Russ Cox [Fri, 10 Nov 2017 19:44:31 +0000 (14:44 -0500)]
cmd/go: cache and replay command output during build
It's nice that
go build -gcflags=-m errors
go build -gcflags=-m errors
uses the cache for the second command.
Even nicer is to make the second command
print the same output as the first command.
Fixes #22587.
Change-Id: I64350839f01c86c9a095d9d22f6924cd7a0b9105
Reviewed-on: https://go-review.googlesource.com/77110 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Daniel Martí [Sun, 12 Nov 2017 16:55:11 +0000 (16:55 +0000)]
cmd/compile: use stringer on types.EType
Another one that is possible thanks to the new -trimprefix stringer
flag.
The only subtle difference is that, in the previous version, some values
such as TUNSAFEPTR were stringified as "TUNSAFEPTR" instead of
"UNSAFEPTR". The new String method is always consistent in removing the
"T" prefix.
Passes toolstash -cmp on std cmd.
Change-Id: I68407f391795403dfcbbfa68c813018c0235bbb5
Reviewed-on: https://go-review.googlesource.com/77250
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com> Reviewed-by: Robert Griesemer <gri@golang.org>
Adam Langley [Mon, 30 Oct 2017 02:20:33 +0000 (19:20 -0700)]
crypto/x509: handle name constraints with cryptobyte
This allows better precision and (the motivation) empty strings to
be handled correctly. With that in place tests for the behaviour of
empty name constraints can be added.
Also fixes a compatibility issue with NSS. See #22616.
David du Colombier [Fri, 10 Nov 2017 20:46:41 +0000 (21:46 +0100)]
cmd/go: skip TestGoTestJSON on uniprocessor machines
CL 76873 added TestGoTestJSON. However, this test
is only succeeding on SMP machines.
This change skips TestGoTestJSON on uniprocessor machines.
Fixes #22665.
Change-Id: I3989d3331fb71193a25a3f0bbb84ff3e1b730890
Reviewed-on: https://go-review.googlesource.com/77130 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Alex Brainman [Mon, 28 Aug 2017 02:21:44 +0000 (12:21 +1000)]
net: ignore duplicate interfaces in TestInterfaceHardwareAddrWithGetmac
Sometimes getmac lists many interfaces for the same MAC address,
while Interfaces returns only single name for that address. Adjust
the test to ignore the names that are not returned by the Interfaces.
Russ Cox [Thu, 9 Nov 2017 15:29:23 +0000 (10:29 -0500)]
cmd/go: redefine -coverpkg to be a pattern list
If you run
go test -coverpkg=all fmt
one possible interpretation is that you want coverage for all the
packages involved in the fmt test, not all the packages in the world.
Because coverpkg was previously defined as a list of packages
to be loaded, however, it meant all packages in the world.
Now that the go command has a concept of package notation
being used as a matching filter instead of a direct enumeration,
apply that to -coverpkg, so that -coverpkg=all now has the
more useful filter interpretation.
Russ Cox [Thu, 9 Nov 2017 03:03:19 +0000 (22:03 -0500)]
cmd/go: add go test -json flag
This CL finally adds one of our longest-requested cmd/go features:
a way for test-running harnesses to access test output in structured form.
In fact the structured json output is more informative than the text
output, because the output from multiple parallel tests can be
interleaved as it becomes available, instead of needing to wait for
the previous test to finish before showing any output from the
next test.
Russ Cox [Thu, 9 Nov 2017 03:01:18 +0000 (22:01 -0500)]
cmd/test2json: go tool test2json converts test output to JSON
Also add cmd/internal/test2json, the actual implementation,
which will be called directly from cmd/go in addition to being
a standalone command (like cmd/buildid and cmd/internal/buildid).
Russ Cox [Fri, 10 Nov 2017 00:54:42 +0000 (19:54 -0500)]
cmd/go: guarantee all test output is on stdout
In past releases, whether test output appears on stdout or stderr
has varied depending on exactly how go test was invoked and
also (indefensibly) on the number of CPUs available.
