File IDs have been expressed using a 10-bit fixed field. This limits us
to 1023 different files in which we can emit diagnostics, where
overflowing would simply crash. With the introduction of macros and
their generated source buffers, we're much more likely to overflow.
Switch to a 5-bit VBR field, which is slightly smaller for the common
case where there are few files involving diagnositcs, and which also
allows us to have up to 2^32-1 file IDs.
When emitting generated source file buffers, we could end up
reallocating the DenseMap that tracks file IDs while still holding a
reference to it. Hilarity ensues. Stop that hilarity by taking a copy.
Additionally, make sure we don't emit an invalid source range, and
instead put in empty source locations.
Extend the serialized diagnostics file format to also support providing
the contents of source files, which can include a reference to the
"original source range" whose text is conceptually replaced with the
contents of the serialized diagnostics file, e.g., due to macro
expansion.
Introduce a new behavior when printing references to modules with an
`export_as` definition. Use the `export_as` name in the public swiftinterface
and the real module name in the private swiftinterface.
This has some limits but should still be an improvement over the current
behavior. First, the we use the `export_as` names only for references to clang
decls, not Swift decls with an underlying module defining an `export_as`.
Second, we always print the `export_as` name in the public swiftinterface,
even in the original swiftinterface file when the `export_as` target is likely
not know, so that generated swiftinterface is still broken.
This behavior is enabled by the flags `-enable-experimental-feature ModuleInterfaceExportAs`
or the `SWIFT_DEBUG_USE_EXPORTED_MODULE_NAME_IN_PUBLIC_ONLY` env var. We may
consider turning it on by default in the future.
rdar://98532918
Introduces a concept of a dependency scanning action context hash, which is used to select an instance of a global dependency scanning cache which gets re-used across dependency scanning actions.
For spatial locality on startup.
Hide collocating metadata functions in a separate section behind a flag.
The default is not to collocate functions.
rdar://101593202
Currently headers produced with `-emit-objc-header` /
`-emit-objc-header-path` produce headers that include modular imports.
If the consumer wishes to operate without modules enabled, these headers
cannot be used. This patch introduces a new flag
(`-emit-clang-header-nonmodular-includes`) that when enabled
attempts to argument each modular import included in such a header with
a set of equivalent textual imports.
When opaque values are enabled, TypeConverter associates to an
address-only type an OpaqueValueTypeLowering. That lowering stores a
single lowered SIL type, and its value category is "object". So long as
the module has not yet been address-lowered, that type has the
appropriate value category. After the module has been address-lowered,
however, that type has the wrong value category: the type is
address-only, and in an address-lowered module, its lowered type's value
category must be "address".
Code that obtains a lowered type expects the value category to reflect
the state of the module. So somewhere, it's necessary to fixup that
single lowered type's value category.
One option would be to update all code that uses lowered types. That
would require many changes across the codebase and all new code that
used lowered types would need to account for this.
Another option would be to update some popular conveniences that call
through to TypeConverter, for example those on SILFunction, and ensure
that all code used those conveniences. Even if this were done
completely, it would be easy enough for new code to be added which
didn't use the conveniences.
A third option would be to update TypeLowering::getLoweredType to take
in the context necessary to determine whether the stored SILType should
be fixed up. That would require each callsite to be changed and
potentially to carry around more context than it already had in order to
be able to pass it along.
A fourth option would be to make TypeConverter aware of the
address-loweredness, and to update its state at the end of
AddressLowering.
Updating TypeConverter's state would entail updating all cached
OpaqueValueTypeLowering instances at the end of the AddressLowering
pass. Additionally, when TypeConverter produces new
OpaqueValueTypeLowerings, they would need to have the "address" value
category from creation.
Of all the options, the last is least invasive and least error-prone, so
it is taken here.
* Introduce TypeLayout Strings
Layout strings encode the structure of a type into a byte string that can be
interpreted by a runtime function to achieve a destroy or copy. Rather than
generating ir for a destroy/assignWithCopy/etc, we instead generate a layout
string which encodes enough information for a called runtime function to
perform the operation for us. Value witness functions tend to be quite large,
so this allows us to replace them with a single call instead. This gives us the
option of making a codesize/runtime cost trade off.
* Added Attribute @_GenerateLayoutBytecode
This marks a type definition that should use generic bytecode based
value witnesses rather than generating the standard suite of
value witness functions. This should reduce the codesize of the binary
for a runtime interpretation of the bytecode cost.
* Statically link in implementation
Summary:
This creates a library to store the runtime functions in to deploy to
runtimes that do not implement bytecode layouts. Right now, that is
everything. Once these are added to the runtime itself, it can be used
to deploy to old runtimes.
* Implement Destroy at Runtime Using LayoutStrings
If GenerateLayoutBytecode is enabled, Create a layout string and use it
to call swift_generic_destroy
* Add Resilient type and Archetype Support for BytecodeLayouts
Add Resilient type and Archetype Support to Bytecode Layouts
* Implement Bytecode assign/init with copy/take
Implements swift_generic_initialize and swift_generic_assign to allow copying
types using bytecode based witnesses.
* Add EnumTag Support
* Add IRGen Bytecode Layouts Test
Added a test to ensure layouts are correct and getting generated
* Implement BytecodeLayouts ObjC retain/release
* Fix for Non static alignments in aligned groups
* Disable MultiEnums
MultiEnums currently have some correctness issues with non fixed multienum
types. Disabling them for now then going to attempt a correct implementation in
a follow up patch
* Fixes after merge
* More fixes
* Possible fix for native unowned
* Use TypeInfoeBasedTypeLayoutEntry for all scalars when ForceStructTypeLayouts is disabled
* Remove @_GenerateBytecodeLayout attribute
* Fix typelayout_based_value_witness.swift
Co-authored-by: Gwen Mittertreiner <gwenm@fb.com>
Co-authored-by: Gwen Mittertreiner <gwen.mittertreiner@gmail.com>
`getValue` -> `value`
`getValueOr` -> `value_or`
`hasValue` -> `has_value`
`map` -> `transform`
The old API will be deprecated in the rebranch.
To avoid merge conflicts, use the new API already in the main branch.
rdar://102362022
Currently headers produced with `-emit-objc-header` /
`-emit-objc-header-path` produce headers that include modular imports.
If the consumer wishes to operate without modules enabled, these headers
cannot be used. This patch introduces a new flag
(`-emit-clang-header-nonmodular-includes`) that when enabled
attempts to argument each modular import included in such a header with
a set of equivalent textual imports.
