This fixes a regression introduced with CL 192937. That change
was intended to fix a problem in arm and arm64 but also added
code to change the behavior in ppc64 and ppc64le even though the
error never occurred there. The change to function sigFetchG
assumes that the register holding 'g' could be clobbered by
vdso code when in fact 'g' is in R30 and that is nonvolatile
in the 64-bit PowerPC ELF ABI so would not be clobbered in vdso code.
So if this happens somehow the path it takes is incorrect,
falling through to a call to badsignal which doesn't seem right.
This regression caused intermittent hangs on the builder dashboard
for ppc64, and can be reproduced consistently when running os/signal
TestStress on some ppc64 systems.
I mentioned this problem is issue #34391 because I thought it was
related to another problem described there.
net: use case-insensitive host string comparison in TestLookupGmailNS
Some nameservers alter the case of NS records they return, e.g.
ns2.google.COm. or ns2.google.coM. Change TestLookupGmailNS to account
for this possibility by comparing host names in lower case.
Keith Randall [Mon, 16 Sep 2019 21:38:12 +0000 (14:38 -0700)]
runtime: allow the Go runtime to return multiple stack frames for a single PC
Upgrade the thread sanitizer to handle mid-stack inlining correctly.
We can now return multiple stack frames for each pc that the thread sanitizer
gives us to symbolize.
To fix #33309, we still need to modify the tsan library with its portion
of this fix, rebuild the .syso files on all supported archs, and check
them into runtime/race.
Daniel Martí [Thu, 19 Sep 2019 14:31:03 +0000 (15:31 +0100)]
cmd/compile: reduce rulegen's output by 200 KiB
First, renove unnecessary "// cond:" lines from the generated files.
This shaves off about ~7k lines.
Second, join "if cond { break }" statements via "||", which allows us to
deduplicate a large number of them. This shaves off another ~25k lines.
This change is not for readability or simplicity; but rather, to avoid
unnecessary verbosity that makes the generated files larger. All in all,
git reports that the generated files overall weigh ~200KiB less, or
about 2.7% less.
While at it, add a -trace flag to rulegen.
Updates #33644.
Change-Id: I3fac0290a6066070cc62400bf970a4ae0929470a
Reviewed-on: https://go-review.googlesource.com/c/go/+/196498
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
Andrew Medvedev [Sat, 21 Sep 2019 17:18:47 +0000 (20:18 +0300)]
crypto/x509: give type hint in error message in marshalPublicKey
Currently if type of public key is unsupported, error message is "only
RSA and ECDSA public keys supported". After adding Ed25519 this message
is no longer correct.
Moreover, it is superfluous because documentation for
MarshalPKIXPublicKey, CreateCertificateRequest and CreateCertificate
already lists supported public key types.
This CL removes unnecessary details from error message.
It also adds reporting the type of unsupported key, which helps
debugging cases when struct (instead of a pointer) to otherwise correct
public key is given.
net: close correct file descriptor when netpoll registration fails
In (*netFD).accept, if initializing the *netFD associated with the
new connection fails, the listen FD is closed, rather than the FD
associated with the new connection. Close the correct FD instead.
Fixes #34392
Change-Id: I7bf3469d661e6d30cbd4b12f5f5fd330a81a541b
Reviewed-on: https://go-review.googlesource.com/c/go/+/196778
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Robert Griesemer [Thu, 19 Sep 2019 21:58:10 +0000 (14:58 -0700)]
go/types: don't clone interface methods when embedding them
https://golang.org/cl/191257 significantly changed (and simplified)
the computation of interface method sets with embedded interfaces.
Specifically, when adding methods from an embedded interface, those
method objects (Func Objects) were cloned so that they could have a
different source position (the embedding position rather than the
original method position) for better error messages.
This causes problems for code that depends on the identity of method
objects that represent the same method, embedded or not.
This CL avoids the cloning. Instead, while computing the method set
of an interface, a position map is carried along that tracks
embedding positions. The map is not needed anymore after type-
checking.
Updates #34421.
Change-Id: I8ce188136c76fa70fba686711167db29a049f46d
Reviewed-on: https://go-review.googlesource.com/c/go/+/196561 Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Andrew Medvedev [Sat, 21 Sep 2019 12:47:34 +0000 (12:47 +0000)]
strings, bytes: clarify usage of EqualFolds
This clarifies meaning of "case folding" Unicode equality with more familiar "case insensitive" wording.
For case folding properties see ftp://ftp.unicode.org/Public/UNIDATA/CaseFolding.txt.
Martin Möhrmann [Mon, 9 Sep 2019 15:50:35 +0000 (17:50 +0200)]
compile: prefer an AND instead of SHR+SHL instructions
On modern 64bit CPUs a SHR, SHL or AND instruction take 1 cycle to execute.
A pair of shifts that operate on the same register will take 2 cycles
and needs to wait for the input register value to be available.
Large constants used to mask the high bits of a register with an AND
instruction can not be encoded as an immediate in the AND instruction
on amd64 and therefore need to be loaded into a register with a MOV
instruction.
However that MOV instruction is not dependent on the output register and
on many CPUs does not compete with the AND or shift instructions for
execution ports.
Using a pair of shifts to mask high bits instead of an AND to mask high
bits of a register has a shorter encoding and uses one less general
purpose register but is slower due to taking one clock cycle longer
if there is no register pressure that would make the AND variant need to
generate a spill.