Standardize on standard output for all test output.
This is easy to explain and makes go test | go tool test2json work nicely.
Daniel Martí [Tue, 7 Nov 2017 11:34:35 +0000 (11:34 +0000)]
cmd/vet: don't warn on escaped newlines in Println
The old code only worked for double-quoted strings, and only checked
that the end of the literal value was \n". This worked most of the time,
except for some strings like "foo\\n", which doesn't actually translate
into a trailing newline when unquoted.
To fix this, unquote the string first and look for a real newline at the
end of it. Ignore errors, as we don't have anything to do with string
literals using back quotes.
Bryan A Ford [Sat, 10 Jun 2017 07:34:40 +0000 (09:34 +0200)]
crypto/subtle: simplify and speed up constant-time primitives
This changes improves the ConstantTimeByteEq and ConstantTimeEq
primitives to both simplify them and improve their performance.
Also, since there were no benchmarks for this package before,
this change adds benchmarks for ConstantTimeByteEq,
ConstantTimeEq, and ConstantTimeLessOrEq.
benchmarks on darwin/amd64, 10 runs on old vs new code:
Alex Brainman [Thu, 9 Nov 2017 23:26:05 +0000 (10:26 +1100)]
cmd/go: close unintentionally left open file
cmd/go/internal/work.Builder.updateBuildID left a file opened.
But opened files cannot be deleted on Windows, so cmd/go just
leaves these files in %TMP% directory.
Close the file so deletion can succeed.
Fixes #22650
Change-Id: Ia3ea62f6ec7208d73972eae2e17fb4a766407914
Reviewed-on: https://go-review.googlesource.com/76810 Reviewed-by: Dave Cheney <dave@cheney.net>
Run-TryBot: Alex Brainman <alex.brainman@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Russ Cox [Wed, 8 Nov 2017 18:34:31 +0000 (13:34 -0500)]
cmd/go: always update mtime during go install / go build -o / go test -c
Even if the go command can see that the target is up-to-date
an mtime-based build system invoking the go command may not
be able to tell. Update the mtime to make clear that the target is
up-to-date, and also to hide exactly how smart the go command
is or is not. This keeps users (and programs) from depending on
the exact details of the go command's staleness determination.
Without this I believe we will get a stream of (completely reasonable)
bug reports that "go install (or go test -c) did not update the binary
after I trivially changed the source code or touched a source file".
Change-Id: I920e4aaed2a57319e3c0c37717f872bc059e484e
Reviewed-on: https://go-review.googlesource.com/76590
Run-TryBot: Russ Cox <rsc@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org> Reviewed-by: David Crawshaw <crawshaw@golang.org>
Russ Cox [Wed, 8 Nov 2017 16:54:34 +0000 (11:54 -0500)]
cmd/go: treat cached test results as satisfying any timeout
We want test caching to work even for people with scripts
that set a non-default test timeout. But then that raises the
question of what to do about runs with different timeouts:
is a cached success with one timeout available for use when
asked to run the test with a different timeout?
This CL answers that question by saying that the timeout applies
to the overall execution of either running the test or displaying
the cached result, and displaying a cached result takes no time.
So it's always OK to record a cached result, regardless of timeout,
and it's always OK to display a cached result, again regardless of timeout.
Fixes #22633.
Change-Id: Iaef3602710e3be107602267bbc6dba9a2250796c
Reviewed-on: https://go-review.googlesource.com/76552
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: roger peppe <rogpeppe@gmail.com> Reviewed-by: David Crawshaw <crawshaw@golang.org>
It has always been problematic that there was no way to specify
tool flags that applied only to the build of certain packages;
it was only to specify flags for all packages being built.
The usual workaround was to install all dependencies of something,
then build just that one thing with different flags. Since the
dependencies appeared to be up-to-date, they were not rebuilt
with the different flags. The new content-based staleness
(up-to-date) checks see through this trick, because they detect
changes in flags. This forces us to address the underlying problem
of providing a way to specify per-package flags.
