Russ Cox [Tue, 8 Mar 2022 23:16:35 +0000 (18:16 -0500)]
cmd/compile: require -p flag
The -p flag specifies the import path of the package being compiled.
This CL makes it required when invoking the compiler and
adjusts tests that invoke the compiler directly to conform to this
new requirement. The go command already passes the flag, so it
is unmodified in this CL. It is expected that any other Go build systems
also already pass -p, or else they will need to arrange to do so before
updating to Go 1.19. Of particular note, Bazel already does for rules
with an importpath= attribute, which includes all Gazelle-generated rules.
There is more cleanup possible now in cmd/compile, cmd/link,
and other consumers of Go object files, but that is left to future CLs.
Additional historical background follows but can be ignored.
Long ago, before the go command, or modules, or any kind of
versioning, symbols in Go archive files were named using just the
package name, so that for example func F in math/rand and func F in
crypto/rand would both be the object file symbol 'rand.F'. This led to
collisions even in small source trees, which made certain packages
unusable in the presence of other packages and generally was a problem
for Go's goal of scaling to very large source trees.
Fixing this problem required changing from package names to import
paths in symbol names, which was mostly straightforward. One wrinkle,
though, is that the compiler did not know the import path of the
package being compiled; it only knew the package name. At the time,
there was no go command, just Makefiles that people had invoking 6g
(now “go tool compile”) and then copying the resulting object file to
an importable location. That is, everyone had a custom build setup for
Go, because there was no standard one. So it was not particularly
attractive to change how the compiler was invoked, since that would
break approximately every Go user at the time. Instead, we arranged
for the compiler to emit, and other tools reading object files to
recognize, a special import path (the empty string, it turned out)
denoting “the import path of this object file”. This worked well
enough at the time and maintained complete command-line compatibility
with existing Go usage.
The changes implementing this transition can be found by searching
the Git history for “package global name space”, which is what they
eliminated. In particular, CL 190076 (a6736fa4), CL 186263 (758f2bc5),
CL 193080 (1cecac81), CL 194053 (19126320), and CL 194071 (531e6b77)
did the bulk of this transformation in January 2010.
Later, in September 2011, we added the -p flag to the compiler for
diagnostic purposes. The problem was that it was easy to create import
cycles, especially in tests, and these could not be diagnosed until
link time. You'd really want the compiler to diagnose these, for
example if the compilation of package sort noticed it was importing a
package that itself imported "sort". But the compilation of package
sort didn't know its own import path, and so it could not tell whether
it had found itself as a transitive dependency. Adding the -p flag
solved this problem, and its use was optional, since the linker would
still diagnose the import cycle in builds that had not updated to
start passing -p. This was CL 4972057 (1e480cd1).
There was still no go command at this point, but when we introduced
the go command we made it pass -p, which it has for many years at this
point.
Over time, parts of the compiler began to depend on the presence of
the -p flag for various reasonable purposes. For example:
In CL 6497074 (041fc8bf; Oct 2012), the race detector used -p to
detect packages that should not have race annotations, such as
runtime/race and sync/atomic.
In CL 13367052 (7276c02b; Sep 2013), a bug fix used -p to detect the
compilation of package reflect.
In CL 30539 (8aadcc55; Oct 2016), the compiler started using -p to
identify package math, to be able to intrinsify calls to Sqrt inside
that package.
In CL 61019 (9daee931; Sep 2017), CL 71430 (2c1d2e06; Oct 2017), and
later related CLs, the compiler started using the -p value when
creating various DWARF debugging information.
In CL 174657 (cc5eaf93; May 2019), the compiler started writing
symbols without the magic empty string whenever -p was used, to reduce
the amount of work required in the linker.
In CL 179861 (dde7c770; Jun 2019), the compiler made the second
argument to //go:linkname optional when -p is used, because in that
case the compiler can derive an appropriate default.
There are more examples. Today it is impossible to compile the Go
standard library without using -p, and DWARF debug information is
incomplete without using -p.
All known Go build systems pass -p. In particular, the go command
does, which is what nearly all Go developers invoke to build Go code.
And Bazel does, for go_library rules that set the importpath
attribute, which is all rules generated by Gazelle.
Gccgo has an equivalent of -p and has required its use in order to
disambiguate packages with the same name but different import paths
since 2010.
On top of all this, various parts of code generation for generics
are made more complicated by needing to cope with the case where -p
is not specified, even though it's essentially always specified.
In summary, the current state is:
- Use of the -p flag with cmd/compile is required for building
the standard library, and for complete DWARF information,
and to enable certain linker speedups.
- The go command and Bazel, which we expect account for just
about 100% of Go builds, both invoke cmd/compile with -p.
- The code in cmd/compile to support builds without -p is
complex and has become more complex with generics, but it is
almost always dead code and therefore not worth maintaining.
- Gccgo already requires its equivalent of -p in any build
where two packages have the same name.
All this supports the change in this CL, which makes -p required
and adjusts tests that invoke cmd/compile to add -p appropriately.
Future CLs will be able to remove all the code dealing with the
possibility of -p not having been specified.
Rhys Hiltner [Wed, 9 Mar 2022 15:56:04 +0000 (07:56 -0800)]
runtime/pprof: fix pcDeck's frame indexing
When building the inlining deck, correctly identify which is the last
frame in the deck. Otherwise, when some forms of inlining cause a PC to
expand to multiple frames, the length of the deck's two slices will
diverge.
Fixes #51567
Change-Id: I24e7ba32cb16b167f4307178b3f03c29e5362c4b
Reviewed-on: https://go-review.googlesource.com/c/go/+/391134 Reviewed-by: Michael Pratt <mpratt@google.com>
Trust: Than McIntosh <thanm@google.com>
Matthew Dempsky [Tue, 8 Mar 2022 22:59:32 +0000 (14:59 -0800)]
cmd/compile: mark instantiated generic functions as DUPOK
Unified IR wasn't marking instantiated generic functions as DUPOK,
even though they can appear in multiple compilation units, which
evidently interfered with cmd/link's dead code elimination logic.
Manually confirmed to fix the issue, but non-trivial to test within
$GOROOT/test currently, because it's only reproducible when
cmd/compile is invoked with -p. @rsc is currently investigating
updating test/run.go appropriately, after which I'll revisit writing a
test case.
Fixes #51519.
Change-Id: I74a79ed0ca15b25b826e419714af5ceb6e567012
Reviewed-on: https://go-review.googlesource.com/c/go/+/390956
Trust: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Cherry Mui <cherryyz@google.com>
Each sequence uses two load or store instructions, actually we can load
or store two registers with one LDP or STP instruction. This CL changes
the sequences as follow.