For example the instructions emitted for (x & 1 << 63) before this CL are: 48c1ea3f SHRQ $0x3f, DX 48c1e23f SHLQ $0x3f, DX
after this CL the instructions are the same as GCC and LLVM use: 48b80000000000000080 MOVQ $0x8000000000000000, AX
4821d0 ANDQ DX, AX
Some platforms such as arm64 already have SSA optimization rules to fuse
two shift instructions back into an AND.
Removing the general rule to rewrite AND to SHR+SHL speeds up this benchmark:
var GlobalU uint
func BenchmarkAndHighBits(b *testing.B) {
x := uint(0)
for i := 0; i < b.N; i++ {
x &= 1 << 63
}
GlobalU = x
}
amd64/darwin on Intel(R) Core(TM) i7-3520M CPU @ 2.90GHz:
name old time/op new time/op delta
AndHighBits-4 0.61ns ± 6% 0.42ns ± 6% -31.42% (p=0.000 n=25+25):
'go run run.go -all_codegen -v codegen' passes with following adjustments:
ARM64: The BFXIL pattern ((x << lc) >> rc | y & ac) needed adjustment
since ORshiftRL generation fusing '>> rc' and '|' interferes
with matching ((x << lc) >> rc) to generate UBFX. Previously
ORshiftLL was created first using the shifts generated for (y & ac).
S390X: Add rules for abs and copysign to match use of AND instead of SHIFTs.
cmd/compile: optimize ssa if blocks for wasm architecture
Check for the next block and accordingly place the successor blocks.
This saves an additional jump instruction if the next block is any one
of the successor blocks.
While at it, inline the logic of goToBlock.
Reduces the size of pkg/js_wasm by 264 bytes.
Change-Id: I671ac4322e6edcb0d7e590dcca27e074268068d5
Reviewed-on: https://go-review.googlesource.com/c/go/+/195204
Run-TryBot: Agniva De Sarker <agniva.quicksilver@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Richard Musiol <neelance@gmail.com>
(*buf).string previously manually searched
through its underlying byte slice until we
encountered a '0'. This change instead uses
bytes.IndexByte that results in a speed up:
$ benchstat before.txt after.txt
name old time/op new time/op delta
BufString-8 257ns ± 1% 174ns ± 1% -32.37% (p=0.000 n=9+8)
name old speed new speed delta
BufString-8 495MB/s ± 1% 732MB/s ± 1% +47.76% (p=0.000 n=10+8)
name old alloc/op new alloc/op delta
BufString-8 162B ± 0% 162B ± 0% ~ (all equal)
name old allocs/op new allocs/op delta
BufString-8 3.00 ± 0% 3.00 ± 0% ~ (all equal)
Ian Lance Taylor [Mon, 16 Sep 2019 19:35:15 +0000 (12:35 -0700)]
runtime: avoid overflow in markrootBlock
In a position independent executable the data or BSS may be located
close to the end of memory. If it is placed closer than
rootBlockBytes, then the calculations in markrootBlock would overflow,
and the test that ensures that n is not larger than n0 would fail.
This would then cause scanblock to scan data that it shouldn't,
using an effectively random ptrmask, leading to program crashes.
No test because the only way to test it is to build a PIE and convince
the kernel to put the data section near the end of memory, and I don't
know how to do that. Or perhaps we could use a linker script, but that
is painful.
The new code is algebraically identical to the original code, but
avoids the potential overflow of b+rootBlockBytes.
Change-Id: Ieb4e5465174bb762b063d2491caeaa745017345e
Reviewed-on: https://go-review.googlesource.com/c/go/+/195717
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
Jay Conrod [Thu, 19 Sep 2019 16:58:02 +0000 (12:58 -0400)]
cmd/go: don't construct module version info if there are import errors
A precondition of modload.PackageBuildInfo is that its path and deps
arguments correspond to paths that have been loaded successfully with
modload.ImportPaths or one of the Load functions. load.Package.load
should not call PackageBuildInfo if there were any errors resolving
imports.
Fixes #34393
Change-Id: I107514f1c535885330ff266c85d3981b71b31c2d
Reviewed-on: https://go-review.googlesource.com/c/go/+/196520
Run-TryBot: Jay Conrod <jayconrod@google.com> Reviewed-by: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Richard Musiol [Thu, 12 Sep 2019 19:05:45 +0000 (21:05 +0200)]
cmd/compile: add 32 bit float registers/variables on wasm
Before this change, wasm only used float variables with a size of 64 bit
and applied rounding to 32 bit precision where necessary. This change
adds proper 32 bit float variables.
cmd/go/internal/modload: use a structured error for 'ambiguous import'
This consolidates the construction of 'ambiguous import' errors to a
single location, ensuring consistency, and lays the groundwork for
automatic resolution in the future.
While we're at it, change "found" to "found package" to try to make
the cause of the error clearer.
Updates #32128
Updates #27899
Change-Id: I14a93593320e5c60d20b0eb686d0d5355763c30c
Reviewed-on: https://go-review.googlesource.com/c/go/+/196298
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Jay Conrod <jayconrod@google.com>
cmd/go: change the gccgo 'package m' regression test to run in GOPATH mode
This test is failing in the builders due to the deployed versions of
gccgo not supporting module mode. However, the bug reproduced in
GOPATH mode too, so that mode should be fine for a regression test.