The solution is to allow -gcflags=pattern=flags, which means
that flags apply to packages matching pattern, in addition to the
usual -gcflags=flags, which is now redefined to apply only to
the packages named on the command line.
David Chase [Sun, 29 Oct 2017 15:53:18 +0000 (11:53 -0400)]
cmd/compile: adjust Pos setting for "empty" blocks
Plain blocks that contain only uninteresting instructions
(that do not have reliable Pos information themselves)
need to have their Pos left unset so that they can
inherit it from their successors. The "uninteresting"
test was not properly applied and not properly defined.
OpFwdRef does not appear in the ssa.html debugging output,
but at the time of the test these instructions did appear,
and it needs to be part of the test.
Fixes #22365.
Change-Id: I99e5b271acd8f6bcfe0f72395f905c7744ea9a02
Reviewed-on: https://go-review.googlesource.com/74252
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
Filippo Valsorda [Tue, 31 Oct 2017 23:43:05 +0000 (19:43 -0400)]
crypto/tls: advertise support for SHA-512 signatures in 1.2
This is the equivalent change to 1c105980 but for SHA-512.
SHA-512 certificates are already supported by default since b53bb2ca,
but some servers will refuse connections if the algorithm is not
advertised in the overloaded signatureAndHash extension (see 09b238f1).
This required adding support for SHA-512 signatures on CertificateVerify
and ServerKeyExchange messages, because of said overloading.
Some testdata/Client-TLSv1{0,1} files changed because they send a 1.2
ClientHello even if the server picks a lower version.
Leigh McCulloch [Tue, 7 Nov 2017 05:33:35 +0000 (05:33 +0000)]
encoding/xml: add Marshal doc about name conflicts
The docs for xml.Marshal state that the XML elements name is derived
from one of five locations in a specific order of precedence, but does
not mention that if the field is a struct type and has its name defined
in a tag and in the types XMLName field that an error will occur. This
is documented in the structFieldInfo function but not in the function
documentation, and the existing docs in Marshal are misleading without
this behavior being discussed.
Fixes #18564
Change-Id: I29042f124a534bd1bc993f1baeddaa0af2e72fed
Reviewed-on: https://go-review.googlesource.com/76321 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Adam Langley [Sun, 15 Oct 2017 19:21:00 +0000 (12:21 -0700)]
crypto/x509: enforce EKU nesting at chain-construction time.
crypto/x509 has always enforced EKUs as a chain property (like CAPI, but
unlike the RFC). With this change, EKUs will be checked at
chain-building time rather than in a target-specific way.
Thus mis-nested EKUs will now cause a failure in Verify, irrespective of
the key usages requested in opts. (This mirrors the new behaviour w.r.t.
name constraints, where an illegal name in the leaf will cause a Verify
failure, even if the verified name is permitted.).
Alex Brainman [Tue, 7 Nov 2017 01:09:59 +0000 (12:09 +1100)]
syscall: change SysProcAttr.Token type to Token
CL 75253 introduced new SysProcAttr.Token field as Handle.
But we already have exact type for it - Token. Use Token
instead of Handle everywhere - it saves few type conversions
and provides better documentation for new API.
Change-Id: Ibc5407a234a1f49804de15a24b27c8e6a6eba7e0
Reviewed-on: https://go-review.googlesource.com/76314 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Alex Brainman [Mon, 23 Oct 2017 05:23:43 +0000 (16:23 +1100)]
net: use WSASocket instead of socket call
WSASocket (unlike socket call) allows to create sockets that
will not be inherited by child process. So call WSASocket to
save on using syscall.ForkLock and calling syscall.CloseOnExec.
Some very old versions of Windows do not have that functionality.
Call socket, if WSASocket failed, to support these.
Change-Id: I2dab9fa00d1a8609dd6feae1c9cc31d4e55b8cb5
Reviewed-on: https://go-review.googlesource.com/72590 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Adam Langley [Sun, 10 Sep 2017 00:05:41 +0000 (17:05 -0700)]
crypto/x509: enforce all name constraints and support IP, email and URI constraints
This change makes crypto/x509 enforce name constraints for all names in
a leaf certificate, not just the name being validated. Thus, after this
change, if a certificate validates then all the names in it can be
trusted – one doesn't have a validate again for each interesting name.