Robert Findley [Tue, 8 Mar 2022 21:43:47 +0000 (16:43 -0500)]
go/printer: don't print unnecesary commas for func type param lists
Type parameter lists are not ambiguous for function declarations in the
way that they are ambiguous for type declarations. Avoid printing an
extra comma to disambiguate.
Fixes #51548
Change-Id: I8ca2b21e271982013653b9e220f92ee74f577ba2
Reviewed-on: https://go-review.googlesource.com/c/go/+/390914
Trust: Robert Findley <rfindley@google.com>
Run-TryBot: Robert Findley <rfindley@google.com> Reviewed-by: Robert Griesemer <gri@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Bryan C. Mills [Tue, 8 Feb 2022 20:24:33 +0000 (15:24 -0500)]
net/http/pprof: skip TestDeltaProfile on all arm and arm64 architectures
Given that we have seen failures with the same failure mode on both
openbsd/arm and android/arm64, it seems likely that the underlying bug
affects at least all ARM-based architectures.
It appears that either these architectures are not able to sample at
the frequency expected by the test, or the samples are for some reason
being dropped.
Michael Pratt [Mon, 13 Dec 2021 22:32:07 +0000 (17:32 -0500)]
cmd/dist: log CPU model when testing
Knowing whether test failures are correlated with specific CPU models on
has proven useful on several issues. Log it for prior to testing so it
is always available.
internal/sysinfo provides the CPU model, but it is not available in the
bootstrap toolchain, so we can't access this in cmd/dist. Instead use a
separate binary which cmd/dist will only build once testing begins.
The addition of new data to the beginning of cmd/dist output will break
x/build/cmd/coordinator's banner parsing, leaving extra lines in the log
output, though information will not be lost.
https://golang.org/cl/372538 fixes up the coordinator and should be
submitted and deployed before this CL is submitted.
This is a redo of CL 371474. It switches back to the original approach
of using a separate binary, as the bootstap toolchain won't allow
cmd/dist to import internal packages.
For #46272.
For #49209.
For #50146.
Change-Id: I906bbda987902a2120c5183290a4e89a2440de58
Reviewed-on: https://go-review.googlesource.com/c/go/+/378589 Reviewed-by: Austin Clements <austin@google.com>
Trust: Michael Pratt <mpratt@google.com>
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Rhys Hiltner [Tue, 8 Feb 2022 22:51:41 +0000 (14:51 -0800)]
runtime/pprof: check if PC is reused for inlining
When describing call stacks that include inlined function calls, the
runtime uses "fake" PCs to represent the frames that inlining removed.
Those PCs correspond to real NOP instructions that the compiler inserts
for this purpose.
Describing the call stack in a protobuf-formatted profile requires the
runtime/pprof package to collapse any sequences of fake call sites back
into single PCs, removing the NOPs but retaining their line info.
But because the NOP instructions are part of the function, they can
appear as leaf nodes in a CPU profile. That results in an address that
should sometimes be ignored (when it appears as a call site) and that
sometimes should be present in the profile (when it is observed
consuming CPU time).
When processing a PC address, consider it first as a fake PC to add to
the current inlining deck, and then as a previously-seen (real) PC.
Fixes #50996
Change-Id: I80802369978bd7ac9969839ecfc9995ea4f84ab4
Reviewed-on: https://go-review.googlesource.com/c/go/+/384239 Reviewed-by: Cherry Mui <cherryyz@google.com> Reviewed-by: Michael Pratt <mpratt@google.com>
Tobias Klauser [Tue, 8 Mar 2022 13:55:26 +0000 (14:55 +0100)]
runtime, syscall: implement syscall.Pipe using syscall.Pipe2 on solaris
All other platforms providing the pipe2 syscall already implement it
that way. Do so as well on solaris, which allows to drop
runtime.syscall_pipe and its dependencies as well.
Change-Id: Icf04777f21d1804da74325d173fefdc87caa42eb
Reviewed-on: https://go-review.googlesource.com/c/go/+/390716
Trust: Tobias Klauser <tobias.klauser@gmail.com>
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
Trust: Matt Layher <mdlayher@gmail.com>
TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Bryan C. Mills [Fri, 4 Mar 2022 22:17:56 +0000 (17:17 -0500)]
cmd/compile/internal/syntax: don't try to parse files in GOROOT/.git
This test was failing locally in my clone of the go repo due to a Git
branch ending in ".go", which the test found and was attempting to
parse as a file. It's fragile to try to parse .go files in
GOROOT/.git, and wasteful to scan GOROOT/pkg and other non-source
directories; instead, let's only parse the directories we actually
expect to contain source files.
Bryan C. Mills [Mon, 7 Mar 2022 16:26:18 +0000 (11:26 -0500)]
internal/fuzz: fix encoding for out-of-range ints and runes
Also switch float64 NaN encoding to use hexadecimal, and accept
hexadecimal encoding for all other integer types too. (That gives us
the flexibility to change the encodings in either direction in the
future without breaking earlier Go versions.)
Out-of-range runes encoded using "%q" were previously replaced with
the Unicode replacement charecter, losing their values.
Out-of-range ints and uints on 32-bit platforms were previously
rejected. Now they are wrapped instead: an “interesting” case with a
large int or uint found on a 64-bit platform likely remains
interesting on a 32-bit platform, even if the specific values differ.
To verify the above changes, I have made TestMarshalUnmarshal accept
(and check for) arbitrary differences between input and output, and
added tests cases that include values in valid but non-canonical
encodings.
I have also added round-trip fuzz tests in the opposite direction for
most of the types affected by this change, verifying that a marshaled
value unmarshals to the same bitwise value.
thepudds [Tue, 8 Mar 2022 14:51:27 +0000 (14:51 +0000)]
cmd/compile/internal/types2: more consistently print "check go.mod" if language version < 1.18
If you attempt to instantiate a generic type or func and run 'go build'
with a language version < 1.18 in the 'go' directive inside the go.mod
file, cmd/compile emits a friendly message that includes the suggestion
to 'check go.mod':
type instantiation requires go1.18 or later (-lang was set to go1.17; check go.mod)
However, if the code instead only declares a generic type or func
without instantiating, cmd/compile currently emits a less friendly
message:
type parameters require go1.18 or later
With this CL, the error in that situation becomes:
type parameter requires go1.18 or later (-lang was set to go1.17; check go.mod)
Within cmd/compile/internal/types2, it already calls check.versionErrorf
in a dozen or so places, including three existing calls to
check.versionErrorf within typeset.go (e.g., for embedding a constraint
interface).
This CL adds two more calls to check.versionErrorf, replacing calls to
check.softErrorf. Both check.versionErrorf and check.softErrorf call
check.err(at, <string>, true) after massaging the string message.
