This CL introduces `VarArgs` to `NativeFunction` signatures. The
`VarArgs` type takes a single type argument. This type argument is a
subtype of `NativeType` if there is a single variadic argument, and a
record with native types if there are multiple variadic arguments.
For example:
`NativeFunction<Void Function(Pointer<Char>, VarArgs<(Int32,Int32)>)>`
for calling refering to a `printf` binding with two `int32_t` arguments
passed as variadic arguments.
The logic of the native calling conventions are detailed in
https://dart-review.googlesource.com/c/sdk/+/278342.
Here we explain how this influences the FFI pipeline.
First, now that `VarArgs` is part of signatures, we have to unwrap
that when with the C types in the CFE transform and checking (analyzer
is in a separate CL), and also in the marshaller when looking up the
C type of arguments.
Second, we have to deal with `BothNativeLocations`. On windows x64,
floating point arguments must be passed both in FPU _and_ CPU
registers. For FFI calls, we solve this in the argument moves by just
copying to both locations. For FFI callbacks, we just take the FPU
register location (which avoids an extra bitcast).
Third, on System-V, we have to pass an upper bound of the number of
XMM registers used in AL. This means we instead RAX, we use R13 for the
target address. For variadic calls, we always pass 8 in AL as the valid
upper bound. We could consider passing the actual number of XMM
registers used.
We keep using RAX as default register for the function address on non-
variadic calls, because changing to R13 (the first free) register
creates more spilling in leaf calls. R13 is callee-saved while RAX is
not, so using R13 instead of RAX causes us to have to spill the value
from RAX on leaf calls.
Fourth, on both x64 and RISC-V, we pass floats in integer locations.
`EmitNativeMove` has been modified to deal with this, so that we do not
have to insert more `BitCastInstr`s.
The tests are generated by a test generator: `tests/ffi/generator/`.
The formatter doesn't support records yet, so the tests are not properly
formatted.
Bug: https://github.com/dart-lang/sdk/issues/50798
TEST=tests/ffi/*_varargs_*
Closes: https://github.com/dart-lang/sdk/issues/38578
Closes: https://github.com/dart-lang/sdk/issues/49460
Closes: https://github.com/dart-lang/sdk/issues/50858
Change-Id: I6a6296fe972527f8a54ac75a630131769e3cc540
Cq-Include-Trybots: luci.dart.try:vm-kernel-reload-rollback-linux-debug-x64-try,vm-kernel-reload-linux-debug-x64-try,vm-kernel-linux-debug-ia32-try,vm-kernel-nnbd-linux-debug-ia32-try,vm-kernel-win-debug-ia32-try,vm-kernel-linux-debug-x64-try,vm-kernel-mac-debug-x64-try,vm-kernel-win-debug-x64-try,vm-kernel-nnbd-win-release-ia32-try,vm-kernel-nnbd-win-debug-x64-try,vm-ffi-android-debug-arm-try,vm-ffi-android-debug-arm64c-try,vm-kernel-precomp-android-release-arm64c-try,vm-kernel-precomp-android-release-arm_x64-try,vm-precomp-ffi-qemu-linux-release-arm-try,vm-precomp-ffi-qemu-linux-release-riscv64-try,vm-kernel-asan-linux-release-x64-try,vm-kernel-precomp-asan-linux-release-x64-try,vm-kernel-msan-linux-release-x64-try,vm-kernel-precomp-msan-linux-release-x64-try,app-kernel-linux-debug-x64-try,vm-kernel-mac-release-arm64-try,vm-kernel-nnbd-mac-debug-arm64-try,vm-kernel-nnbd-mac-debug-x64-try
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/276921
Reviewed-by: Devon Carew <devoncarew@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
On ARM64 macos and ios, when varargs are used, the first vararg blocks
all cpu and fpu registers.
On Windows x64, when varargs are used, floating point arguments are
passed _both_ in the integer and double register.