Making extended key usage work in this fashion still remains to be done.
Currently dead goroutines retain their assist credit. This credit can
be used if the goroutine gets recycled, but in general this can make
assist pacing over-aggressive by hiding an amount of credit
proportional to the number of exited (and not reused) goroutines.
Fix this "hidden credit" by flushing assist credit to the global
credit pool when a goroutine exits.
Ian Lance Taylor [Tue, 7 Nov 2017 04:51:36 +0000 (20:51 -0800)]
runtime: only call netpoll if netpollinited returns true
This fixes a race on old Linux kernels, in which we might temporarily
set epfd to an invalid value other than -1. It's also the right thing
to do. No test because the problem only occurs on old kernels.
Fixes #22606
Change-Id: Id84bdd6ae6d7c5d47c39e97b74da27576cb51a54
Reviewed-on: https://go-review.googlesource.com/76319
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Michael Munday [Tue, 7 Nov 2017 09:40:56 +0000 (04:40 -0500)]
cmd/compile: fix SSA immediate sign extension on s390x
The CMPWUconst op (32-bit unsigned comparison with immediate) takes
an unsigned immediate value. In SSA this should be sign extended to
64-bits to match the Int32 type given in the op and then zero
extended when producing the final assembly. Before this CL we were
zero extending in SSA which caused ssacheck to fail.
While we are here also ensure other 32-bit immediates are sign
extended in SSA.
Than McIntosh [Mon, 6 Nov 2017 20:33:31 +0000 (15:33 -0500)]
runtime/pprof: harden CPU profile test against smart backend
A couple of the CPU profiling testpoints make calls to helper
functions (cpuHog1, for example) where the computed value is always
thrown away by the caller without being used. A smart compiler back
end (in this case LLVM) can detect this fact and delete the contents
of the called function, which can cause tests to fail. Harden the test
slighly by passing in a value read from a global and insuring that the
caller stores the value back to a global; this prevents any optimizer
mischief.
Change-Id: Icbd6e3e32ff299c68a6397dc1404a52b21eaeaab
Reviewed-on: https://go-review.googlesource.com/76230
Run-TryBot: Than McIntosh <thanm@google.com> Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
Joe Tsai [Thu, 2 Nov 2017 20:53:16 +0000 (13:53 -0700)]
archive/zip: add FileHeader.NonUTF8 field
The NonUTF8 field provides users with a way to explictly tell the
ZIP writer to avoid setting the UTF-8 flag.
This is necessary because many readers:
1) (Still) do not support UTF-8
2) And use the local system encoding instead
Thus, even though character encodings other than CP-437 and UTF-8
are not officially supported by the ZIP specification, pragmatically
the world has permitted use of them.
When a non-standard encoding is used, it is the user's responsibility
to ensure that the target system is expecting the encoding used
(e.g., producing a ZIP file you know is used on a Chinese version of Windows).
We adjust the detectUTF8 function to account for Shift-JIS and EUC-KR
not being identical to ASCII for two characters.
We don't need an API for users to explicitly specify that they are encoding
with UTF-8 since all single byte characters are compatible with all other
common encodings (Windows-1256, Windows-1252, Windows-1251, Windows-1250,
IEC-8859, EUC-KR, KOI8-R, Latin-1, Shift-JIS, GB-2312, GBK) except for
the non-printable characters and the backslash character (all of which
are invalid characters in a path name anyways).
Fixes #10741
Change-Id: I9004542d1d522c9137973f1b6e2b623fa54dfd66
Reviewed-on: https://go-review.googlesource.com/75592
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Austin Clements [Thu, 2 Nov 2017 16:37:25 +0000 (12:37 -0400)]
cmd/compile: []T where T is go:notinheap does not need write barriers
Currently, assigning a []T where T is a go:notinheap type generates an
unnecessary write barrier for storing the slice pointer.
