263141 "an" should have been changed to "a" but was likely
missed in the process.

This commit corrects the spelling of the documentation.

Change-Id: I0609c45878291f8f01560efc3f3e6fba191e095b
GitHub-Last-Rev: e3257ca272dff42ed7d07b6e6a5fc49493772653
GitHub-Pull-Request: golang/go#45190
Reviewed-on: https://go-review.googlesource.com/c/go/+/304209
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Robert Griesemer <gri@golang.org>

4 years agocmd/compile/internal/test: update abi tests for recent spec changes
Than McIntosh [Tue, 26 Jan 2021 13:24:00 +0000 (08:24 -0500)]
cmd/compile/internal/test: update abi tests for recent spec changes

The internal ABI spec was recently updated to include specific
language covering "past-the-end" pointers and structs containing
trailing zero-sized fields. Add a unit test that makes sure we do the
right thing in this case. Fix a couple comments in other unit tests.

Change-Id: I18d373d11e122aec74b316837843887272676c63
Reviewed-on: https://go-review.googlesource.com/c/go/+/303809
Trust: Than McIntosh <thanm@google.com>
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
4 years agocmd/go: make -coverpkg=all skip test-only packages
Daniel Martí [Sat, 30 Jan 2021 11:41:29 +0000 (11:41 +0000)]
cmd/go: make -coverpkg=all skip test-only packages

Otherwise, the added test would fail in an unnecessary way:

go build example.com/cov/onlytest: no non-test Go files ...

The test script is mimicking other cover_pkgall_*.txt scripts, so it
similarly tests both GOPATH and module modes.

Fixes #27333.

Change-Id: Ie60be569b31d49b173a78556c0669a87ada6799e
Reviewed-on: https://go-review.googlesource.com/c/go/+/288292
Trust: Daniel Martí <mvdan@mvdan.cc>
Trust: Jay Conrod <jayconrod@google.com>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Jay Conrod <jayconrod@google.com>
4 years agonet/http: treat MaxBytesReader's negative limits as equivalent to zero limit
Artur M. Wolff [Sun, 21 Mar 2021 00:18:21 +0000 (01:18 +0100)]
net/http: treat MaxBytesReader's negative limits as equivalent to zero limit

Current MaxBytesReader behaviour differs from its documentation. It's
not similar enough to io.LimitReader. It panics when limit (n) < -1 and
returns [-1, <nil>] when limit (n) = -1. To fix that, we treat all
negative limits as equivalent to 0.

It would be possible to make MaxBytesReader analogically identical in
behaviour to io.LimitReader, but that would require to stop
maxBytesReader's Read from reading past the limit. Read always reads one
more byte (if possible) for non-negative limits and returns a non-EOF
error. This behaviour will now apply to all limits.

Fixes #45101

Change-Id: I25d1877dbff1eb4b195c8741fe5e4a025d01ebc0
Reviewed-on: https://go-review.googlesource.com/c/go/+/303171
Reviewed-by: Damien Neil <dneil@google.com>
Reviewed-by: Dmitri Shuralyov <dmitshur@golang.org>
Trust: Damien Neil <dneil@google.com>
Trust: Dmitri Shuralyov <dmitshur@golang.org>
Run-TryBot: Damien Neil <dneil@google.com>
TryBot-Result: Go Bot <gobot@golang.org>

4 years agocmd/compile: remove AuxCall.results, cleanup ssagen/ssa.go
David Chase [Fri, 19 Mar 2021 20:52:55 +0000 (16:52 -0400)]
cmd/compile: remove AuxCall.results, cleanup ssagen/ssa.go

More cleanup to remove unnecessary parts of AuxCall.
Passed testing on arm64 (a link-register architecture)
in addition to amd64 so very likely okay.

(Gratuitously updated commit message to see if it will
correctly this time.)

Updates #40724

Change-Id: Iaece952ceb5066149a5d32aaa14b36755f26bb8e
Reviewed-on: https://go-review.googlesource.com/c/go/+/303433
Trust: David Chase <drchase@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
4 years agonet: make go resolver aware of network parameter
Zhang Boyang [Mon, 15 Mar 2021 15:27:51 +0000 (15:27 +0000)]
net: make go resolver aware of network parameter

Currently, the go resolver always send two DNS queries (A and AAAA) even
if tcp4/udp4/ip4 or tcp6/udp6/ip6 is used. This can cause unwanted
latencies when making IPv4-only or IPv6-only connections.

This change make go resolver aware of network parameter. Now, only one A
query is sent when tcp4/udp4/ip4 is used, and vice versa for
tcp6/udp6/ip6.

