go.dev/cl/339591 changed the code generation to use a constant string,
so that the ~88KiB table can be marked read-only.
The compiled code became a lot better, but unfortunately,
the generated Go source became significantly more inefficient.
The numbers below compare "gofmt -l" and "go tool compile" of said file,
where "old" is the file as of Go 1.17, and "new" as of master in 2022/01/19:
name old time/op new time/op delta
Gofmt 22.8ms ± 6% 898.5ms ± 3% +3837.32% (p=0.000 n=8+8)
GoToolCompile 26.9ms ± 2% 371.1ms ± 2% +1278.36% (p=0.000 n=7+8)
name old user-time/op new user-time/op delta
Gofmt 25.7ms ±65% 897.1ms ± 3% +3383.86% (p=0.000 n=8+8)
GoToolCompile 35.1ms ±26% 367.2ms ± 3% +945.80% (p=0.000 n=8+8)
name old sys-time/op new sys-time/op delta
Gofmt 6.42ms ±276% 7.23ms ±38% ~ (p=0.412 n=8+6)
GoToolCompile 9.20ms ±100% 13.90ms ±53% ~ (p=0.105 n=8+8)
name old peak-RSS-bytes new peak-RSS-bytes delta
Gofmt 9.11MB ± 7% 22.79MB ± 1% +150.23% (p=0.000 n=8+8)
GoToolCompile 25.1MB ± 2% 68.6MB ± 2% +173.57% (p=0.000 n=8+8)
"+" operators are binary expressions at the syntax tree level,
which are represented by packages like go/ast as roughly:
struct {
X Expr
Op Token
Y Expr
}
Since each node is a pointer, chains of "+" operators act like linked lists.
The generated code has about 14k lines, and 8 "+" operators per line,
meaning that we end up with a linked list with over 11k elements.
This explains the slow-down in gofmt; the printer must walk said list,
and it does so more than once to work out how to format it.
It seems like the compiler is similarly affected by the huge length.
To remedy the effect of the linked list, use go:embed instead.
This results in the same string variable with the binary table,
but it greatly reduces the amount of syntax and its cost above.
We still keep the generator around, but modified so it produces the
binary file to be embedded rather than a large Go file.
Finally, we update go/build/deps_test.go to allow crypto/elliptic to
depend on embed; it's a tiny package and crypto/elliptic was already
manually embedding assets via code generation.
The change to deps_test.go was briefly discussed in the issue below.
Fixes#50995.
Change-Id: I0c8b432710e971a296a2e9c99147cc2cad9662aa
Reviewed-on: https://go-review.googlesource.com/c/go/+/380475
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
Trust: Daniel Martí <mvdan@mvdan.cc>
Trust: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Filippo Valsorda <filippo@golang.org>
Declare receiver type parameters in the function scope, but don't
resolve them (for now), as ast.Object.Decl is not documented to hold
*ast.Idents. This avoids incorrect resolution of identifiers to names
outside the function scope.
Also make tracing and error reporting more consistent.
For golang/go#50956
Change-Id: I8cd61dd25f4c0f6b974221599b00e23d8da206a1
Reviewed-on: https://go-review.googlesource.com/c/go/+/382247
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>
Due to delayed computation of constraint type sets it is possible that
a type argument is checked against a constraint before that constraint
has been wrapped into an implicit interface (if needed).
When the type checker is about to check whether a type argument
implements a constraint, it's always safe to force wrapping of
a constraint in an implicit interface (if necessary) before doing
the implements check.
Also, use a better position for tracing output related to constraint
type set computation.
Fixes#51048.
Change-Id: I52fecbf57814f09b62762452d7e17c2a230fdd59
Reviewed-on: https://go-review.googlesource.com/c/go/+/383834
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
Record the actual argument type for a cap/len call, not the
underlying type.
Fixes#51055.