The Windows logic requires a new kind of native location:
`BothNativeLocations`, which signals that a value needs to be copied
to both locations before an FFI call, and can be copied from any of
the two locations when getting an FFI callback.
TEST=runtime/vm/compiler/ffi/unit_tests/variadic_double/x64_win.expect
Note that integer arguments already block out the corresponding xmm
registers on Windows x64.
On System-V, an upper bound of the number of XMM registers used must
be passed in AL. (Not reflected in the unit tests here, but will be in
the dependent CL.)
On ARM (32 bit), using varargs forces the calling convention to be in
softfp mode even on hardfp supported devices.
On RISC-V, the FPU registers are blocked when using varargs.
TEST=runtime/vm/compiler/ffi/native_calling_convention_test.cc
Test outputs in: runtime/vm/compiler/ffi/unit_tests/variadic_*
Run test with `tools/test.py ffi_unit`.
Bug: https://github.com/dart-lang/sdk/issues/38578
Change-Id: Ic568f8156c1c28ac3d6a2144805edf8caaa0169c
Cq-Include-Trybots: luci.dart.try:vm-precomp-ffi-qemu-linux-release-arm-try
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/278342
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Unnecessary boxing/unboxing instruction can be eliminated if
their operand already has a desired representation.
This change also schedules Canonicalize pass before
TypePropagation/SelectRepresentations_Final so it has a final chance
to cleanup IL after preceeding passes but before the representations are frozen
and certain canonicalizations are disallowed.
TEST=vm/cc/IL_Canonicalize_RepresentationChange
Related issue: https://github.com/dart-lang/sdk/issues/51019
Change-Id: I187767e3efa69497cdd1f44dd916a8b9fb9f7261
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/279274
Reviewed-by: Slava Egorov <vegorov@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Fixes a scaling limitation where compiling N functions using the same guarded field or CHA guarded interface will result in O(N^2) comparisons when registering the dependencies.
TEST=ci
Change-Id: Ic48f29fa0ce8f43a6f60bc00fa95a60763d333a0
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/278943
Commit-Queue: Ryan Macnak <rmacnak@google.com>
Reviewed-by: Alexander Aprelev <aam@google.com>
We should not use dart_sysroot outside of our own
build/config/sysroot.gni, Flutter buildroot does not
define this variable.
We don't actually need -Wall -Werror -Wextra in dart_config
because default config already contains chromium_code config
which does this.
Change-Id: Id7843809dcecc8871af1df6b8d3899511545cd51
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/279355
Commit-Queue: Slava Egorov <vegorov@google.com>
Reviewed-by: Daco Harkes <dacoharkes@google.com>
Reviewed-by: Alexander Thomas <athom@google.com>
Adds support for `asset: 'asset-id'` in `FfiNative`s.
Symbols in assets are resolved by
1. the provided [asset] (the bulk of this CL),
2. the native resolver set with `Dart_SetFfiNativeResolver` in
`dart_api.h` (old behavior), and
3. the current process.
Symbols in assets are resolved by the Dart VM through the mapping
provided in the `vm:ffi:native_assets` library:
```
@pragma('vm:ffi:native-assets', {
"linux_x64": {
"<asset1>": [
"absolute",
"<absolute-path>"
],
"<asset2>": [
"relative",
"<relative-path>"
],
"<asset3>": [
"system",
"<system-path>"
],
"<asset4>": [
"process",
],
"<asset5>": [
"executable",
],
}
})
library;
```
There are 5 asset path types. Symbols resolved in these asset types are
resolved as follows.
A. Absolute paths are dlopened, after which the symbol is looked up.
B. Relative paths are resolved relative to `Platform.script.uri`, and
dlopened, after which the symbol is looked up.
C. System paths are treated as absolute paths. (But we might explicitly
try to use `PATH` in the future.)
D. Symbols looked up in Process assets are looked up in the current
process.
