These search paths will not get used during Swift module compilation and can only hinder module sharing among different targets.
Resolves rdar://119217774
Generalize the existing `-playground-high-performance` flag into a set of options that control various aspects of the "playground transformation" in Sema.
This commit adds the first two of those controllable parts of the transform, matching what the existing flag already controls (scope entry/exit and function arguments), but in an extensible way. The intent is for this to be a scalable way to control a larger set of upcoming options.
So instead of a single flag, we represent the playground transform options as a set of well-defined choices, with a new `-playground-option` flag to individually enable or disable those options (when prefixed with "No", the corresponding option is instead disabled). Enabling an already-enabled option or disabling an already-disabled option is a no-op.
For compatibility, the existing `-playground-high-performance` flag causes "expensive" transforms to be disabled, as before. We can also leave it as a useful shorthand to include or exclude new options even in the future, based on their cost. There is a comment on the old function indicating that new code should use the more general form, but it remains for clients like LLDB until they can switch over.
The machinery for implementing the playground options is similar to how `Features.def` works, with a new `PlaygroundOptions.def` that defines the supported playground transform options. Each playground definition specifies the name and description, as well as whether the option is enabled by default, and whether it's also enabled in the "high performance" case.
Adding a new option in the future only requires adding it to `PlaygroundOptions.def`, deciding whether it should be on or off by default, deciding whether it should also be on or off in `-playground-high-performance` mode, and checking for its presence from the appropriate places in `PlaygroundTransform.cpp`.
Note that this is intended to control the types of user-visible results that the invoker of the compiler wants, from an externally detectable standpoint. Other flags, such as whether or not to use the extended form of the callbacks, remain as experimental features, since those deal with the mechanics and not the desired observed behavior.
rdar://109911673
Add a new flag to enable package interface loading.
Use the last value of package-name in case of dupes.
Rename PrintInterfaceContentMode as InterfaceMode.
Update diagnostics.
Test package interface loading with various scenarios.
Test duplicate package-name.
The current implementation of `-application-extension` has a problem that affects the generation of ObjC headers for regular Swift modules.
The primary purpose of `-application-extension` is to prevent the use of unavailable APIs in app extensions. However, it has an impact on the generation of -Swift.h headers and exposes Swift's internal declarations to ObjC. This behavior is appropriate for mixed modules that are not consumed externally, such as app extensions, but it fails to address the situation when a module is not an extension itself but is consumed by the extension (c90cd11).
To resolve this issue while maintaining the desired behavior, we can introduce a new flag for this particular use-case.
With the growing popularity of using Features, I think it's important to
ensure our Features querying set is as fast as possible. In particular,
we sometimes may end up putting a feature check in a hot function, so
the `contains` overhead is important.
While I haven't done *any* measurements to verify this, I think a
`llvm::SmallSet<Feature, 2>` is slower than a FixedBitSet. For the
former, if the set grows more than the constant size, it switches over
to using a `std::set`. Setting the size to `numFeatures()` is not
particularly attractive or possible, because `SmallSet` is limited to a
size of 32 to be considered small, as it uses linear search over a
SmallVector to service a `contains` query.
Meanwhile, a FixedBitSet has a small constant factor overhead for
querying if the element is contained: two divisions by a
constant power-of-two, a bit shift, and a memory-read / compare. I think
that'll beat a SmallSet in all cases.
Conflicts:
- `lib/AST/TypeCheckRequests.cpp` renamed `isMoveOnly` which requires
a static_cast on rebranch because `Optional` is now a `std::optional`.
By default the compiler will now replace the bodies of unavailable functions
with stubs that call `_diagnoseUnavailableCodeReached()` instead.
Resolves rdar://116019744
Using `-Rmodule-api-import` the compiler prints a remark about the
import bringing in every decl used in public function signatures or
inlinable code. It also remarks on the source of conformances where they
are used and the source of typealias underlying types.