Change-Id: Ia0e746a462377f030424ccaec0babf72b78da420
Reviewed-on: https://go-review.googlesource.com/c/go/+/383474
Trust: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
For the gofrontend, it's not; there are some trailing one bits,
which is permitted by the spec.
Change-Id: I58f5a5347068bb488306ec8e73da7c59376f1ae1
Reviewed-on: https://go-review.googlesource.com/c/go/+/383635
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Change-Id: I119cfb1a0da9af89ced78935b8fcdfdb8d9b4ab6
Reviewed-on: https://go-review.googlesource.com/c/go/+/383794
Trust: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
These tests use a slice to represent the base C compiler command (with
flags). Appending to that slice can cause subtle aliasing bugs, such
as commands that silently corrupt the arguments of other concurrent
commands in parallel tests.
In this change, we explicitly reduce the capacity of the command slice
to force appends to it to always allocate unique new slices.
Fixes#49693
Change-Id: Ide466bf65f12cb6cead3dcba69f513cccb60a160
Reviewed-on: https://go-review.googlesource.com/c/go/+/383754
Trust: Bryan Mills <bcmills@google.com>
Run-TryBot: Bryan Mills <bcmills@google.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Change run.go to apply the GO_TEST_TIMEOUT_SCALE scaling factor to
test timeouts (mentioned in "-t" clause in test header).
Also with this patch, bump up the timeout for fixedbugs/issue46234.go
from 30 to 45 seconds, to avoid flakes on very slow builders.
Updates #50973.
Change-Id: Icbafa482860e24cc1e72fee53511bcc764d06bf1
Reviewed-on: https://go-review.googlesource.com/c/go/+/382774
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Trust: Than McIntosh <thanm@google.com>
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
The dir argument is already in prefix, we shouldn't add it again.
Change-Id: I42a158bec3a43950fce24f57b808da3ad8c5ef5b
Reviewed-on: https://go-review.googlesource.com/c/go/+/383636
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
The previous workaround for issue #39466 only disabled this test for
Linux. However, the issue manifests for all arm64 systems with gcc 9.4
and above. The new netbsd-arm64 builder uses NetBSD-current with gcc
10.3, so it fails in the same way.
Updates #39466
Change-Id: I276a99a5e60914e5c22f74a680e461bea17cfe92
Reviewed-on: https://go-review.googlesource.com/c/go/+/383554
Trust: Benny Siegert <bsiegert@gmail.com>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
The code was updated, the comments were not.
Change-Id: If387779f3abd5e8a1b487fe34c33dcf9ce5fa7ff
Reviewed-on: https://go-review.googlesource.com/c/go/+/383495
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Further simplify and regularize Checker.missingMethod and use the
same code in both type checkers.
This enables case-folding lookup for go/types.
Adjusted test case that looks for alternative methods.
Change-Id: I5b8cc598c295c329ff93b1c65787cc6140f0900e
Reviewed-on: https://go-review.googlesource.com/c/go/+/382858
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
Remove special case where we don't have a *Checker and always use
Checker.missingMethodReason in Checker.implements.
Look for zero methods rather than empty interface to exit early
from Checker.missingMethod, and remove the extra test in
Checker.implements.
With this change we get consistent and more detailed error messages
from all places where we do a form of the "implements" test.
To make this possible, allow for the receiver to be nil in
- Checker.sprintf
- Checker.missingMethodReason
- Checker.interfacePtrError
- Checker.funcString
Allowing Checker.sprintf with nil Checker permits further simplifying
in a couple of places.
Change-Id: I0ea7178c9efbcd4a25ded2a66e2b058db52dc4d4
Reviewed-on: https://go-review.googlesource.com/c/go/+/383054
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
- Remove the xtyp argument from the Checker.typeAssertion parameter
list; it was confusing and not needed. Adjusted call sites.
- Simplify logic in Checker.typeAssertion.
- Use the same code in both types2 and go/types, specifically
use the same error positions.