Fixes #51531
Change-Id: If54e179f5952b97701d1dfde4abb08101de07811
GitHub-Last-Rev: b0b7c1346f3a92f70e6cd5ff9ef047f441b09895
GitHub-Pull-Request: golang/go#51536
Reviewed-on: https://go-review.googlesource.com/c/go/+/390578 Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Robert Findley <rfindley@google.com>
Joel Sing [Fri, 21 Jan 2022 08:48:26 +0000 (19:48 +1100)]
internal/bytealg: optimise compare on riscv64
Implement compare using loops that process 32 bytes, 16 bytes, 4 bytes
or 1 byte depending on size and alignment. For comparisons that are less
than 32 bytes the overhead of checking and adjusting alignment usually
exceeds the overhead of reading and processing 4 bytes at a time.
Change-Id: Ie18712a9009d425c75e1ab49d5a673d84e73a1eb
Reviewed-on: https://go-review.googlesource.com/c/go/+/380076
Trust: Joel Sing <joel@sing.id.au>
Trust: mzh <mzh@golangcn.org> Reviewed-by: Cherry Mui <cherryyz@google.com>
Joel Sing [Fri, 21 Jan 2022 07:34:03 +0000 (18:34 +1100)]
internal/bytealg: optimise memequal on riscv64
Implement memequal using loops that process 32 bytes, 16 bytes, 4 bytes
or 1 byte depending on size and alignment. For comparisons that are less
than 32 bytes the overhead of checking and adjusting alignment usually
exceeds the overhead of reading and processing 4 bytes at a time.
Russ Cox [Tue, 22 Feb 2022 19:43:41 +0000 (14:43 -0500)]
cmd/compile: allow fieldtrack of unexported fields
The fieldtrack support is experimental and used mainly inside Google,
where we have included this change for years. No reason not to make
it in the public copy.
Russ Cox [Tue, 22 Feb 2022 22:07:49 +0000 (17:07 -0500)]
crypto/rand: simplify Prime to use only rejection sampling
The old code picks a random number n and then tests n, n+2, n+4, up to
n+(1<<20) for primality before giving up and picking a new n.
(The chance of finishing the loop and picking a new n is infinitesimally
small.) This approach, called “incremental search” in the Handbook of
Applied Cryptography, section 4.51, demands fewer bits from the random
source and amortizes some of the cost of the small-prime division
checks across the incremented values.
This commit deletes the n+2, n+4, ... checks, instead picking a series
of random n and stopping at the first one that is probably prime.
This approach is called “rejection sampling.”
Reasons to make this change, in decreasing order of importance:
1. Rejection sampling is simpler, and simpler is more clearly correct.
2. The main benefit of incremental search was performance, and that is
less important than before. Incremental search required fewer random
bits and was able to amortize the checks for small primes across the
entire sequence. However, both random bit generation and primality
checks have gotten faster much quicker than typical primes have
gotten longer, so the benefits are not as important today.
Also, random prime generation is not typically on the critical path.
Negating any lingering concerns about performance, rejection sampling
no slower in practice than the incremental search, perhaps because
the incremental search was using a somewhat inefficient test to
eliminate multiples of small primes; ProbablyPrime does it better.
name old time/op new time/op delta
Prime/MathRand 69.3ms ±23% 68.0ms ±37% ~ (p=0.531 n=20+19)
Prime/CryptoRand 69.2ms ±27% 63.8ms ±36% ~ (p=0.076 n=20+20)
(Here, Prime/MathRand is the current Prime benchmark,
and Prime/CryptoRand is an adaptation to use crypto/rand.Reader
instead of math/rand's non-cryptographic randomness source,
just in case the quality of the bits affects the outcome.
If anything, rejection sampling is even better with cryptographically
random bits, but really the two are statistically indistinguishable
over 20 runs.)
3. Incremental search has a clear bias when generating small primes:
a prime is more likely to be returned the larger the gap between
it and the next smaller prime. Although the bias is negligible in
practice for cryptographically large primes, people can measure the
bias for smaller prime sizes, and we have received such reports
extrapolating the bias to larger sizes and claiming a security bug
(which, to be clear, does not exist).
However, given that rejection sampling is simpler, more clearly
correct and at least no slower than incremental search, the bias
is indefensible.
4. Incremental search has a timing leak. If you can tell the incremental
search ran 10 times, then you know that p is such that there are no
primes in the range [p-20, p). To be clear, there are other timing
leaks in our current primality testing, so there's no definitive
benefit to eliminating this one, but there's also no reason to keep
it around.
(See https://bugs.chromium.org/p/boringssl/issues/detail?id=238 for
all the work that would be needed to make RSA key generation
constant-time, which is definitely not something we have planned for
Go crypto.)
5. Rejection sampling moves from matching OpenSSL to matching BoringSSL.
As a general rule BoringSSL is the better role model.
(Everyone started out using incremental search; BoringSSL switched
to rejection sampling in 2019, as part of the constant-time work
linked above.)
Robert Griesemer [Mon, 7 Mar 2022 22:48:28 +0000 (14:48 -0800)]
spec: document that type inference doesn't apply to generic types
Type inference for types was always a "nice to have" feature.
Given the under-appreciated complexity of making it work in all
cases, and the fact that we don't have a good understanding of
how it might affect readability of generic code, require explicit
type arguments for generic types.
This matches the current implementation.
Change-Id: Ie7ff6293d3fbea92ddc54c46285a4cabece7fe01
Reviewed-on: https://go-review.googlesource.com/c/go/+/390577
Trust: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org> Reviewed-by: Robert Findley <rfindley@google.com>
Dmitri Shuralyov [Tue, 8 Mar 2022 00:08:49 +0000 (00:08 +0000)]
fmt: use tabs for indentation
Replace 24 spaces added in CL 389434 with 3 tabs,
so the new line is indented like other lines around it.
Updates #51419.
Change-Id: Ic3e50023a01f233c52dda53c36de2c461222d95c
Reviewed-on: https://go-review.googlesource.com/c/go/+/390674
Trust: Dmitri Shuralyov <dmitshur@golang.org>
Run-TryBot: Dmitri Shuralyov <dmitshur@golang.org>
Auto-Submit: Dmitri Shuralyov <dmitshur@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Adam Shannon <adamkshannon@gmail.com> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Jorropo [Mon, 7 Mar 2022 12:56:01 +0000 (12:56 +0000)]
io: add WriterTo to MultiReader
This patch allows to zerocopy using MultiReader.
This is done by MultiReader implementing WriterTo.
Each sub reader is copied using usual io copy helper and thus use
WriterTo or ReadFrom with reflection.