Updates #34358
Change-Id: I954132a96849e80e8783d4de10389fcab7b14af2
Reviewed-on: https://go-review.googlesource.com/c/go/+/196518
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Than McIntosh <thanm@google.com>
Than McIntosh [Wed, 18 Sep 2019 19:49:25 +0000 (15:49 -0400)]
cmd/link: enhance linker's dwarf test
Couple of changes to the linker's dwarf test, including:
- add some code to the DWARF tests inlining coverage to verify the
call_file attribute attached to inlined routine DIEs. If function
main.F is inlined into function main.G, we want to see that the
call_file attribute in the inlined routine DIE for main.F is the
same file as that reported for main.G.
- fix a glitch with the way the DW_AT_decl_file attribute was
being checked. The previous code relied on hard-coded indices
into the line table files table, which is very brittle (since
there is no requirement that files be ordered in any specific
way). Instead, add machinery to look up the actual file string
via the line table reader.
Change-Id: I44e71c69b6e676238cf4b805e7170de17b50939f
Reviewed-on: https://go-review.googlesource.com/c/go/+/196517
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Jeremy Faller <jeremy@golang.org>
cmd/go/internal/mvs: recompute build list in Reqs before minimizing
modload.MinReqs was passing modload.buildList to mvs.Reqs explicitly,
apparently as an optimization. However, we do not always have the
invariant that modload.buildList is complete: in particular, 'go mod
tidy' begins by reducing modload.buildList to only the set of modules
that provide packages to the build, which may be substantially smaller
than the final build list.
Other operations, such as 'go mod graph', do not load the entire
import graph, and therefore call Reqs with the unreduced build list.
Since Reqs retains modules according to a post-order traversal of the
list, an incomplete list may produce a different traversal order — and
therefore a different minimal solution, when multiple minimal
solutions exist. That caused 'go mod tidy' to produce different output
from other 'go' subcommands when certain patterns of dependencies are
present.
Since passing in the build list is only an optimization anyway, remove
the parameter and recompute the actual (complete) list at the
beginning of mvs.Reqs itself. That way, it is guaranteed to be
complete and in canonical order.
Fixes #34086
Change-Id: I3101bb81a1853c4a5e773010da3e44d2d90a570c
Reviewed-on: https://go-review.googlesource.com/c/go/+/193397
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Jay Conrod <jayconrod@google.com>
Jason A. Donenfeld [Wed, 18 Sep 2019 06:10:07 +0000 (00:10 -0600)]
syscall: avoid memory corruption in mksyscall_windows.go with *bool parameters
Windows type PBOOL is a pointer to a 4 byte value, where 0 means false
and not-0 means true. That means we should use uint32 here, not bool,
since Go bools can be 1 byte. Since a *bool is never a "real" valid
Windows type, converting on both in and out is probably sufficient,
since *bool shouldn't ever be used as something with significance for
its particular address.
Updates: #34364
Change-Id: I4c1b91cd9a39d91e23dae6f894b9a49f7fba2c0a
Reviewed-on: https://go-review.googlesource.com/c/go/+/196122
Run-TryBot: Jason A. Donenfeld <Jason@zx2c4.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
Gert Cuykens [Thu, 27 Jun 2019 01:04:21 +0000 (03:04 +0200)]
cmd/doc: add option to output a clean one-line symbol representation
Currently there is no way for go doc to output a clean
one-line symbol representation of types, functions, vars
and consts without documentation lines or other text lines
added.
For example `go doc fmt` has a huge introduction so if you
pass that to grep or fzf to search a symbol let say scan
`go doc fmt | grep scan` you get way to many false
positives.
Added a `-short` flag to be able to do
`go doc -short fmt | grep scan` instead which will result in
just the symbols you are looking for.
func Fscan(r io.Reader, a ...interface{}) (n int, err error)
func Fscanf(r io.Reader, format string, a ...interface{}) (n int, err error)
func Fscanln(r io.Reader, a ...interface{}) (n int, err error)
func Sscan(str string, a ...interface{}) (n int, err error)
func Sscanf(str string, format string, a ...interface{}) (n int, err error)
func Sscanln(str string, a ...interface{}) (n int, err error)
Fixes #32597
Change-Id: I77a73838adc512c8d1490f5a82075de6b0462a31
Reviewed-on: https://go-review.googlesource.com/c/go/+/184017
Run-TryBot: Andrew Bonventre <andybons@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Andrew Bonventre <andybons@golang.org>
cmd/go/internal/modload: do not prune the module root when walking directories
When walking filesystem paths to locate packages, we normally prune
out subdirectories with names beginning with ".", "_", or equal to
"testdata". However, we should not prune out such a directory if it is
at or above the module root, since its name is not part of the package
path.
Fixes #28481
Updates #27852
Change-Id: Ice82b1f908afaab50f5592f6c38ca6a0fe911edf
Reviewed-on: https://go-review.googlesource.com/c/go/+/196297
Run-TryBot: Bryan C. Mills <bcmills@google.com> Reviewed-by: Daniel Martí <mvdan@mvdan.cc> Reviewed-by: Jay Conrod <jayconrod@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
The current msanwrite() segfaults during libpreinit
when built with -msan on arm64. The cause is msancall()
in runtime/msan_arm64.s called by msanwrite() assumes
that it is always called with a valid g, leading to a
segfult.