This fixes this by teaching HasHeapPointer that this type does not
have a heap pointer, and tweaking the lowering of slice assignments so
the pointer store retains the correct type rather than simply lowering
it to a *uint8 store.
Russ Cox [Mon, 6 Nov 2017 00:36:35 +0000 (19:36 -0500)]
cmd/dist: do not reinstall runtime/cgo with -tags lldb on ios builders
The cache will take care of keeping go test -tags lldb fast.
Installing runtime/cgo this way just makes all the checkNotStale
tests think runtime/cgo is out of date.
Russ Cox [Sun, 5 Nov 2017 22:09:54 +0000 (17:09 -0500)]
cmd/dist, cmd/cgo, cmd/go: allow per-goos/goarch default CC
Even though cmd/dist has historically distinguished "CC for gohostos/gohostarch"
from "CC for target goos/goarch", it has not recorded that distinction
for later use by cmd/cgo and cmd/go. Now that content-based staleness
includes the CC setting in the decision about when to rebuild packages,
the go command needs to know the details of which CC to use when.
Otherwise lots of things look out of date and (worse) may be rebuilt with
the wrong CC.
A related issue is that users may want to be able to build a toolchain
capable of cross-compiling for two different non-host targets, and
to date we've required that CC_FOR_TARGET apply to both.
This CL introduces CC_FOR_${GOOS}_${GOARCH}, so that you can
(for example) set CC_FOR_linux_arm and CC_FOR_linux_arm64
separately on a linux/ppc64 host and be able to cross-compile to
either arm or arm64 with the right toolchain.
Russ Cox [Sun, 5 Nov 2017 21:22:02 +0000 (16:22 -0500)]
cmd/go: drop runtime, runtime/internal/sys, runtime/internal/atomic, unsafe as deps of everything
This was a hack to make a new make.bash avoid reusing installed packages.
The new content-based staleness is precise enough not to need this hack;
now it's just causing unnecessary rebuilds: if a package doesn't import "runtime",
for example, it doesn't need to be recompiled when runtime changes.
(It does need to be relinked, and we still arrange that.)
Cherry Zhang [Mon, 6 Nov 2017 15:47:02 +0000 (10:47 -0500)]
cmd/compile: enable ssacheck for tests in ssa_test.go
I thought SSA check was enabled for those tests, but in fact it
was not. Enable it. So we have SSA check on for at least some
tests on all architectures.
Updates #22499.
Change-Id: I51fcdda3af7faab5aeb33bf46c6db309285ce42c
Reviewed-on: https://go-review.googlesource.com/76024 Reviewed-by: Keith Randall <khr@golang.org>
Joe Tsai [Mon, 28 Aug 2017 19:07:58 +0000 (12:07 -0700)]
archive/zip: add FileHeader.Modified field
The ModifiedTime and ModifiedDate fields are not expressive enough
for many of the time extensions that have since been added to ZIP,
nor are they easy to access since they in a legacy MS-DOS format,
and must be set and retrieved via the SetModTime and ModTime methods.
Instead, we add new field Modified of time.Time type that contains
all of the previous information and more.
Support for extended timestamps have been attempted before, but the
change was reverted because it provided no ability for the user to
specify the timezone of the legacy MS-DOS fields.
Technically the old API did not either, but users were manually offsetting
the timestamp to achieve the same effect.
The Writer now writes the legacy timestamps according to the timezone
of the FileHeader.Modified field. When the Modified field is set via
the SetModTime method, it is in UTC, which preserves the old behavior.
The Reader attempts to determine the timezone if both the legacy
and extended timestamps are present since it can compute the delta
between the two values.
Since Modified is a superset of the information in ModifiedTime and ModifiedDate,
we mark ModifiedTime, ModifiedDate, ModTime, and SetModTime as deprecated.