There is a special case for when a subreader is a MultiReader.
Instead of using copyBuffer which would call multiReader.WriteTo,
multiReader.writeToWithBuffer is used instead, the difference
is that the temporary copy buffer is passed along, saving
allocations for nested MultiReaders.
The workflow looks like this:
- multiReader.WriteTo (allocates 32k buffer)
- multiReader.writeToWithBuffer
- for each subReader:
- is instance of multiReader ?
- yes, call multiReader.writeToWithBuffer
- no, call copyBuffer(writer, currentReader, buffer)
- does currentReader implements WriterTo ?
- yes, use use currentReader.WriteTo
- no, does writer implement ReadFrom ?
- yes, use writer.ReadFrom
- no, copy using Read / Write with buffer
This can be improved by lazy allocating the 32k buffer.
For example a MultiReader of such types:
MultiReader(
bytes.Reader, // WriterTo-able
bytes.Reader, // WriterTo-able
bytes.Reader, // WriterTo-able
)
Doesn't need any allocation, all copy can be done using bytes.Reader's
internal data slice. However currently we still allocate a 32k buffer
for nothing.
This optimisation has been omitted for a future patch because of high
complexity costs for a non obvious performance cost (it needs a benchmark).
This patch at least is on par with the previous MultiReader.Read
workflow allocation wise.
Fixes #50842
Change-Id: Ib070c8f36337d9dd86090df8a703c5df97a773ae
GitHub-Last-Rev: 8ebe60ceacec6bd52b63d9bdc05cd5b4ada57a6e
GitHub-Pull-Request: golang/go#51502
Reviewed-on: https://go-review.googlesource.com/c/go/+/390215
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org> Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Trust: Daniel Martí <mvdan@mvdan.cc>
Daniel Martí [Sun, 6 Mar 2022 20:31:33 +0000 (20:31 +0000)]
go/build: use static maps rather than an init func
go/build is one of the packages that contributes the most towards
cmd/go's init cost, which adds up to any call to the tool.
One piece of low-hanging fruit is knownOS and knownArch,
maps which are filled via an init func from a space-separated list.
Using GODEBUG=inittrace=1, we can get three samples:
The speedup isn't huge, but it helps, and also reduces allocs.
One can also imagine that the compiler may get better with static,
read-only maps in the future, whereas the init func will likely always
have a linear cost and extra allocations.
Change-Id: I430212bad03d25358d2cc7b1eab4536ad88d05a8
Reviewed-on: https://go-review.googlesource.com/c/go/+/390274
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
Trust: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Robert Findley [Fri, 4 Mar 2022 23:52:49 +0000 (18:52 -0500)]
go/types, types2: record all type instances, even duplicates
Due to instance de-duplication, we were failing to record some type
instances in types.Info.Instances. Fix this by moving the instance
recording out of the resolver.
Fixes #51494
Change-Id: Iddd8989307d95886eedb321efa4ab98cd2b3573a
Reviewed-on: https://go-review.googlesource.com/c/go/+/390041
Trust: Robert Findley <rfindley@google.com>
Run-TryBot: Robert Findley <rfindley@google.com> Reviewed-by: Robert Griesemer <gri@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Robert Griesemer [Fri, 4 Mar 2022 23:07:17 +0000 (15:07 -0800)]
go/types, types2: correctly include comparable in type set intersection
The comparable bit was handled incorrectly. This CL establishes
a clear invariant for a type set's terms and its comparable bit
and correctly uses the bit when computing term intersections.
Relevant changes:
- Introduce a new function intersectTermLists that does the
correct intersection computation.
Minor:
- Moved the comparable bit after terms in _TypeSet to make it
clearer that they belong together.
- Simplify and clarify _TypeSet.IsAll predicate.
- Remove the IsTypeSet predicate which was only used for error
reporting in union.go, and use the existing predicates instead.
- Rename/introduce local variables in computeInterfaceTypeSet
for consistency and to avoid confusion.
- Update some tests whose output has changed because the comparable
bit is now only set if we have have the set of all types.
For instance, for interface{comparable; int} the type set doesn't
set the comparable bit because the intersection of comparable and
int is just int; etc.
- Add many more comments to make the code clearer.
Fixes #51472.
Change-Id: I8a5661eb1693a41a17ce5f70d7e10774301f38ab
Reviewed-on: https://go-review.googlesource.com/c/go/+/390025
Trust: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org> Reviewed-by: Robert Findley <rfindley@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Matthew Dempsky [Mon, 7 Mar 2022 16:54:28 +0000 (08:54 -0800)]
cmd/compile: remove duplicate const logic from typecheck
Now that we always use types2 to validate user source code, we can
remove the constSet logic from typecheck for detecting duplicate
expression switch cases and duplicate map literal keys. This logic is
redundant with types2, and currently causes unified IR to report
inappropriate duplicate constant errors that only appear after type
substitution.
Updates #42758.
Change-Id: I51ee2c5106eec9abf40eba2480dc52603c68ba21
Reviewed-on: https://go-review.googlesource.com/c/go/+/390474
Trust: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Robert Griesemer <gri@golang.org>
Roland Shoemaker [Sat, 5 Mar 2022 16:47:33 +0000 (08:47 -0800)]
internal/fuzz: fix TestUnmarshalMarshal on MIPS
Previous value used in the float32 roundtrip used float32(math.NaN())-1
which caused the quiet/signal bit to flip, which seemed to break the
test on MIPS platforms. Instead switch to using float32(math.NaN())+1,
which preserves the bit and makes the test happy.
Possibly related to #37455
Fixes #51258
Change-Id: Ia85c649e89a5d02027c0ec197f0ff318aa819c19
Reviewed-on: https://go-review.googlesource.com/c/go/+/390214
Trust: Dmitri Shuralyov <dmitshur@golang.org> Reviewed-by: Bryan Mills <bcmills@google.com>
Trust: Roland Shoemaker <roland@golang.org>
Run-TryBot: Roland Shoemaker <roland@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Matthew Dempsky [Mon, 7 Mar 2022 07:47:27 +0000 (23:47 -0800)]
cmd/compile: represent derived types with ir.DynamicType in unified IR
This CL switches unified IR to using ir.DynamicType for derived
types. This has an immediate effect of fixing compilation of generic
code that when fully stenciled results in statically invalid type
assertions. This does require updating typecheck to expect
ODYNAMICTYPE in type switches, but this is straightforward to
implement.
For now, we still statically resolve the runtime type (or itab)
pointer. However, a subsequent CL will allow reading these pointers
from the runtime dictionary.