Liz Rice [Wed, 20 Jun 2018 19:33:37 +0000 (20:33 +0100)]
regexp: add more examples for Regexp methods
Since I first started on this CL, most of the methods have had examples
added by other folks, so this is now one new example, and additions to
two existing examples for extra clarity.
The issue has a comment about not necessarily having examples for all
methods, but I recall finding this package pretty confusing when I first
used it, and having concrete examples would have really helped me
navigate all the different options. There are more
String methods with examples now, but I think seeing how the byte-slice
methods work could also be helpful to explain the differences.
Updates #21450
Change-Id: I27b4eeb634fb8ab59f791c0961cce79a67889826
Reviewed-on: https://go-review.googlesource.com/c/go/+/120145 Reviewed-by: Daniel Martí <mvdan@mvdan.cc> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Jean de Klerk [Thu, 20 Jun 2019 00:02:09 +0000 (18:02 -0600)]
reflect: give type hints in error messages
Currently, if you call various reflect methods you might get a panic with a
message like, "reflect: Field of non-struct type". Sometimes it's easy to
grok what's going on, but other times you need to laboriously go perform
reflect.ValueOf(myType).Kind().
This CL just adds that detail to the error message, saving debuggers the
extra step and making the error message more clear.
Change-Id: I7e0c211a3001e6b217b828cbcf50518080b5cb1e
Reviewed-on: https://go-review.googlesource.com/c/go/+/183097 Reviewed-by: Daniel Martí <mvdan@mvdan.cc> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Daniel Martí [Wed, 18 Sep 2019 15:33:54 +0000 (16:33 +0100)]
cmd/compile: parallelize one more bit of rulegen
'go tool trace' pointed at an obvious inefficiency; roughly the first
fifth of the program's life was CPU-heavy and making use of only one CPU
core at a time.
This was due to genOp being run before genLower. We did make genLower
use goroutines to parallelize the work between architectures, but we
didn't make genOp run in parallel too.
Do that. To avoid having two layers of goroutines, simply fire off all
goroutines from the main function, and inline genLower, since it now
becomes just two lines of code.
Overall, this shaves another ~300ms from 'go run *.go' on my laptop.
name old time/op new time/op delta
Rulegen 2.04s ± 2% 1.76s ± 2% -13.93% (p=0.008 n=5+5)
name old user-time/op new user-time/op delta
Rulegen 9.04s ± 1% 9.25s ± 1% +2.37% (p=0.008 n=5+5)
name old sys-time/op new sys-time/op delta
Rulegen 235ms ±14% 245ms ±16% ~ (p=0.690 n=5+5)
name old peak-RSS-bytes new peak-RSS-bytes delta
Rulegen 179MB ± 1% 190MB ± 2% +6.21% (p=0.008 n=5+5)
Austin Clements [Wed, 21 Aug 2019 20:21:06 +0000 (16:21 -0400)]
debug/dwarf: expose file name table from line table reader
Currently, the line table reader keeps the file name table internal.
However, there are various attributes like AttrDeclFile and
AttrCallFile whose value is an index into this table. Hence, in order
to interpret these attributes, we need access to the file name table.
This CL adds a method to LineReader that exposes the file table of the
current compilation unit in order to allow consumers to interpret
attributes that index into this table.
Change-Id: I6b64b815f23b3b0695036ddabe1a67c3954867dd
Reviewed-on: https://go-review.googlesource.com/c/go/+/192699
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Than McIntosh <thanm@google.com> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Austin Clements [Wed, 21 Aug 2019 19:05:26 +0000 (15:05 -0400)]
debug/dwarf: expose CU byte order
Currently, dwarf.Reader exposes the current compilation unit's address
size, but doesn't expose its byte order. Both are important for
decoding many attributes. For example, location descriptions include
addresses that are encoded in native form for the CU.
This CL exposes the byte order of the compilation unit in the same way
we already expose its address size, which makes it possible to decode
attributes containing native addresses.
Change-Id: I92f156818fe92b049d1dfc1613816bb1689cfadf
Reviewed-on: https://go-review.googlesource.com/c/go/+/192698
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Tom Thorogood [Sat, 14 Sep 2019 01:16:20 +0000 (01:16 +0000)]
net/http: fix HTTP/2 idle pool tracing
CL 140357 caused HTTP/2 connections to be put in the idle pool, but
failed to properly guard the trace.GotConn call in getConn. dialConn
returns a minimal persistConn with conn == nil for HTTP/2 connections.
This persistConn was then returned from queueForIdleConn and caused the
httptrace.GotConnInfo passed into GotConn to have a nil Conn field.
HTTP/2 connections call GotConn themselves so leave it for HTTP/2 to call
GotConn as is done directly below.
Than McIntosh [Wed, 18 Sep 2019 18:07:55 +0000 (14:07 -0400)]
cmd/go: fix buglet in alternate gccgo debug_modinfo recipe
Fix bug in previous CL 171768 -- with Go 1.13 the proper entry point
to call is runtime.setmodinfo, not runtime..z2fdebug.setmodinfo (this
changed when we moved from 1.12). [ Unclear why trybots and runs of
all.bash didn't catch this, but hand testing made it apparent. ]
Updates #30344.