Fixes #18359
Change-Id: I29c6bc0a62908095d02740df3e6902f50d3152f1
Reviewed-on: https://go-review.googlesource.com/74970
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Tim Wright [Sun, 5 Nov 2017 02:35:23 +0000 (19:35 -0700)]
syscall: fix NaCl Link syscall error handling
The existing NaCl filesystem Link system call erroneously allowed
a caller to call Link on an existing target which violates the POSIX
standard and effectively corrupted the internal filesystem
representation.
Russ Cox [Sun, 5 Nov 2017 21:09:46 +0000 (16:09 -0500)]
cmd/go: fix corner case missed rebuild of binary
If the only thing changing in the binary is the embedded main.a action ID,
go install was declining to install the binary, but go list could see that the
binary needed reinstalling (was stale).
Russ Cox [Mon, 6 Nov 2017 01:30:35 +0000 (20:30 -0500)]
cmd/dist: read dependencies from source files
I do not remember why we require deps.go to have a hard-coded
copy of the dependency information for cmd/go, when we can
read it from the source files instead. The answer probably involves
cmd/dist once being a C program.
In any event, stop doing that, which will eliminate the builder-only
failures in the builder-only deps test.
Russ Cox [Sat, 4 Nov 2017 17:14:19 +0000 (13:14 -0400)]
cmd/go: do not print entire help text for unrecognized flag
I typed 'go list -josn' without realizing I'd mistyped json, and I was confused for
quite a while as to why I was staring at the 'go help json' text: the actual problem
(a missing flag) scrolls far off the screen. If people want the full text, they can
easily ask for it, but don't drown the important bit - unrecognized flag or other
improper usage - with pages of supporting commentary. The help text does not
help people who just need to be told about a typo.
Michael Munday [Fri, 31 Mar 2017 15:08:40 +0000 (11:08 -0400)]
crypto/aes: use s390x KMA instruction for AES-GCM if available
Adds support for the cipher message with authentication (KMA)
instruction added in message-security-assist extension 8. This
instruction encapsulates most of the operations required for
AES-GCM and is faster than executing the operations independently.
David du Colombier [Mon, 6 Nov 2017 06:35:42 +0000 (07:35 +0100)]
os: fix RemoveAll on large directories on Plan 9 and NaCl
On Plan 9, some file servers, like ramfs, handle the read
offset when reading directories. However, the offset isn't
valid anymore after directory entries have been removed
between successive calls to read.
This issue happens when os.RemoveAll is called on a
directory that doesn't fit on a single 9P response message.
In this case, the first part of the directory is read,
then directory entries are removed and the second read
will be incomplete because the read offset won't be valid
anymore. Consequently, the content of the directory will
only be partially removed.
We change RemoveAll to call fd.Seek(0, 0) before calling
fd.Readdirnames, so the read offset will always be reset
after removing the directory entries.
After adding TestRemoveAllLarge, we noticed the same issue
appears on NaCl and the same fix applies as well.
Fixes #22572.
Change-Id: Ifc76ea7ccaf0168c34dc8ec0f400dc04db1baf8f
Reviewed-on: https://go-review.googlesource.com/75974
Run-TryBot: David du Colombier <0intro@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Daniel Martí [Fri, 3 Nov 2017 18:14:37 +0000 (18:14 +0000)]
cmd/go: skip "exclude all Go files" error in fmt
Otherwise, one can't run "go fmt" on a directory containing Go files if
none of them are buildable (e.g. because of build tags). This is
counter-intuitive, as fmt will format all Go files anyway.
If we encounter such a load error, ignore it and carry on. All other
load errors, such as when a package can't be found, should still be
shown to the user.
Add a test for the two kinds of load errors. Use fmt -n so that any
changes to the formatting of the files in testdata don't actually get
applied. The load errors still occur with -n, so the test does its job.
David Chase [Fri, 27 Oct 2017 21:02:11 +0000 (17:02 -0400)]
cmd/compile: adjust locationlist lifetimes
A statement like
foo = bar + qux
might compile to
AX := AX + BX
resulting in a regkill for AX before this instruction.