Change-Id: I1666678fcc588bc9cb8b97871bd02b9059848e6d
Reviewed-on: https://go-review.googlesource.com/c/go/+/390336
Trust: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Cuong Manh Le <cuong.manhle.vn@gmail.com>
Daniel Martí [Sun, 6 Mar 2022 11:26:11 +0000 (11:26 +0000)]
flag: make tests silent
A few of the tests were printing garbage to stderr,
since FlagSet's default Output is os.Stderr:
$ go test
flag provided but not defined: -x
invalid value "1" for flag -v: test error
Usage of test:
flag needs an argument: -b
Usage of test:
-b usage
PASS
ok flag 0.008s
Add the remaining SetOutput(io.Discard) method calls.
Note that TestUserDefinedFunc was a tricky one.
Even with the added SetOutput calls,
the last part of the test would still print usage text to stderr.
It took me a while to figure out the problem was copying FlagSet.
I've filed go.dev/issue/51507 to record this particular sharp edge,
and the test code now avoids making FlagSet copies to avoid the bug.
Change-Id: I323f24091b98386312aa72df3eb890af6625628d
Reviewed-on: https://go-review.googlesource.com/c/go/+/390234
Trust: Daniel Martí <mvdan@mvdan.cc>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Matthew Dempsky [Mon, 7 Mar 2022 03:52:08 +0000 (19:52 -0800)]
cmd/compile: remove unneeded type alias code in unified IR
Before #46477, the Go generics proposal allowed `type T = U` where `U`
was an uninstantiated generic type. However, we decided not to allow
that, and go/types and types2 have already been updated to disallow
it. This CL just removes the analogous code from unified IR.
Change-Id: I0fe6d1754c96790b498c1d5185b948333646d7de
Reviewed-on: https://go-review.googlesource.com/c/go/+/390315
Trust: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Cuong Manh Le <cuong.manhle.vn@gmail.com>
Matthew Dempsky [Mon, 7 Mar 2022 04:11:21 +0000 (20:11 -0800)]
cmd/compile: fix reentrancy issue in unified IR function body reading
We shouldn't need to read in function bodies for new functions found
during inlining, but something is expecting them to still be read
in. We should fix that code to not depend on them being read in, but
in the mean time reading them in anyway is at least correct, albeit
less efficient in time and space.
Fixes #49536.
Updates #50552.
Change-Id: I949ef45e7be09406e5a8149e251d78e015aca5fa
Reviewed-on: https://go-review.googlesource.com/c/go/+/390335
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Trust: Matthew Dempsky <mdempsky@google.com> Reviewed-by: Cuong Manh Le <cuong.manhle.vn@gmail.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Dan Kortschak [Sat, 5 Mar 2022 08:51:15 +0000 (19:21 +1030)]
all: fix some typos
Change-Id: I7dfae0fc91c2d70873ec7ec920be7c0a4888153a
Reviewed-on: https://go-review.googlesource.com/c/go/+/390175 Reviewed-by: Ian Lance Taylor <iant@golang.org> Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Trust: Daniel Martí <mvdan@mvdan.cc>
Adam Shannon [Wed, 2 Mar 2022 23:05:39 +0000 (15:05 -0800)]
fmt: clarify right-padded strings use spaces
Fixes #51419
Change-Id: I0a32f41a6e6e01481ad58c7dddb57ec7085d77af
Reviewed-on: https://go-review.googlesource.com/c/go/+/389434 Reviewed-by: Rob Pike <r@golang.org>
Trust: Ian Lance Taylor <iant@golang.org>
Iskander Sharipov [Wed, 17 Nov 2021 14:46:22 +0000 (17:46 +0300)]
encoding/xml: improve the test coverage, fix minor bugs
Improve the test coverage of encoding/xml package by adding
the test cases for the execution paths that were not covered before.
Since it reveals a couple of issues, fix them as well while we're at it.
As I used an `strings.EqualFold` instead of adding one more `strings.ToLower`,
our fix to `autoClose()` tends to run faster as well as a result.
name old time/op new time/op delta
HTMLAutoClose-8 5.93µs ± 2% 5.75µs ± 3% -3.16% (p=0.000 n=10+10)
name old alloc/op new alloc/op delta
HTMLAutoClose-8 2.60kB ± 0% 2.58kB ± 0% -0.46% (p=0.000 n=10+10)
name old allocs/op new allocs/op delta
HTMLAutoClose-8 72.0 ± 0% 67.0 ± 0% -6.94% (p=0.000 n=10+10)
The overall `encoding/xml` test coverage increase is `88.1% -> 89.9%`;
although it may look insignificant, this CL covers some important corner cases,
like `autoClose()` functionality (that was not tested at all).
Fixes #49635
Fixes #49636
Change-Id: I50b2769896c197eb285672313b7148f4fe8bdb38
Reviewed-on: https://go-review.googlesource.com/c/go/+/364734
Trust: Bryan Mills <bcmills@google.com> Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Trust: Daniel Martí <mvdan@mvdan.cc>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gopher Robot <gobot@golang.org>
Lynn Boger [Thu, 3 Mar 2022 13:28:44 +0000 (07:28 -0600)]
crypto/sha512: fix stack size for previous change
In a recent change CL 388654 a function was updated so it
no longer needed stack space, but the TEXT statement was
not updated to reflect that change. This corrects that problem.
Daniel Martí [Thu, 13 Jan 2022 12:46:34 +0000 (12:46 +0000)]
cmd/go: avoid rebuilding itself in TestMain
An extra "go build" was happening, for the sake of -tags=testgo,
which would insert some extra behavior into ./internal/work.
Instead, reuse the test binary as cmd/go directly,
by calling the main func when a special env var is set.
We still duplicate the test binary into testBin,
because we need a "go" executable in that directory for $PATH.
Finally, the special behavior is instead inserted via TestMain.
The numbers below represent how long it takes to run zero tests,
measured via:
benchcmd GoTestNothing go test -run=-
That is, the time it takes to run the first test is reduced by half.
Note that these numbers are on a warm build cache,
so if the -tags=testgo build were to be done from scratch,
the speed-up would be significantly more noticeable.
name old time/op new time/op delta
GoTestNothing 830ms ± 2% 380ms ± 7% -54.23% (p=0.008 n=5+5)
name old user-time/op new user-time/op delta
GoTestNothing 1.64s ± 1% 0.82s ± 3% -50.24% (p=0.008 n=5+5)
name old sys-time/op new sys-time/op delta
GoTestNothing 306ms ± 7% 159ms ±28% -48.15% (p=0.008 n=5+5)
name old peak-RSS-bytes new peak-RSS-bytes delta
GoTestNothing 173MB ± 1% 147MB ± 1% -14.96% (p=0.008 n=5+5)
Change-Id: I1f8fc71269a7b45bc5b82b7228e13f56589d44c3
Reviewed-on: https://go-review.googlesource.com/c/go/+/378294
Trust: Daniel Martí <mvdan@mvdan.cc>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Bryan Mills <bcmills@google.com>
Jason A. Donenfeld [Mon, 3 Jan 2022 17:48:51 +0000 (18:48 +0100)]
crypto/rand: use fast key erasure RNG on plan9 instead of ANSI X9.31
This should be a bit faster and slicker than the very old ANSI X9.31,
which relied on the system time. Uses AES instead of ChaCha because it's
in the standard library.