- Adjust error messages as needed.
This removes another subtle discrepancy between types2 and go/types.
The go/types error messages don't have the have/want appendix for
the affected error messages yet because we don't use case folding
in lookups yet.
Change-Id: Id39f5c473da36c9baad60082f85cf1f34dc26c50
Reviewed-on: https://go-review.googlesource.com/c/go/+/383014
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
Added a funcString helper so we don't need to rewrite strings
with strings.Replace.
Use compiler format for error message about wrong method type;
this removes another unnecessary variation.
Simplify conditions for pointer-to-interface related error:
if one of the involved types is an interface pointer, it can't
have any methods.
Rewrite logic so we don't need all the else-if branches.
Adjusted a test case for types2 accordingly. The go/types version
of this test case has a different error because the implementation
of Checker.typeAssertion is different in the two type checkers
(the types2 version gives errors closer to the 1.17 compiler).
Change-Id: I19e604d7063c3e31e8290bd78384a9f38bb0d740
Reviewed-on: https://go-review.googlesource.com/c/go/+/382694
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
Because Checker.missingMethod also looks up methods matching
matching case-folded names, when Checker.missingMethod returns
an alternative method, that method does not automatically have
the wrong type. It may be a method with a different name.
Adjust types2.MissingMethod to check the alternative method
name before reporting a wrong type.
Add API test that verifies (now correct) behavior for this
case.
Ported the code also to go/types, though it was not a bug
there yet because looking up with case-folding is not yet
enabled.
Change-Id: Iaa48808535c9265a9879338ea666c6c021e93a2b
Reviewed-on: https://go-review.googlesource.com/c/go/+/382634
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
There is no need to check for length equality of type parameter
lists in Checker.missingMethod: the Identical predicate does
this check. Furthermore, we don't have methods with their own
type parameters. Remove the unnecessary (duplicate) code.
Also, update doc string on missingMethod and rename the 2nd
result parameter for clarity, and clarify internal comments.
For go/types, include the same case-folding code as for types2
but leave it disabled or now. Adjust any other differences in
the missingMethod implementation.
With this change, the types2 and go/types code of missingMethod
is identical again except for the disabled case-folding lookup.
No functionality changes.
Preparation for fixing some of the outstanding error reporting issues.
Change-Id: I4778d006c17f4e084ecc2cac7386c68e86aa49eb
Reviewed-on: https://go-review.googlesource.com/c/go/+/382614
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
types2 has some code to handle case-folding when doing method lookups
which was missing in go/types. This change is a first step to match
the implementations. Specifically:
In types2:
- remove the lookupMethodFold names in favor of just lookupMethod,
but with the foldCase flag (e.g., instead if lookupMethodFold,
we just use lookupMethod)
- rename checkFold to foldCase everywhere
- add foldCase parameter where it was missing
- moved foldCase paremeter to the end in lookupFieldOrMethod
- no functionality changes
In go/types:
- match function signatures with types2 use
- always provide false as argument for foldCase for now
- no functionality changes
Preparation for fixing some of the outstanding error reporting issues.
Change-Id: If129a5feb89ddf96a3596e8d73b23afa591875a0
Reviewed-on: https://go-review.googlesource.com/c/go/+/382461
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
Refactor Checker.comparison such that its logic is easier to reason
about and so that special cases can be handled more directly.
Use the appropriate operand (of 1st or 2nd operand) for error
reporting (position and type), rather than always using the
first operand.
Use an extra parameter to indicate a switch case
comparison; in this case the error is always reported at
the position of the first operand. (The error messages are
not yet adjusted for switches; see next CL.)
Introduce a new kindString function which is used to print simplified
types in error messages (related to comparisons only): instead of
printing the details of a struct type, we just print "struct" where
the details are not relevant. This matches the 1.17 compiler behavior.