Change-Id: I91f47bd0c279ad2d84875051be582818b13735b6
Reviewed-on: https://go-review.googlesource.com/c/go/+/196237
Run-TryBot: Than McIntosh <thanm@google.com> Reviewed-by: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Daniel Martí [Wed, 18 Sep 2019 16:13:46 +0000 (17:13 +0100)]
all: remove trailing whitespace from HTML files
I noticed lots of trailing whitespace in one of cmd/trace's HTML files.
While at it, remove a few others from still-maintained files. Leave old
documents alone, such as doc/devel/weekly.html.
Than McIntosh [Fri, 12 Apr 2019 13:47:43 +0000 (09:47 -0400)]
cmd/go: use alternate debug_modinfo recipe for gccgo
Use a different recipe for capturing debug modinfo if we're compiling
with the gccgo toolchain, to avoid applying a go:linkname directive to
a variable (not supported by gccgo).
Fixes #30344.
Change-Id: I9ce3d42c3bbb809fd68b140f56f9bbe3406c351b
Reviewed-on: https://go-review.googlesource.com/c/go/+/171768 Reviewed-by: Bryan C. Mills <bcmills@google.com>
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
cmd/asm,cmd/compile: clean up isel codegen on ppc64x
This cleans up the isel code generation in ssa for ppc64x.
Current there is no isel op and the isel code is only
generated from pseudo ops in ppc64/ssa.go, and only using
operands with values 0 or 1. When the isel is generated,
there is always a load of 1 into the temp register before it.
This change implements the isel op so it can be used in PPC64.rules,
and can recognize operand values other than 0 or 1. This also
eliminates the forced load of 1, so it will be loaded only if
needed.
This will make the isel code generation consistent with other ops,
and allow future rule changes that can take advantage of having
a more general purpose isel rule.
Change-Id: I363e1dbd3f7f5dfecb53187ad51cce409a8d1f8d
Reviewed-on: https://go-review.googlesource.com/c/go/+/195057
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Carlos Eduardo Seo <cseo@linux.vnet.ibm.com>
Matthew Dempsky [Fri, 13 Sep 2019 23:02:23 +0000 (16:02 -0700)]
cmd/compile: optimize switch on strings
When compiling expression switches, we try to optimize runs of
constants into binary searches. The ordering used isn't visible to the
application, so it's unimportant as long as we're consistent between
sorting and searching.
For strings, it's much cheaper to compare string lengths than strings
themselves, so instead of ordering strings by "si <= sj", we currently
order them by "len(si) < len(sj) || len(si) == len(sj) && si <= sj"
(i.e., the lexicographical ordering on the 2-tuple (len(s), s)).
However, it's also somewhat cheaper to compare strings for equality
(i.e., ==) than for ordering (i.e., <=). And if there were two or
three string constants of the same length in a switch statement, we
might unnecessarily emit ordering comparisons.
For example, given:
switch s {
case "", "1", "2", "3": // ordered by length then content
goto L
}
we currently compile this as:
if len(s) < 1 || len(s) == 1 && s <= "1" {
if s == "" { goto L }
else if s == "1" { goto L }
} else {
if s == "2" { goto L }
else if s == "3" { goto L }
}
This CL switches to using a 2-level binary search---first on len(s),
then on s itself---so that string ordering comparisons are only needed
when there are 4 or more strings of the same length. (4 being the
cut-off for when using binary search is actually worthwhile.)
So the above switch instead now compiles to:
if len(s) == 0 {
if s == "" { goto L }
} else if len(s) == 1 {
if s == "1" { goto L }
else if s == "2" { goto L }
else if s == "3" { goto L }
}
which is better optimized by walk and SSA. (Notably, because there are
only two distinct lengths and no more than three strings of any
particular length, this example ends up falling back to simply using
linear search.)
Test case by khr@ from CL 195138.
Fixes #33934.
Change-Id: I8eeebcaf7e26343223be5f443d6a97a0daf84f07
Reviewed-on: https://go-review.googlesource.com/c/go/+/195340
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
Robert Griesemer [Tue, 17 Sep 2019 23:32:11 +0000 (16:32 -0700)]
go/parser: return partial result from ParseExpr in case of error
Remove redundant code and improve documentation in the process.
Fixes #34211.
Change-Id: I9a6d1467f1a2c98a163f41f9df147fc6500c6fad
Reviewed-on: https://go-review.googlesource.com/c/go/+/196077 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Matthew Dempsky [Wed, 18 Sep 2019 01:32:04 +0000 (18:32 -0700)]
cmd/compile: remove OCASE and rename OXCASE to OCASE
We used to use OXCASE to represent general, possibly multi-valued
cases, and then desugar these during walk into single-value cases
represented by OCASE.
In CL 194660, we switched to eliminated the desugaring step and
instead handle the multi-valued cases directly, which eliminates the
need for an OCASE Op. Instead, we can simply remove OCASE, and rename
OXCASE to just OCASE.
Passes toolstash-check.
Change-Id: I3cc184340f9081d37453927cca1c059267fdbc12
Reviewed-on: https://go-review.googlesource.com/c/go/+/196117 Reviewed-by: Keith Randall <khr@golang.org>
Robert Griesemer [Mon, 16 Sep 2019 21:34:02 +0000 (14:34 -0700)]
go/types: make sure interfaces are complete before comparing them
Complete interfaces before comparing them with Checker.identical.