Jason A. Donenfeld [Thu, 9 Dec 2021 14:24:38 +0000 (15:24 +0100)]
crypto/rand: separate out plan9 X9.31 /dev/random expander
The X9.31 expander is now only used for plan9. Perhaps once upon a time
there was a use for abstraction, but the code is now covered in hacky
"fileName == urandomDevice" and "GOOS == plan9" checks, to the point
where the abstraction is much too leaky. Since plan9 is the only
platform that has a /dev/random without a /dev/urandom, we can simplify
both the generic urandom code and the plan9 X9.31 code by separating
them into different files, each focusing on doing one thing well.
Change-Id: I0ca43b748a0fbbd60f2ec7819688a540506d34df
Reviewed-on: https://go-review.googlesource.com/c/go/+/370580
Trust: Jason Donenfeld <Jason@zx2c4.com>
Run-TryBot: Jason Donenfeld <Jason@zx2c4.com>
TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Filippo Valsorda <filippo@golang.org>
Tobias Klauser [Thu, 3 Mar 2022 09:19:07 +0000 (10:19 +0100)]
runtime: remove fallback to pipe on platforms with pipe2
On Linux, the minimum required kernel version for Go 1.18 was be changed
to 2.6.32, see #45964. The pipe2 syscall was added in 2.6.27.
All other platforms already provide the pipe2 syscall in the minimum
supported version:
- DragonFly BSD added it in version 4.2, see
https://www.dragonflybsd.org/release42/
- FreeBSD added it in version 10.0, see
https://www.freebsd.org/cgi/man.cgi?pipe(2)#end
- NetBSD added it in version 6.0, see
https://man.netbsd.org/pipe2.2#HISTORY
- OpenBSD added it in version 5.7, see
https://man.openbsd.org/pipe.2#HISTORY
- Illumos supports it since 2013, see
https://www.illumos.org/issues/3714
- Solaris supports it since 11.4
This also allows to remove setNonblock which was only used in the pipe
fallback path on these platforms.
Cherry Mui [Thu, 3 Mar 2022 01:16:54 +0000 (20:16 -0500)]
runtime: count spill slot for frame size at finalizer call
The finalizer is called using reflectcall. When register ABI is
used, the finalizer's argument is passed in register(s). But the
frame size calculation does not include the spill slot. When the
argument actually spills, it may clobber the caller's stack frame.
This CL fixes it.
Change-Id: Ibcc7507c518ba65c1c5a7759e5cab0ae3fc7efce
Reviewed-on: https://go-review.googlesource.com/c/go/+/389574
Trust: Cherry Mui <cherryyz@google.com>
Run-TryBot: Cherry Mui <cherryyz@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Michael Knyszek <mknyszek@google.com>
Paul E. Murphy [Mon, 14 Feb 2022 23:15:05 +0000 (17:15 -0600)]
syscall, runtime/internal/syscall: always return 0 in r2 on ppc64{,le} linux syscalls
Both endians perform syscalls similarly. Only CR0S0 and R3 hold
the resultant status of a syscall. A random value may be stored into
the second return value (r2) result in some cases. Always set it to
zero.
Fixes #51192
Change-Id: Ida6a5692578d2cdadf3099af28478b3bc364f623
Reviewed-on: https://go-review.googlesource.com/c/go/+/385796
Run-TryBot: Paul Murphy <murp@ibm.com> Reviewed-by: Cherry Mui <cherryyz@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
Trust: Paul Murphy <murp@ibm.com>
Paul E. Murphy [Tue, 8 Feb 2022 15:09:36 +0000 (09:09 -0600)]
runtime: fix 32B backward copy on ppc64x
The test to enter the 32b copy loop always fails, and execution
falls back to a single 8B/iteration copy loop for copies of more
than 7 bytes. Likewise, the 32B loop has SRC/DST args mixed,
and fails to truncate DWORDS after completing.
Fix these, and unroll the 8B/iteration loop as it will only
execute 1-3 times if reached.
Eli Bendersky [Tue, 28 Sep 2021 13:18:30 +0000 (06:18 -0700)]
sort: use a different codegen strategy
The existing codegen strategy in sort.go relied on parsing the sort.go source
with go/ast and a combination of an AST rewrite + code text rewrite with regexes
to generate zfuncversion -- the same sort functionality with a different variant
of data.
In preparation for implementing #47619, we need a more robust codegen
strategy. To generate variants required for the generic sort functions
in the slices package, we'd need significanly more complicated AST
rewrites, which would make genzfunc.go much heavier.
Instead, redo the codegen strategy to use text/template instead of AST rewrites.
gen_sort_variants.go now contains the code for the underlying sort functions,
and generates multiple versions of them based on Variant configuration structs.
With this approach, adding new variants to generate generic sort functions for
the slices package becomes trivial.
See the discussion in #47619 for more details on the design decisions.
Change-Id: I8af784c41b1dc8ef92aaf6321359e8faa5fe106c
Reviewed-on: https://go-review.googlesource.com/c/go/+/353069 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Trust: Than McIntosh <thanm@google.com>
Lynn Boger [Thu, 26 Aug 2021 12:09:50 +0000 (07:09 -0500)]
crypto/aes: improve performance for aes-cbc on ppc64le
This adds an asm implementation of aes-cbc for ppc64le to
improve performance. This is ported from the
cryptogams implementation as are other functions in
crypto/aes with further description at the top of
the asm file.
Improvements on a power10:
name old time/op new time/op delta
AESCBCEncrypt1K 1.67µs ± 0% 0.87µs ±-48.15%
AESCBCDecrypt1K 1.35µs ± 0% 0.43µs ±-68.48%
name old speed new speed delta
AESCBCEncrypt1K 614MB/s ± 0% 1184MB/s ± 0%+92.84%
AESCBCDecrypt1K 757MB/s ± 0% 2403M/s ± 0 +217.21%
A fuzz test to compare the generic Go implemenation
against the asm implementation has been added.
eric fang [Wed, 19 Jan 2022 04:02:25 +0000 (04:02 +0000)]
runtime: use stp/ldp to save and restore all registers on arm64
Async preemption needs to save and restore almost all of the registers,
currently this is done by ldr and str on arm64. We can do it with ldp
and stp as they are more efficient.
