Add support for availability-style features #10098
Conversation
|
#9815 has the codegen and serialization changes. |
ahatanaka
requested review from
jroelofs,
nkcsgexi,
ributzka,
tshortli,
rjmccall and
devincoughlin
clang/lib/Parse/ParseExpr.cpp
Outdated
| @@ -3999,6 +4008,23 @@ std::optional<AvailabilitySpec> Parser::ParseAvailabilitySpec() { | |||
| Actions.CodeCompletion().CodeCompleteAvailabilityPlatformName(); | |||
| return std::nullopt; | |||
| } | |||
|
|
|||
| if (Tok.is(tok::identifier) && GetLookAheadToken(1).is(tok::equal)) { | |||
There was a problem hiding this comment.
suggest: tok::colon instead of tok:equal, so it matches the feat: <feature> style ones
clang/include/clang/Sema/ScopeInfo.h
Outdated
| ObjCShouldCallSuper(false), ObjCIsDesignatedInit(false), | ||
| ObjCWarnForNoDesignatedInitChain(false), ObjCIsSecondaryInit(false), | ||
| ObjCWarnForNoInitDelegation(false), NeedsCoroutineSuspends(true), | ||
| FoundImmediateEscalatingExpression(false), ErrorTrap(Diag) {} |
There was a problem hiding this comment.
suggest to not clang-format this after adding the new line, so we reduce the downstream diff.
clang/lib/AST/ASTContext.cpp
Outdated
| } | ||
|
|
||
| bool ASTContext::hasUnavailableFeature(const Decl *D) const { | ||
| if (D->hasAttrs()) |
There was a problem hiding this comment.
nitpick: using early exist instead of nesting the for loop inside the if check.
| @@ -3539,6 +3539,10 @@ def fno_builtin : Flag<["-"], "fno-builtin">, Group<f_Group>, | |||
| def fno_builtin_ : Joined<["-"], "fno-builtin-">, Group<f_Group>, | |||
| Visibility<[ClangOption, CC1Option, CLOption, DXCOption]>, | |||
| HelpText<"Disable implicit builtin knowledge of a specific function">; | |||
| def ffeature_availability_EQ : Joined<["-"], "ffeature-availability=">, Group<f_Group>, | |||
There was a problem hiding this comment.
Silly question: why do we need this flag if all the features are defined in a header?
There was a problem hiding this comment.
I added this flag for testing purposes before we decided on the availability domain definition spec. I can remove it as it's not needed anymore. I left it in as I thought it could be handy when testing or debugging the feature.
There was a problem hiding this comment.
Thanks for the explanation! this matches with what I thought.
|
@tshortli I added an API for retrieving the information about availability domain on a decl. 81991a7 The function returns a pair of feature name and |
|
It's still possible to get the |
Yes, that looks reasonable. Thanks! |
ahatanaka
force-pushed
the
feature-availability-sema
branch
from
95fa011 to
6d736cf
Compare
This commit extends the `__availability` attribute to accept a feature
name and a boolean-like integer argument (0 or 1) to indicate
availability based on feature presence or absence.
`__builtin_available` and `@available` take a feature name as their
arguments. References to annotated declarations in unguarded contexts
are rejected.
For example:
```
__attribute__((availability(domain:feature1, 0)))
void available_func(void);
__attribute__((availability(domain:feature1, 1)))
void unavailable_func(void);
if (__builtin_available(feature1)) {
available_func();
unavailable_func(); // error
} else {
available_func(); // error
unavailable_func();
}
```
rdar://137999979
ahatanaka
force-pushed
the
feature-availability-sema
branch
from
6d736cf to
b53735b
Compare
This patch doesn't include the codegen and serialization changes that are needed to support the availability-style features.
rdar://137999979