This requires passing through a *Checker to various functions that
didn't need this before.
Verified that none of the exported API entry points for interfaces
that rely on completed interfaces are used internally except for
Interface.Empty. Verified that interfaces are complete before
calling Empty on them, and added a dynamic check in the exported
functions.
Unfortunately, this fix exposed another problem with an esoteric
test case (#33656) which we need to reopen.
Fixes #34151.
Updates #33656.
Change-Id: I4e14bae3df74a2c21b565c24fdd07135f22e11c0
Reviewed-on: https://go-review.googlesource.com/c/go/+/195837
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Than McIntosh [Mon, 16 Sep 2019 20:11:01 +0000 (16:11 -0400)]
debug/elf: apply more relocations when reading DWARF data sections
The elf reader's method for reading in DWARF section data has support
for applying selected relocations when the debug/dwarf readers are
being used on relocatable objects. This patch extends the set of
relocations applied slightly. In particlar, prior to this for some
architectures we were only applying relocations whose target symbol
was a section symbol; now we also include some relocations that target
other symbols. This is needed to get meaningful values for compilation
unit DIE low_pc attributes, which typically target a specific function
symbol in text.
Fixes #31363.
Change-Id: I34b02e7904cd7f2dea74197f73fa648141d15212
Reviewed-on: https://go-review.googlesource.com/c/go/+/195679 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Daniel Martí [Tue, 17 Sep 2019 13:39:54 +0000 (14:39 +0100)]
cmd/compile: reduce the regexp work in rulegen
As correctly pointed out by Giovanni Bajo, doing a single regexp pass
should be much faster than doing hundreds per architecture. We can then
use a map to keep track of what ops are handled in each file. And the
amount of saved work is evident:
name old time/op new time/op delta
Rulegen 2.48s ± 1% 2.02s ± 1% -18.44% (p=0.008 n=5+5)
name old user-time/op new user-time/op delta
Rulegen 10.9s ± 1% 8.9s ± 0% -18.27% (p=0.008 n=5+5)
name old sys-time/op new sys-time/op delta
Rulegen 209ms ±28% 236ms ±18% ~ (p=0.310 n=5+5)
name old peak-RSS-bytes new peak-RSS-bytes delta
Rulegen 178MB ± 3% 176MB ± 3% ~ (p=0.548 n=5+5)
The speed-up is so large that we don't need to parallelize it anymore;
the numbers above are with the removed goroutines. Adding them back in
doesn't improve performance noticeably at all:
name old time/op new time/op delta
Rulegen 2.02s ± 1% 2.01s ± 1% ~ (p=0.421 n=5+5)
name old user-time/op new user-time/op delta
Rulegen 8.90s ± 0% 8.96s ± 1% ~ (p=0.095 n=5+5)
While at it, remove an unused method.
Change-Id: I328b56e63b64a9ab48147e67e7d5a385c795ec54
Reviewed-on: https://go-review.googlesource.com/c/go/+/195739
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
LE Manh Cuong [Sun, 16 Jun 2019 18:15:53 +0000 (01:15 +0700)]
cmd/compile: support more length types for slice extension optimization
golang.org/cl/109517 optimized the compiler to avoid the allocation for make in
append(x, make([]T, y)...). This was only implemented for the case that y has type int.
This change extends the optimization to trigger for all integer types where the value
is known at compile time to fit into an int.
name old time/op new time/op delta
ExtendInt-12 106ns ± 4% 106ns ± 0% ~ (p=0.351 n=10+6)
ExtendUint64-12 1.03µs ± 5% 0.10µs ± 4% -90.01% (p=0.000 n=9+10)
name old alloc/op new alloc/op delta
ExtendInt-12 0.00B 0.00B ~ (all equal)
ExtendUint64-12 13.6kB ± 0% 0.0kB -100.00% (p=0.000 n=10+10)
name old allocs/op new allocs/op delta
ExtendInt-12 0.00 0.00 ~ (all equal)
ExtendUint64-12 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10)
Updates #29785
Change-Id: Ief7760097c285abd591712da98c5b02bc3961fcd
Reviewed-on: https://go-review.googlesource.com/c/go/+/182559
Run-TryBot: Cuong Manh Le <cuong.manhle.vn@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
Matthew Dempsky [Tue, 10 Sep 2019 23:03:03 +0000 (16:03 -0700)]
cmd/compile: major refactoring of switch walking
There are a lot of complexities to handling switches efficiently:
1. Order matters for expression switches with non-constant cases and
for type expressions with interface types. We have to respect
side-effects, and we also can't allow later cases to accidentally take
precedence over earlier cases.
2. For runs of integers, floats, and string constants in expression
switches or runs of concrete types in type switches, we want to emit
efficient binary searches.
3. For runs of consecutive integers in expression switches, we want to
collapse them into range comparisons.
4. For binary searches of strings, we want to compare by length first,
because that's more efficient and we don't need to respect any
particular ordering.
5. For "switch true { ... }" and "switch false { ... }", we want to
optimize "case x:" as simply "if x" or "if !x", respectively, unless x
is interface-typed.
The current swt.go code reflects how these constraints have been
incrementally added over time, with each of them being handled ad
hocly in different parts of the code. Also, the existing code tries
very hard to reuse logic between expression and type switches, even
though the similarities are very superficial.