The buggy behavior is to kill AX "at" this instruction,
before it has executed. (Code generation of no-instruction
values like RegKills applies their effects at the
next actual instruction emitted).
However, bar is still associated with AX until after the
instruction executes, so the effect of the regkill must
occur at the boundary between this instruction and the
next. Similarly, the new value bound to AX is not visible
until this instruction executes (and in the case of values
that require multiple instructions in code generation, until
all of them have executed).
The ranges are adjusted so that a value's start occurs
at the next following instruction after its evaluation,
and the end occurs after (execution of) the first
instruction following the end of the lifetime as a value.
(Notice the asymmetry; the entire value must be finished
before it is visible, but execution of a single instruction
invalidates. However, the value *is* visible before that
next instruction executes).
The test was adjusted to make it insensitive to the result
numbering for variables printed by gdb, since that is not
relevant to the test and makes the differences introduced
by small changes larger than necessary/useful.
The test was also improved to present variable probes
more intuitively, and also to allow explicit indication
of "this variable was optimized out"
David Chase [Tue, 24 Oct 2017 21:51:05 +0000 (17:51 -0400)]
cmd/compile: repair name propagation into aggregate parts
For structs, slices, strings, interfaces, etc, propagation of
names to their components (e.g., complex.real, complex.imag)
is fragile (depends on phase ordering) and not done right
for the "dec" pass.
The dec pass is subsumed into decomposeBuiltin,
and then names are pushed into the args of all
OpFooMake opcodes.
compile/ssa/debug_test.go was fixed to pay attention to
variable values, and the reference files include checks
for the fixes in this CL (which make debugging better).
Hugues Bruant [Sat, 21 Oct 2017 22:58:37 +0000 (15:58 -0700)]
cmd/compile: inline closures with captures
When inlining a closure with captured variables, walk up the
param chain to find the one that is defined inside the scope
into which the function is being inlined, and map occurrences
of the captures to temporary inlvars, similarly to what is
done for function parameters.
No noticeable impact on compilation speed and binary size.
Minor improvements to go1 benchmarks on darwin/amd64
Keith Randall [Sat, 4 Nov 2017 04:57:08 +0000 (21:57 -0700)]
bytes: reduce work in IndexNearPageBoundary test
This test was taking too long on ppc64x.
There were a few reasons.
The first is that the page size on ppc64x is 64k instead of 4k.
That's 16x more work.
The second is that the generic Index is pretty bad in this case.
It first calls IndexByte which does a bunch of setup work only to find
the byte we're looking for at index 0. Then it calls Equal which
has to look at the whole string to find a difference on the last byte.
To fix, just limit our attention to near the end of the page.
Alberto Donizetti [Fri, 3 Nov 2017 10:11:27 +0000 (11:11 +0100)]
cmd/compile: add mul by ±2ⁿ code-generation tests for arm/arm64
This change adds code generation tests for multiplication by ±2ⁿ for
arm and arm64, in preparation for a future CL which will remove the
relevant architecture-specific SSA rules (the reduction is already
performed by rules in generic.rules added in CL 36323).
Leigh McCulloch [Fri, 3 Nov 2017 05:48:43 +0000 (05:48 +0000)]
all: change github.com issue links to golang.org
The go repository contains a mix of github.com/golang/go/issues/xxxxx
and golang.org/issues/xxxxx URLs for references to issues in the issue
tracker. We should use one for consistency, and golang.org is preferred
in case the project moves the issue tracker in the future.
This reasoning is taken from a comment Sam Whited left on a CL I
recently opened: https://go-review.googlesource.com/c/go/+/73890.
In that CL I referenced an issue using its github.com URL, because other
tests in the file I was changing contained references to issues using
their github.com URL. Sam Whited left a comment on the CL stating I
should change it to the golang.org URL.
If new code is intended to reference issues via golang.org and not
github.com, existing code should be updated so that precedence exists
for contributors who are looking at the existing code as a guide for the
code they should write.
Tim Cooper [Fri, 3 Nov 2017 23:17:08 +0000 (20:17 -0300)]
hash: add marshaling, unmarshaling example
Example usage of functionality implemented in CL 66710.