Added a "reportf" parameter to the internal comparable function so we
can report an error cause in addition to the boolean result. Rather
than passing a *string for cause, we pass a function to record the
cause so that we can use the *Checker context for printing (needed
for proper type qualification). This mechanism reports the same
details now as the 1.17 compiler.
Adjusted various tests as needed added new test files.
Fixes#50918.
Change-Id: I1f0e7af22f09db4d31679c667c71a9038a8dc9d2
Reviewed-on: https://go-review.googlesource.com/c/go/+/381964
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
In bespoke build environments default libraries may be specificied with
LIBRARY_PATH, C_INCLUDE_PATH enviroment variables to overide the system
(i.e glibc). Allow them though to allow cgo testing of these enviroments.
Fixes#50985
Change-Id: I7497a7715d9b635a6ae97efaab94a7ff01cdf8e2
Reviewed-on: https://go-review.googlesource.com/c/go/+/383334
Reviewed-by: Bryan Mills <bcmills@google.com>
Run-TryBot: Bryan Mills <bcmills@google.com>
Trust: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Change-Id: Ibf51c0687197c0d791916b21cba7f8408aa5300a
Reviewed-on: https://go-review.googlesource.com/c/go/+/383216
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Trust: Kevin Burke <kevin@burke.dev>
The go/parser package is updated to report an error on method type
parameters, and to not store them in the AST. Tests are updated
accordingly, and error messages are normalized accross go/parser and the
compiler syntax package.
Before this CL, go/parser would parse type parameters on method
declarations and interface methods, leaving it to go/types to complain.
There are several problems with this:
- Interface Methods and Method declarations do not have type parameters
in the spec. We try to align the parser with the productions in the
spec.
- Parsing method type parameters means that downstream libraries
(go/doc, go/format, etc.) theoretically need to handle them, even
though they are not part of the language.
- Relatedly, go/types has inconsistent handling of method type
parameters due to support being removed, leading to the crasher in
#50427.
It is more consistent and safer to disallow type parameters on methods
in the parser.
Fixes#50427
Change-Id: I555766da0c76c4cf1cfe0baa9416863088088b4e
Reviewed-on: https://go-review.googlesource.com/c/go/+/382538
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>
GCC 9 warns about a change in the ABI of passing structs with bitfields,
but we don't care.
Fixes#50987
Change-Id: Ica658d04172a42a7be788f94d31a714bb8c4766f
Reviewed-on: https://go-review.googlesource.com/c/go/+/382956
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Benny Siegert <bsiegert@gmail.com>
Trust: Benny Siegert <bsiegert@gmail.com>
Add a regression test for issue 50936 which coerces the runtime into
frequent execution of the cgocall dropg/execute curg assignment race by
making many concurrent cgo calls eligible for P retake by sysmon. This
results in no P during exitsyscall, at which point they will update curg
and will crash if SIGPROF arrives in between updating mp.curg and
mp.curg.m.
This test is conceptually similar to the basic cgo callback test in
aprof.go but with additional concurrency and a sleep in C.
On my machine this test fails ~5% of the time prior to CL 382079.
For #50936.
Change-Id: I21b6c7f2594f9a615a64580ef70a88b692505678
Reviewed-on: https://go-review.googlesource.com/c/go/+/382244
Trust: Michael Pratt <mpratt@google.com>
Run-TryBot: Michael Pratt <mpratt@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
This failure mode has been present since at least 2020-06-08. We have
enough information to diagnose it, and further failures don't seem to
be adding any new information at this point: they can only add noise,
both on the Go project's builders and in users' own modules (for
example, when run as part of 'go test all').
For #39457
Change-Id: I2379631da0c8af69598fa61c0cc5ac0ea6ba8267
Reviewed-on: https://go-review.googlesource.com/c/go/+/382395
Trust: Bryan Mills <bcmills@google.com>
Run-TryBot: Bryan Mills <bcmills@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Since at least CL 121857, the conversion logic in
(*modfetch).codeRepo.Stat has had a short-circuit to use the version
requested by the caller if it successfully resolves and is already
canonical.