Change-Id: Ida5a6f0a8d825a56af607ba2c2cd91fdc2e8f67f
Reviewed-on: https://go-review.googlesource.com/c/go/+/379715 Reviewed-by: Cherry Mui <cherryyz@google.com>
Trust: Eric Fang <eric.fang@arm.com>
Run-TryBot: Eric Fang <eric.fang@arm.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Ian Lance Taylor [Tue, 15 Feb 2022 21:46:56 +0000 (13:46 -0800)]
net: send EDNS(0) packet length in DNS query
Advertise to DNS resolvers that we are willing and able to accept up
to 1232 bytes in a DNS packet. The value 1232 was chosen based on
https://dnsflagday.net/2020/.
For #6464
For #21160
For #44135
For #51127
Fixes #51153
Change-Id: If9182d5210bfe047cf0a4d46163effc6812ab677
Reviewed-on: https://go-review.googlesource.com/c/go/+/386016
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org> Reviewed-by: Damien Neil <dneil@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Keith Randall [Wed, 2 Mar 2022 23:08:23 +0000 (15:08 -0800)]
cmd/compile: don't include instantiating types in type hash
This CL is a bit overkill, but it is pretty safe for 1.18. We'll
want to revisit for 1.19 so we can avoid the hash collisions between
types, e.g. G[int] and G[float64], that will cause some slowdowns
(but not incorrect behavior). Thanks Cherry for the simple idea.
Fixes #51250
Change-Id: I68130e09ba68e7cc35687bc623f63547bc552867
Reviewed-on: https://go-review.googlesource.com/c/go/+/389474
Trust: Keith Randall <khr@golang.org>
Run-TryBot: Keith Randall <khr@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com> Reviewed-by: Cherry Mui <cherryyz@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Carlos Amedee [Mon, 28 Feb 2022 23:04:20 +0000 (18:04 -0500)]
doc: start draft of go1.19 release notes, move go1.18 to x/website
This template is based on CL 342070 and previous ones like it.
Continue to eagerly include often-used sections, and clarify that
the TODO is about completing the section, or removing if it turns
out not to be needed.
Move the Go 1.18 release notes to x/website, since that's the new
home for past Go release notes as of CL 291711. They're added to
x/website in CL 388556.
For #51400
Updates #47694
Change-Id: I7b5213e039ad6e14a7ff7ad486311efcecc06824
Reviewed-on: https://go-review.googlesource.com/c/go/+/388515
Trust: Carlos Amedee <carlos@golang.org>
Run-TryBot: Carlos Amedee <carlos@golang.org> Reviewed-by: Dmitri Shuralyov <dmitshur@golang.org> Reviewed-by: Alex Rakoczy <alex@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Lynn Boger [Mon, 28 Feb 2022 21:36:11 +0000 (15:36 -0600)]
crypto/sha512: add BE support to PPC64 asm implementation
This adds big endian support for the assembly implementation of
sha512. There was a recent request to do this for sha256 for
AIX users; for completeness, the same is being done for sha512.
The majority of the code is common between big and little
endian with a few differences controlled by ifdefs: with LE
the generation of a mask is needed along with VPERM instructions
to put bytes in the correct order; some VPERMs need the V
registers in a different order.
name old time/op new time/op delta
Hash8Bytes 1.02µs ± 0% 0.38µs ± 0% -62.68%
Hash1K 7.01µs ± 0% 2.43µs ± 0% -65.42%
Hash8K 50.2µs ± 0% 14.6µs ± 0% -70.89%
Joe Tsai [Wed, 16 Feb 2022 00:59:59 +0000 (16:59 -0800)]
encoding/binary: add AppendByteOrder
AppendByteOrder specifies new methods for LittleEndian and BigEndian
for appending an unsigned integer to a byte slice.
The performance of AppendXXX methods are slower than PutXXX methods
since the former needs to do a few slice operations,
while the latter is essentially a single integer store.
In practice, existing usages of PutXXX performed slicing operations
around the call such that this cost was present, regardless.
Joe Tsai [Tue, 2 Nov 2021 19:01:24 +0000 (12:01 -0700)]
encoding/json: use reflect.Value.UnsafePointer over Pointer
The latter returns a uintptr, while the former returns a unsafe.Pointer.
A uintptr is unsafe if Go ever switches to a moving GC,
while a unsafe.Pointer will be properly tracked by the GC.
We do not use unsafe.Pointer for any unsafe type conversions,
and only use it for comparability purposes, which is relatively safe.
Updates #40592
Change-Id: I813e218668704b63a3043acda4331205a3835a66
Reviewed-on: https://go-review.googlesource.com/c/go/+/360855
Trust: Joseph Tsai <joetsai@digital-static.net>
Trust: Daniel Martí <mvdan@mvdan.cc> Reviewed-by: Daniel Martí <mvdan@mvdan.cc> Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Joseph Tsai <joetsai@digital-static.net>
TryBot-Result: Gopher Robot <gobot@golang.org>
Phil Bracikowski [Wed, 2 Mar 2022 01:51:28 +0000 (01:51 +0000)]
compress/gzip: return unexpected EOF for certain truncated streams
For cases where RFC 1952 requires a field, the code returns the error
io.ErrUnexpectedEOF except in two places: for the FNAME flag or the
FCOMMENT flag. These flags expect a null-terminated string and
readString may return an EOF if the Reader is truncated before a
null byte is found. For consistency with parsing other parts of the
header, this is converted to an unexpected EOF herein.
Follow-up to CL 14832.
Fixes #51417
Change-Id: I173283a6ae309e4a8e52fc15df404ce5db06eff1
GitHub-Last-Rev: 2e573cd961c0b1e4296fbde53097cf5723a2ca29
GitHub-Pull-Request: golang/go#51418
Reviewed-on: https://go-review.googlesource.com/c/go/+/389034 Reviewed-by: Joseph Tsai <joetsai@digital-static.net>
Trust: Joseph Tsai <joetsai@digital-static.net>
Trust: Ian Lance Taylor <iant@golang.org>
MoZhonghua [Fri, 17 Dec 2021 06:53:21 +0000 (06:53 +0000)]
cmd/link: don't generate typedef DWARF entry for noalg.struct{...}
cmd/compile uses "noalg.struct {...}" as type name when hash and eq algorithm generation of this struct type is suppressed. This should be treated as normal struct type, that is, link shouldn't generate DW_TAG_typedef DIE for it.