Change-Id: I87d6e4d2fb7a60e4ba1e6ef02715480eb7e8f8bd
Reviewed-on: https://go-review.googlesource.com/76011
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Russ Cox [Fri, 3 Nov 2017 05:24:19 +0000 (01:24 -0400)]
cmd/go: do not install dependencies during "go install"
This CL makes "go install" behave the way many users expect:
install only the things named on the command line.
Future builds still run as fast, thanks to the new build cache (CL 75473).
To install dependencies as well (the old behavior), use "go install -i".
Actual definitions aside, what most users know and expect of "go install"
is that (1) it installs what you asked, and (2) it's fast, unlike "go build".
It was fast because it installed dependencies, but installing dependencies
confused users repeatedly (see for example #5065, #6424, #10998, #12329,
"go build" and "go test" so that they could be "fast" too, but that only
created new opportunities for confusion. We also had to add -installsuffix
and then -pkgdir, to allow "fast" even when dependencies could not be
installed in the usual place.
The recent introduction of precise content-based staleness logic means that
the go command detects the need for rebuilding packages more often than it
used to, with the consequence that "go install" rebuilds and reinstalls
dependencies more than it used to. This will create more new opportunities
for confusion and will certainly lead to more issues filed like the ones
listed above.
CL 75743 introduced a build cache, separate from the install locations.
That cache makes all operations equally incremental and fast, whether or
not the operation is "install" or "build", and whether or not "-i" is used.
Installing dependencies is no longer necessary for speed, it has confused
users in the past, and the more accurate rebuilds mean that it will confuse
users even more often in the future. This CL aims to end all that confusion
by not installing dependencies by default.
By analogy with "go build -i" and "go test -i", which still install
dependencies, this CL introduces "go install -i", which installs
dependencies in addition to the things named on the command line.
Russ Cox [Tue, 31 Oct 2017 17:00:51 +0000 (13:00 -0400)]
cmd/go: run vet automatically during go test
This CL adds an automatic, limited "go vet" to "go test".
If the building of a test package fails, vet is not run.
If vet fails, the test is not run.
The goal is that users don't notice vet as part of the "go test"
process at all, until vet speaks up and says something important.
This should help users find real problems in their code faster
(vet can just point to them instead of needing to debug a
test failure) and expands the scope of what kinds of things
vet can help with.
The "go vet" runs in parallel with the linking of the test binary,
so for incremental builds it typically does not slow the overall
"go test" at all: there's spare machine capacity during the link.
all.bash has less spare machine capacity. This CL increases
the time for all.bash on my laptop from 4m41s to 4m48s (+2.5%)
To opt out for a given run, use "go test -vet=off".
The vet checks used during "go test" are a subset of the full set,
restricted to ones that are 100% correct and therefore acceptable
to make mandatory. In this CL, that set is atomic, bool, buildtags,
nilfunc, and printf. Including printf is debatable, but I want to
include it for now and find out what needs to be scaled back.
(It already found one real problem in package os's tests that
previous go vet os had not turned up.)
Now that we can rely on type information it may be that printf
should make its function-name-based heuristic less aggressive
and have a whitelist of known print/printf functions.
Determining the exact set for Go 1.10 is #18085.
Running vet also means that programs now have to type-check
with both cmd/compile and go/types in order to pass "go test".
We don't start vet until cmd/compile has built the test package,
so normally the added go/types check doesn't find anything.
However, there is at least one instance where go/types is more
precise than cmd/compile: declared and not used errors involving
variables captured into closures.
This CL includes a printf fix to os/os_test.go and many declared
and not used fixes in the race detector tests.
Fixes #18084.
Change-Id: I353e00b9d1f9fec540c7557db5653e7501f5e1c9
Reviewed-on: https://go-review.googlesource.com/74356
Run-TryBot: Russ Cox <rsc@golang.org> Reviewed-by: Rob Pike <r@golang.org> Reviewed-by: David Crawshaw <crawshaw@golang.org>