However, we should not use that version if it refers to a branch
instead of a tag, because branches (unlike tags) usually do not refer
to a single, stable release: a branch named "v1.0.0" may be for the
development of the v1.0.0 release, or for the development of patches
based on v1.0.0, but only one commit (perhaps at the end of that
branch — but possibly not even written yet!) can be that specific
version.
We already have some logic to prefer tags that are semver-equivalent
to the version requested by the caller. That more general case
suffices for exact equality too — so we can eliminate the
special-case, fixing the bug and (happily!) also somewhat simplifying
the code.
Fixes#35671
Updates #41512
Change-Id: I2fd290190b8a99a580deec7e26d15659b58a50b0
Reviewed-on: https://go-review.googlesource.com/c/go/+/378400
Trust: Bryan Mills <bcmills@google.com>
Run-TryBot: Bryan Mills <bcmills@google.com>
Reviewed-by: Russ Cox <rsc@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
In the 2nd phase of function argument type inference we only
consider parameters with types that are single type parameters.
Thus there is no need to collect anything else in the first
phase.
This matches the algorithm description in the forthcoming spec
more closely.
Change-Id: Ie5c29f30ff43b1e37d719ecbe1688b50ed2177f3
Reviewed-on: https://go-review.googlesource.com/c/go/+/381554
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Findley <rfindley@google.com>
Looking at the condition actually exercised by the test it seems
unnecessary: assuming that the Write succeeds (checked earlier in the
test), the Read must have a nonzero number of bytes available to read
immediately. (That is not the case in TestUDPZeroByteBuffer, from
which this test appears to have been derived.)
Fixes#50870
Change-Id: Ia6040a2d5dc320f0b86ec9d6f6b91dc72e8f3b84
Reviewed-on: https://go-review.googlesource.com/c/go/+/382537
Trust: Bryan Mills <bcmills@google.com>
Run-TryBot: Bryan Mills <bcmills@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Otherwise, the failure mode for these subcommands refers to an empty file path:
go: open : no such file or directory
Fixes#50964
Change-Id: I8776431a294d2b2246d7d147b6059054f31bc255
Reviewed-on: https://go-review.googlesource.com/c/go/+/382246
Trust: Bryan Mills <bcmills@google.com>
Run-TryBot: Bryan Mills <bcmills@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Michael Matloob <matloob@golang.org>
It has moved to golang.org/x/exp/constraints. Perhaps it will move
back to the standard library in a future release.
For golang/go#45458Fixesgolang/go#50792
Change-Id: I93aa251a7afe7b329a3d3faadc0c5d6388b1f0e9
Reviewed-on: https://go-review.googlesource.com/c/go/+/382460
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
crypto/ecdsa was long overdue a cleanup. Bump the FIPS 186 version, and
make sure we consistently reference that and SEC 1, not the paywalled
ANSI standard.
Change-Id: Idd90bd6c14b334941fdcd829d89b796a60a8b174
Reviewed-on: https://go-review.googlesource.com/c/go/+/352529
Run-TryBot: Filippo Valsorda <filippo@golang.org>
Trust: Filippo Valsorda <filippo@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Roland Shoemaker <roland@golang.org>
CL 339669 changed freedefer to not mention every field of _defer,
so no need to call it out in the _defer comment.
Change-Id: Id8b67ba2298685f609bf901b5948fd30666bd6e3
Reviewed-on: https://go-review.googlesource.com/c/go/+/382251
Trust: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Add a panicAtUnificationDepthLimit const to replace the use of the debug
const to control whether to panic when the unification recursion depth
is reached. Our tests should pass when debug==true.
Change-Id: I58847f49d66010bd4ca01c0408ec10acac95cba6
Reviewed-on: https://go-review.googlesource.com/c/go/+/382534
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>