An existing test (Frontend/skip-function-bodies.swift) was designed under the
assumption that multiple `-debug-forbid-typecheck-prefix` arguments were
already supported, and as a result the test was not actually asserting what it
was written to assert.
Macro implementations can come from various locations associated with
different search paths. Add a frontend flag `-Rmacro-loading` to emit
a remark when each macro implementation module is resolved, providing
the kind of macro (shared library, executable, shared library loaded
via the plugin server) and appropriate paths. This allows one to tell
from the build load which macros are used.
Addresses rdar://110780311.
When `-warn-on-potentially-unavailable-enum-case` was introduced, the build
system was required to invoke `swift-frontend` at artificially low deployment
targets when emitting `.swiftinterface` files for legacy architectures. Because
the deployment target was low, some availability diagnostics needed to be
de-fanged in order to allow module interface emission to succeed. Today, the
build system is able to use the correct deployment target when emitting module
interfaces and the `-warn-on-potentially-unavailable-enum-case` is superfluous,
so deprecate it.
Resolves rdar://114092047
Clang dependency scanning produces scanner PCMs which we may want to live in a
different filesystem location than the main build module cache.
Resolves rdar://113222853
Reformatting everything now that we have `llvm` namespaces. I've
separated this from the main commit to help manage merge-conflicts and
for making it a bit easier to read the mega-patch.
This is phase-1 of switching from llvm::Optional to std::optional in the
next rebranch. llvm::Optional was removed from upstream LLVM, so we need
to migrate off rather soon. On Darwin, std::optional, and llvm::Optional
have the same layout, so we don't need to be as concerned about ABI
beyond the name mangling. `llvm::Optional` is only returned from one
function in
```
getStandardTypeSubst(StringRef TypeName,
bool allowConcurrencyManglings);
```
It's the return value, so it should not impact the mangling of the
function, and the layout is the same as `std::optional`, so it should be
mostly okay. This function doesn't appear to have users, and the ABI was
already broken 2 years ago for concurrency and no one seemed to notice
so this should be "okay".
I'm doing the migration incrementally so that folks working on main can
cherry-pick back to the release/5.9 branch. Once 5.9 is done and locked
away, then we can go through and finish the replacement. Since `None`
and `Optional` show up in contexts where they are not `llvm::None` and
`llvm::Optional`, I'm preparing the work now by going through and
removing the namespace unwrapping and making the `llvm` namespace
explicit. This should make it fairly mechanical to go through and
replace llvm::Optional with std::optional, and llvm::None with
std::nullopt. It's also a change that can be brought onto the
release/5.9 with minimal impact. This should be an NFC change.
Teach swift dependency scanner to use CAS to capture the full dependencies for a build and construct build commands with immutable inputs from CAS.
This allows swift compilation caching using CAS.
Lazy member loading has been in use and the default for several years
now. However, the lazy loading was disabled for any type whose primary
definition was parsed even though some of its extensions could have
been deserialized, e.g., from a Clang module. Moreover, the non-lazy
path walked all of the extensions of such a type for all member name
lookup operations. Faced with a large number of extensions to the same
type (in my example, 6,000), this walk of the list of the extensions
could dominate type-checking time.
Eliminate all effects of the `-disable-named-lazy-member-loading`
flag, and always use the "lazy" path, which effectively does no work
for parsed type definitions and extensions thereof. The example with
6,000 extensions of a single type goes from type checking in 6 seconds
down to type checking in 0.6 seconds, and name lookup completely
disappears from the profiling trace.
The deleted tests relied on the flag that is now inert. They aren't by
themselves providing much value nowadays, and it's better to have the
simpler (and more efficient) implementation of member name lookup be
the only one.
Intro a deserialization mode controlled by the flag
`-experimental-force-workaround-broken-modules` to attempt unsafe
recovery from deserialization failures caused by project issues.
The one issue handled at this time is when a type moves from one module
to another. With this new mode the compiler may be able to pick a
matching type in a different module. This is risky to use, but may help
in a pinch for a client to fix and issue in a library over which they
have no control.