Fixes #45024

Change-Id: I815f909e6df5f7242cfc900f7dfecca628c3a2c8
GitHub-Last-Rev: 3d30c486dedd0d211366b1989034480a22ef2ffc
GitHub-Pull-Request: golang/go#45016
Reviewed-on: https://go-review.googlesource.com/c/go/+/301709
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Trust: Emmanuel Odeke <emmanuel@orijtech.com>

4 years agotest: enable fixedbugs/bug193.go for -G compiler option
Robert Griesemer [Fri, 19 Mar 2021 01:35:53 +0000 (18:35 -0700)]
test: enable fixedbugs/bug193.go for -G compiler option

Temporarily disable a questionable test case in fixedbugs/bug193.go
and enable the test as a whole. See the issues below for details.

Updates #45114.
Updates #45117.

Change-Id: I1de6f8d79b592eeeec139cd92b6c9cac56a9a74b
Reviewed-on: https://go-review.googlesource.com/c/go/+/303094
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Emmanuel Odeke <emmanuel@orijtech.com>
4 years agogo/types: remove superfluous code for shift checking
Robert Griesemer [Fri, 19 Mar 2021 00:29:02 +0000 (17:29 -0700)]
go/types: remove superfluous code for shift checking

Negative constant shift counts are already handled
earlier in the code. No need anymore for this extra
section.

With this change, the shift code matches types2
with respect to the function logic.

Change-Id: Ic8b7f382271c79ab66021e30955cd9bac092332b
Reviewed-on: https://go-review.googlesource.com/c/go/+/303093
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
4 years agogo/types, types2: add a test case for shifts that used to fail
Robert Griesemer [Thu, 18 Mar 2021 22:10:44 +0000 (15:10 -0700)]
go/types, types2: add a test case for shifts that used to fail

For #42989.

Change-Id: Ic354edde23ed7bfb4cb3eec2794f7213ed862b47
Reviewed-on: https://go-review.googlesource.com/c/go/+/303090
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
4 years agocmd/compile/internal/types2: refactor untyped conversions
Robert Griesemer [Thu, 18 Mar 2021 05:13:16 +0000 (22:13 -0700)]
cmd/compile/internal/types2: refactor untyped conversions

Based on https://golang.org/cl/284256 for go/types.
Brings this code more in line with go/types.

Adjusted various tests to match new error messages which
generally are now better: for assignment errors, instead
of a generic "cannot convert" we now say "cannot use"
followed by a clearer reason as to why not.

Major differences to go/types with respect to the changed
files:

- Some of the new code now returns error codes, but they
  are only used internally for now, and not reported with
  errors.

- go/types does not "convert" untyped nil values to target
  types, but here we do. This is unchanged from how types2
  handled this before this CL.

Change-Id: If45336d7ee679ece100f6d9d9f291a6ea55004d8
Reviewed-on: https://go-review.googlesource.com/c/go/+/302757
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
4 years agocmd/compile: replace calls to typecheck with transform functions
Dan Scales [Thu, 18 Mar 2021 21:36:39 +0000 (14:36 -0700)]
cmd/compile: replace calls to typecheck with transform functions

For additions, compares, and slices, create transform functions that do
just the transformations for those nodes by the typecheck package (given
that the code has been fully typechecked by types2). For nodes that have
no args with typeparams, we call these transform functions directly in
noder2. But for nodes that have args with typeparams, we have to delay
and call the tranform functions during stenciling, since we don't know
the specific types involved.

We indicate that a node still needs transformation by setting Typecheck
to a new value 3. This value means the current type of the node has been
set (via types2), but the node may still need transformation.

Had to export typcheck.IsCmp and typecheck.Assignop from the typecheck
package.

Added new tests list2.go (required delaying compare typecheck/transform
because of != compare in checkList) and adder.go (requires delaying add
typecheck/transform, since it can do addition for numbers or strings).

There are several more transformation functions needed for expressions
(indexing, calls, etc.) and several more complicated ones needed for
statements (mainly various kinds of assignments).

Change-Id: I7d89d13a4108308ea0304a4b815ab60b40c59b0a
Reviewed-on: https://go-review.googlesource.com/c/go/+/303091
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Dan Scales <danscales@google.com>
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
4 years agoruntime: support long paths without fixup on Windows 10 >= 1607
Jason A. Donenfeld [Thu, 11 Feb 2021 18:10:10 +0000 (19:10 +0100)]
runtime: support long paths without fixup on Windows 10 >= 1607

Windows 10 >= 1607 allows CreateFile and friends to use long paths if
bit 0x80 of the PEB's BitField member is set.

In time this means we'll be able to entirely drop our long path hacks,
which have never really worked right (see bugs below). Until that point,
we'll simply have things working well on recent Windows.

Updates #41734.
Updates #21782.
Updates #36375.

Change-Id: I765de6ea4859dd4e4b8ca80af7f337994734118e
Reviewed-on: https://go-review.googlesource.com/c/go/+/291291
Trust: Jason A. Donenfeld <Jason@zx2c4.com>
Run-TryBot: Jason A. Donenfeld <Jason@zx2c4.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>