Change-Id: Ifada8a818bcfa2e5615f85ead9582cead923b86c
GitHub-Last-Rev: 15de3e4a846fcd79e231384539be20b06197b826
GitHub-Pull-Request: golang/go#50237
Reviewed-on: https://go-review.googlesource.com/c/go/+/373054 Reviewed-by: Than McIntosh <thanm@google.com> Reviewed-by: Cherry Mui <cherryyz@google.com>
Run-TryBot: Cherry Mui <cherryyz@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Jinwook Jeong [Mon, 1 Nov 2021 14:46:47 +0000 (23:46 +0900)]
regexp: fix typo in the overview
Correct the slice expression in the description of Index functions.
Change-Id: I97a1b670c4c7e600d858f6550b647f677ef90b41
Reviewed-on: https://go-review.googlesource.com/c/go/+/360058 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Trust: Daniel Martí <mvdan@mvdan.cc>
Alan Donovan [Wed, 5 Jan 2022 14:20:15 +0000 (09:20 -0500)]
unicode/utf8: optimize Valid to parity with ValidString
The benchmarks added in this change revealed that ValidString
runs ~17% faster than Valid([]byte) on the ASCII prefix
of the input. Inspection of the assembly revealed that the
code generated for p[8:] required recomputing the slice capacity
to handle the cap=0 special case, which added an ADD -8 instruction.
By making len=cap, the capacity becomes a common subexpression
with the length, saving the ADD instruction.
(Thanks to khr for the tip.)
Incidentally, I tried a number of other optimizations but was
unable to make consistent gains across all benchmarks. The most
promising was to retain the bitmask of non-ASCII bytes from the
fast loop; the slow loop would shift it, and when it becomes zero,
return to the fast loop. This made the MostlyASCII benchmark 4x
faster, but made the other cases slower by up to 10%.
Robert Griesemer [Fri, 25 Feb 2022 06:11:40 +0000 (22:11 -0800)]
go/types, types2: correctly consider ~ (tilde) in constraint type inference
When doing constraint type inference, we must consider whether the
constraint's core type is precise (no tilde) or imprecise (tilde,
or not a single specific type). In the latter case, we cannot infer
an unknown type argument from the (imprecise) core type because there
are infinitely many possible types. For instance, given
[E ~byte]
if we don't know E, we cannot infer that E must be byte (it could be
myByte, etc.). On the other hand, if we do know the type argument,
say for S in this example:
[S ~[]E, E any]
we must consider the underlying type of S when matching against ~[]E
because we have a tilde.
Because constraint type inference may infer type arguments that were
not eligible initially (because they were unknown and the core type
is imprecise), we must iterate the process until nothing changes any-
more. For instance, given
[S ~[]E, M ~map[string]S, E any]
where we initially only know the type argument for M, we must ignore
S (and E) at first. After one iteration of constraint type inference,
S is known at which point we can infer E as well.
The change is large-ish but the actual functional changes are small:
- There's a new method "unknowns" to determine the number of as of yet
unknown type arguments.
- The adjCoreType function has been adjusted to also return tilde
and single-type information. This is now conveniently returned
as (*term, bool), and the function has been renamed to coreTerm.
- The original constraint type inference loop has been adjusted to
consider tilde information.
- This adjusted original constraint type inference loop has been
nested in another loop for iteration, together with some minimal
logic to control termination.
The remaining changes are modifications to tests:
- There's a substantial new test for this issue.
- Several existing test cases were adjusted to accomodate the
fact that they inferred incorrect types: tildes have been
removed throughout. Most of these tests are for pathological
cases.
- A couple of tests were adjusted where there was a difference
between the go/types and types2 version.
Fixes #51229.
Change-Id: If0bf5fb70ec22913b5a2da89adbf8a27fbc921d9
Reviewed-on: https://go-review.googlesource.com/c/go/+/387977
Trust: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org> Reviewed-by: Robert Findley <rfindley@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Keith Randall [Sun, 6 Feb 2022 17:43:39 +0000 (09:43 -0800)]
cmd/compile: include all entries in map literal hint size
Currently we only include static entries in the hint for sizing
the map when allocating a map for a map literal. Change that to
include all entries.
This will be an overallocation if the dynamic entries in the map have
equal keys, but equal keys in map literals are rare, and at worst we
waste a bit of space.
Cherry Mui [Tue, 1 Mar 2022 17:02:38 +0000 (12:02 -0500)]
cmd/compile: use AutogeneratedPos for method value wrapper
We use AutogeneratedPos for most compiler-generated functions. But
for method value wrappers we currently don't. Instead, we use the
Pos for their (direct) declaration if there is one, otherwise
not set it in methodValueWrapper, which will probably cause it to
inherit from the caller, i.e. the Pos of that method value
expression. If that Pos has inline information, it will cause the
method wrapper to have bogus inline information, which could lead
to infinite loop when printing a stack trace.
Change it to use AutogeneratedPos instead.
Fixes #51401.
Change-Id: I398dfe85f9f875e1fd82dc2f489dab63ada6570d
Reviewed-on: https://go-review.googlesource.com/c/go/+/388794
Trust: Cherry Mui <cherryyz@google.com>
Run-TryBot: Cherry Mui <cherryyz@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Matthew Dempsky [Tue, 1 Mar 2022 20:21:04 +0000 (12:21 -0800)]
test: workaround codegen bug in typeparam/mdempsky/13.go
This test case is failing on the noopt builder, because it disables
inlining. Evidently the explicit -gcflags flag in all of our generics
tests was overriding the noopt builder's default mode.
This CL restores a noop -gcflags to get the builder green again until
the issue can be properly fixed.
Updates #51413.
Change-Id: I61d22a007105f756104ba690b73f1d68ce4be281
Reviewed-on: https://go-review.googlesource.com/c/go/+/388894
Trust: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Matthew Dempsky <mdempsky@google.com> Reviewed-by: Keith Randall <khr@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
ir.PkgName was only used by the old -G=0 frontend for representing
identifiers that refer to a package name. The new types2-based
frontends directly resolve the qualified identifier to the respective
object during IR construction.
Similarly, most of the ir.*Type nodes were only needed for
representing types in the IR prior to type checking. The new
types2-based frontends directly construct the corresponding types.Type
instead.
Exception: The internal typecheck.DeclFunc API used for
compiler-generated functions still depends on ir.FuncType, so that IR
node type is retained for now. (Eventually, we should update
typecheck.DeclFunc and callers to not depend on it, but it's not
urgent.)
Change-Id: I982f1bbd41eef5b42ce0f32676c7dc4a8ab6d0ee
Reviewed-on: https://go-review.googlesource.com/c/go/+/388538
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com> Reviewed-by: Keith Randall <khr@golang.org>