This CL rewrites switch handling to better abstract away the logic
involved in constructing the binary searches. In particular, it's
intended to make further optimizations to switch dispatch much easier.
It also eliminates the need for both OXCASE and OCASE ops, and a
subsequent CL can collapse the two.
Passes toolstash-check.
Change-Id: Ifcd1e56f81f858117a412971d82e98abe7c4481f
Reviewed-on: https://go-review.googlesource.com/c/go/+/194660
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
Daniel Martí [Fri, 13 Sep 2019 13:42:05 +0000 (14:42 +0100)]
cmd/compile: parallelize another big chunk of rulegen
rulegen has a sanity check that ensures all the arch-specific opcodes
are handled by each of the gen files.
This is an expensive chunk of work, particularly since there are a lot
of opcodes in total, and each one of them compiles and runs a regular
expression.
Parallelize that for each architecture, which greatly speeds up 'go run
*.go' on my laptop with four real CPU cores.
name old time/op new time/op delta
Rulegen 3.39s ± 1% 2.53s ± 2% -25.34% (p=0.008 n=5+5)
name old user-time/op new user-time/op delta
Rulegen 10.6s ± 1% 11.2s ± 1% +6.09% (p=0.008 n=5+5)
name old sys-time/op new sys-time/op delta
Rulegen 201ms ± 7% 218ms ±17% ~ (p=0.548 n=5+5)
name old peak-RSS-bytes new peak-RSS-bytes delta
Rulegen 182MB ± 3% 184MB ± 3% ~ (p=0.690 n=5+5)
Change-Id: Iec538ed0fa7eb867eeeeaab3da1e2615ce32cbb9
Reviewed-on: https://go-review.googlesource.com/c/go/+/195218
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
Matthew Dempsky [Mon, 16 Sep 2019 18:28:05 +0000 (11:28 -0700)]
cmd/compile: require -lang=go1.14 for overlapping interfaces
Support for overlapping interfaces is a new (proposed) Go language
feature to be supported in Go 1.14, so it shouldn't be supported under
-lang=go1.13 or earlier.
Fixes #34329.
Change-Id: I5fea5716b7d135476980bc40b4f6e8c611b67735
Reviewed-on: https://go-review.googlesource.com/c/go/+/195678
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
Jay Conrod [Fri, 13 Sep 2019 17:58:58 +0000 (13:58 -0400)]
cmd/go: don't include package dir in cache key when -trimpath is set
The '-trimpath' flag tells 'go build' to trim any paths from the
output files that are tied to the current workspace or toolchain. When
this flag is set, we do not need to include the package directory in
the text hashed to construct the action ID for each package.
Fixes #33772
Change-Id: I20b902d2f58019709b15864ca79aa0d9255ae707
Reviewed-on: https://go-review.googlesource.com/c/go/+/195318
Run-TryBot: Jay Conrod <jayconrod@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Bryan C. Mills <bcmills@google.com>
Matthew Dempsky [Thu, 12 Sep 2019 17:18:03 +0000 (10:18 -0700)]
cmd/compile: trim function name prefix from escape diagnostics
This information is redundant with the position information already
provided. Also, no other -m diagnostics print out function name.
While here, report parameter leak diagnostics against the parameter
declaration position rather than the function, and use Warnl for
"moved to heap" messages.
Test cases updated programmatically by removing the first word from
every "no match for" error emitted by run.go:
go run run.go |& \
sed -E -n 's/^(.*):(.*): no match for `([^ ]* (.*))` in:$/\1!\2!\3!\4/p' | \
while IFS='!' read -r fn line before after; do
before=$(echo "$before" | sed 's/[.[\*^$()+?{|]/\\&/g')
after=$(echo "$after" | sed -E 's/(\&|\\)/\\&/g')
fn=$(find . -name "${fn}" | head -1)
sed -i -E -e "${line}s/\"${before}\"/\"${after}\"/" "${fn}"
done
Ian Lance Taylor [Thu, 12 Sep 2019 00:57:58 +0000 (17:57 -0700)]
os/signal: split up sleeps waiting for signal
Try to deflake TestNohup.
The kernel will deliver a signal as a thread returns from a syscall.
If the only active thread is sleeping, and the system is busy,
the kernel may not get around to waking up a thread to catch the signal.
Try splitting up the sleep, to give the kernel another change to deliver.
I don't know if this will help, but it seems worth a try.
Fixes #33174
Change-Id: I34b3240af706501ab8538cb25c4846d1d30d7691
Reviewed-on: https://go-review.googlesource.com/c/go/+/194879 Reviewed-by: Bryan C. Mills <bcmills@google.com>
Lynn Boger [Tue, 25 Jun 2019 17:15:59 +0000 (13:15 -0400)]
doc: update ppc64 section for asm.html
Update the section in asm.html related to PPC64. Remove the line
that says it is in an experimental state, add a link to the
new doc.go file that has all the detail for the Go assembler for
PPC64.
Change-Id: I45d9891669e01d94e2721be576d572e02cd9d2db
Reviewed-on: https://go-review.googlesource.com/c/go/+/183840
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com> Reviewed-by: Carlos Eduardo Seo <cseo@linux.vnet.ibm.com> Reviewed-by: Daniel Martí <mvdan@mvdan.cc> Reviewed-by: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Lucas Bremgartner [Fri, 13 Sep 2019 19:46:50 +0000 (19:46 +0000)]
encoding/json: make Number with the ,string option marshal with quotes
Add quotes when marshaling a json.Number with the string option
set via a struct tag. This ensures that the resulting json
can be unmarshaled into the source struct without error.