E. Symbols looked up in Executable assets are looked up in the current
executable.
The native assets mapping can be provided in three ways.
X. In the invocation to `gen_kernel` with the `--native-assets`
argument. This uses the gen_kernel entry point for compiling
kernel.
Y. By placing a `native_assets.yaml` next to the
`package_config.json` used by the isolate group. This works for
`dart <source file>` and `Isolate.spawnUri(<dart source file>)`.
This uses the kernel_service entry point for compiling kernel.
Z. By sending a String containing the native assets mapping when
invoking the kernel_service directly from within the VM. Currently,
only used for unit tests.
TEST=tests/ffi/native_assets/asset_*_test.dart (X)
TEST=tests/ffi/native_assets/infer_native_assets_yaml_*.dart (Y)
TEST=runtime/vm/ffi/native_assets_test.cc (Z)
The library is synthesized from yaml in pkg:vm as AST.
Design doc: go/dart-native-assets
Bug: https://github.com/dart-lang/sdk/issues/49803
Change-Id: I8bf7000bfcc03b362948efd3baf168838948e6b9
Cq-Include-Trybots: luci.dart.try:vm-ffi-android-debug-arm-try,vm-ffi-android-debug-arm64c-try,vm-precomp-ffi-qemu-linux-release-arm-try,vm-kernel-nnbd-mac-debug-arm64-try,vm-kernel-nnbd-mac-debug-x64-try,vm-kernel-win-debug-x64-try,vm-kernel-win-debug-ia32-try,vm-kernel-precomp-win-debug-x64c-try,vm-kernel-reload-linux-debug-x64-try,vm-kernel-reload-rollback-linux-debug-x64-try,vm-kernel-precomp-nnbd-linux-debug-x64-try,vm-kernel-precomp-win-debug-x64c-try,vm-kernel-precomp-nnbd-mac-release-arm64-try
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/264842
Reviewed-by: Martin Kustermann <kustermann@google.com>
In dartfuzz_api_table.dart, both `DartType.VOID` and `DartType.VOID_NULLABLE` are added to the Map keys. At runtime, the last value for a given key wins, so I removed the `DartType.VOID` key-value pair.
In gen_api_table.dart, both `assertionErrorEncoding` and `errorEncoding` are added to the Map keys. At runtime, the last value for a given key wins, so I removed the `assertionErrorEncoding` key-value pair.
This CL should be a no-op.
Bug: https://github.com/dart-lang/sdk/issues/50796
Change-Id: Ifcfeb3e84930267f60efd2eb009a722f8ac2453e
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/279234
Reviewed-by: Tess Strickland <sstrickl@google.com>
Commit-Queue: Samuel Rawlins <srawlins@google.com>
Our compiler shouldn't depend on current isolate, since it can
run on any isolate within an IG.
Doing this change, reveals two existing dependencies on current
isolate from compiler
- resolving native symbols in unoptimized compilations
- issuing of debug events for breakpoints
For the former we'll re-enter the currently active isolate that
triggered unoptimized compilation.
=> We may want to change that embedder API to not be based on
handles and instead give embedder a simple `const char*`.
For the ladder we'll enter the isolate corresponding to the
breakpoint debug event to be issued. We are at place where
all mutators are stopped, so that does seem okish.
=> Future could remove this by making Object Id Ring per-IG
Issue https://github.com/dart-lang/sdk/issues/48523
TEST=service_2/break_on_function_many_child_isolates_test/dds
Change-Id: Id246db5972ae505e82f637ce04bb2302bed76257
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/278901
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
The DWARF generator would always attempt to deobfuscate the URL
retrieved from `Script` objects, whether it came from the `url`
or `resolved_url` field. However, the `resolved_url` field is not
obfuscated, and thus attempting to deobfuscate it could result in a different (and incorrect) string. Thus, only deobfuscate when not
resolving DWARF paths.