Fixes #34268
Change-Id: Ide167d9dec77019554870b5957b37dc258119d81
GitHub-Last-Rev: dde81b71208be01c253bb87dbb6f81ac6e0785be
GitHub-Pull-Request: golang/go#34269
Reviewed-on: https://go-review.googlesource.com/c/go/+/195043 Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Cherry Zhang [Wed, 28 Aug 2019 18:42:06 +0000 (14:42 -0400)]
test/codegen: document -all_codegen option in README
It is useful to know about the -all_codegen option for running
codegen tests for all platforms. I was puzzling that some codegen
test was not failing on my local machine or on trybot, until I
found this option.
Change-Id: I062cf4d73f6a6c9ebc2258195779d2dab21bc36d
Reviewed-on: https://go-review.googlesource.com/c/go/+/192101 Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
The wasm_exec.js wrapper tries to set up the argv and envp following
the UNIX conventions, but doesn't get it quite right, which can cause
runtime.goenv to crash if you get unlucky.
The main problem was that the envp array wasn't terminated with a nil
pointer, so the runtime didn't know when to stop reading the array.
This CL adds that nil pointer to the end of the envp array.
The other problem was harmless, but confusing. In the UNIX convention,
the argv array consists of argc pointers followed by a nil pointer,
followed by the envp array. However, wasm_exec.js put the environment
variable count between the two pointer arrays rather than a nil
pointer. The runtime never looks at this slot, so it didn't matter,
but the break from convention left Cherry and I trying to debug why it
*wasn't* losing any environment variables before we realized that that
layouts happened to be close enough to work. This CL switches to the
UNIX convention of simply terminating the argv array with a nil
pointer.
Tim Cooper [Mon, 15 Jul 2019 22:54:23 +0000 (17:54 -0500)]
log: add Lmsgprefix flag
The Lmsgprefix flag moves the logger's prefix from the
beginning of the line to after the log header. For example,
a logger with the prefix "LOG " and LstdFlags would output:
LOG 2009/11/10 23:00:00 entry text
Adding the Lmsgprefix flag would output:
2009/11/10 23:00:00 LOG entry text
Fixes #32062
Change-Id: I9f7c9739abeb53c424112aaeed33444eeefdfbbc
Reviewed-on: https://go-review.googlesource.com/c/go/+/186182
Run-TryBot: Rob Pike <r@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Rob Pike <r@golang.org>
Jay Conrod [Fri, 13 Sep 2019 15:10:36 +0000 (11:10 -0400)]
cmd/go: fix link error for -coverpkg in GOPATH mode
If a generated test main package transitively depends on a main
package, the main package will now always be rebuilt as a library and
will not be compiled with '-p main'.
This expands the fix for #30907, which only applied to packages with
the BuildInfo set (main packages built in module mode). Linking
multiple packages with BuildInfo caused link errors, but it appears
these errors apply to some symbols in GOPATH mode.
Fixes #34114
Change-Id: Ic1e53437942269a950dd7e45d163707922c92edd
Reviewed-on: https://go-review.googlesource.com/c/go/+/195279
Run-TryBot: Jay Conrod <jayconrod@google.com> Reviewed-by: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Joel Sing [Thu, 12 Sep 2019 18:33:06 +0000 (04:33 +1000)]
cmd/link: simplify determineLinkMode
Simplify determineLinkMode by calling mustLinkExternal upfront,
then doing a first pass for LinkModeAuto, followed by a second pass
that determines if the link mode is valid.
Change-Id: I9d7668107c159f8fe330b8c05fee035bbe9875fd
Reviewed-on: https://go-review.googlesource.com/c/go/+/195078 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Ruixin Bao [Sun, 8 Sep 2019 22:50:24 +0000 (18:50 -0400)]
cmd/compile: add math/bits.Mul64 intrinsic on s390x
This change adds an intrinsic for Mul64 on s390x. To achieve that,
a new assembly instruction, MLGR, is introduced in s390x/asmz.go. This assembly
instruction directly uses an existing instruction on Z and supports multiplication
of two 64 bit unsigned integer and stores the result in two separate registers.
In this case, we require the multiplcand to be stored in register R3 and
the output result (the high and low 64 bit of the product) to be stored in
R2 and R3 respectively.
Joel Sing [Sat, 7 Sep 2019 15:56:26 +0000 (01:56 +1000)]
cmd/asm,cmd/internal/obj: initial support for riscv64 assembler
Provide the initial framework for the riscv64 assembler. For now this
only supports raw WORD instructions, but at least allows for basic
testing. Additional functionality will be added in separate changes.
This change updates the status of gccgo 8 and gccgo 9, which are now
released.
It also replaces every instance of two-spaces with one space in text
paragraphs, which is the preferred style in Go documentation.
Fixes #34167
Change-Id: I94a4d85c06281f2623d39a68db7b6c95b5867999
Reviewed-on: https://go-review.googlesource.com/c/go/+/193842 Reviewed-by: Ian Lance Taylor <iant@golang.org>