TEST=vm/dart{,_2}/use_resolve_dwarf_paths_flag
Bug: https://github.com/dart-lang/sdk/issues/44325
Change-Id: I57a3d47cc78035508082ccba23a2b1cd726ed6bb
Cq-Include-Trybots: luci.dart.try:vm-kernel-precomp-linux-release-x64-try
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/279086
Commit-Queue: Tess Strickland <sstrickl@google.com>
Reviewed-by: Daco Harkes <dacoharkes@google.com>
This enables running other C/C++ code before the Dart main function
without having to implement the full embedder API by including
main_lib and friends in a build but excluding `main.cc`.
TEST=ci build + flutter bot
Change-Id: I5739eb21fa6918b0a2c0f33f7d97dad73c5cf0b1
Cq-Include-Trybots: luci.dart.try:flutter-engine-linux-try,flutter-linux-try
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/278518
Commit-Queue: Daco Harkes <dacoharkes@google.com>
Reviewed-by: Slava Egorov <vegorov@google.com>
We do not want to lose coverage when no longer running _2 test suites.
Those tests were originally written at a time when package:front_end
wasn't migrated yet - and as such couldn't be run in sound null safety.
TEST=ci
Change-Id: Id26da7e5896345a9e165829ba4a22d36695b6da3
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/279007
Reviewed-by: William Hesse <whesse@google.com>
Commit-Queue: Martin Kustermann <kustermann@google.com>
* ARGUMENT_TYPE_NOT_ASSIGNABLE_TO_ERROR_HANDLER
* BODY_MIGHT_COMPLETE_NORMALLY_CATCH_ERROR
* BODY_MIGHT_COMPLETE_NORMALLY_NULLABLE
* DEPRECATED_EXTENDS_FUNCTION
* DEPRECATED_IMPLEMENTS_FUNCTION
* DEPRECATED_MIXIN_FUNCTION
* DEPRECATED_NEW_IN_COMMENT_REFERENCE
This requires allowing these warnings to be reported in language tests without being expected.
Also, observatory now must comply with BODY_MIGHT_COMPLETE_NORMALLY_NULLABLE (or we could ignore the diagnostic).
TESTED=unit tests
Change-Id: I8facfa48a8645bac5bf8037163076a8f0a0ad05e
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/275740
Reviewed-by: Ben Konyi <bkonyi@google.com>
Reviewed-by: Phil Quitslund <pquitslund@google.com>
Reviewed-by: Brian Wilkerson <brianwilkerson@google.com>
Reviewed-by: Bob Nystrom <rnystrom@google.com>
Commit-Queue: Samuel Rawlins <srawlins@google.com>
* Do not print redundant ParallelMove instructions unless
`--print_redundant_il` is specified.
* Do not print `;; Inlined` comments if
`--no_include_inlining_info_in_disassembly` is specified.
TEST=manually
Change-Id: I9a8c83514e2a749484863a4cc010c8ea7950b0fb
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/278748
Commit-Queue: Slava Egorov <vegorov@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Number of parent type arguments in function types is critical
for instantiation and subtype checks when there are nested
generic function types. So it is important to update
number of parent type arguments when parent function type is updated.
This change adds mechanism for updating number of parent
type arguments and uses the new mechanism in 2 cases:
1) When a type parameter is substituted during type instantiation.
The substituted type can have nested generic function types and
it should be updated if substituted inside another generic
function type.
2) When creating a constructor tear-off for a generic class.
Parameter types of the constructor could have nested generic
function types and they should be updated as tear-off itself is a
generic function.
TEST=language/function_subtype/generic_function_type_substitution_test
TEST=language/regress/regress50905_test
Fixes https://github.com/dart-lang/sdk/issues/50614
Fixes https://github.com/dart-lang/sdk/issues/50905
Change-Id: If13baa420939f5ac002ce32f3909953fb6998bd2
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/278525
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Josh Soref