✔ More informative error messages in case of crashes.
✔ Handling and documenting different cases.
✔ Test cases for different cases.
✔ Make SDKDependencies.swift pass again.
We generate .swiftsourceinfo for stdlib in the build directory because ABI checker
could issue diagnostics to the stdlib source. However, this may also change other
diagnostic tests. Both Brent and Jordan have raised concern about this. After
adding this flag, other diagnostic tests could ignore .swiftsourceinfo files even
though when they are present so our tests will reflect what most users experience
when sources for stdlib are unavailable.
After setting up the .swiftsourceinfo file, this patch starts to actually serialize
and de-serialize source locations for declaration. The binary format of .swiftsourceinfo
currently contains these three records:
BasicDeclLocs: a hash table mapping from a USR ID to a list of basic source locations. The USR id
could be retrieved from the following DeclUSRs record using an actual decl USR. The basic source locations
include a file ID and the results from Decl::getLoc(), ValueDecl::getNameLoc(), Decl::getStartLoc() and Decl::getEndLoc().
The file ID could be used to retrieve the actual file name from the following SourceFilePaths record.
Each location is encoded as a line:column pair.
DeclUSRS: a hash table mapping from USR to a USR ID used by location records.
SourceFilePaths: a hash table mapping from a file ID to actual file name.
BasicDeclLocs should be sufficient for most diagnostic cases. If additional source locations
are needed, we could always add new source location records without breaking the backward compatibility.
When de-serializing the source location from a module-imported decl, we calculate its USR, retrieve the USR ID
from the DeclUSRS record, and use the USR ID to look up the basic location list in the BasicDeclLocs record.
For more details about .swiftsourceinfo file: https://forums.swift.org/t/proposal-emitting-source-information-file-during-compilation
Structurally prevent a number of common anti-patterns involving generic
signatures by separating the interface into GenericSignature and the
implementation into GenericSignatureBase. In particular, this allows
the comparison operators to be deleted which forces callers to
canonicalize the signature or ask to compare pointers explicitly.
Most of AST, Parse, and Sema deal with FileUnits regularly, but SIL
and IRGen certainly don't. Split FileUnit out into its own header to
cut down on recompilation times when something changes.
No functionality change.
Removes duplicated logic from the implementations of
FileUnit::lookupValue, and simplifies the interface to
ModuleDecl::lookupValue, where everyone was passing an empty
(non-filtering) access path anyway /except/ during actual lookup from
source code. No functionality change.
...fulfilling the promised audit from 0747d9a339. No intended
functionality change /other/ than the order of already-unsorted lists.
This affected a number of SIL tests that relied on deserialization
order matching the original source order; I have no idea why the old
hash logic would make that the case. If we think that's a valuable
property, we should serialize a list of functions in addition to the
iterable table. (Maybe just in SIB mode?)
A generic environment is always serialized as a GenericSignature with
a lazily-recreated environment, though sometimes it has to include
extra info specifically for generic environments used by SIL. The code
that was doing this claimed a bit for disambiguating between the two,
shrinking the permitted size of a compiled module from 2^31 bits to
2^30. (The code isn't just needlessly complicated; GenericEnvironments
used to be serialized with more information.)
Rather than have two representations for GenericEnvironmentID, this
commit just drops it altogether in favor of referencing
GenericSignatures directly. This causes a negligible file size
shrinkage for swiftmodules in addition to eliminating the problematic
disambiguation bit.
For now, the Deserialization logic will continue to cache
GenericEnvironments that are used directly by Deserialization, but
really that should probably be done at the AST level. Then we can
simplify further to ModuleFile tracking a plain list of
GenericSignatures.
ABI checker imports Swift frameworks by using Swift interfaces for various
reasons. The existing way of controlling preferred importing mechanism is by
setting an environment variable (SWIFT_FORCE_MODULE_LOADING), which may lead
to performance issues because the stdlib could also be loaded in this way.
This patch adds a new front-end option to specify module names for
which we prefer to importing via Swift interface. The option currently is only
accessible via swift-api-digester.
rdar://54559888
VBR fields can store up to 64 bits of info in N-bit units, where the
top bit describes whether there are further units to come. See
http://llvm.org/docs/BitCodeFormat.html for more information. In the
cases of the fields changed here, these are all usually-small values
that can /occasionally/ get a lot larger; a VBR field allows us to
pick a reasonable default while still not setting a maximum.
Swiftmodule size micro-optimization. No functionality change; due to
LLVM bitstream being a self-describing container format, this does not
actually break compatibility.
...by making it a tagged union of either a DeclID or a
LocalDeclContextID. This should lead to smaller module files and be
slightly more efficient to deserialize, and also means that every
AST entity kind is serialized in exactly one way, which allows for
the following commit's refactoring.
Rather than storing the set of input requirements in a
(SIL)SpecializeAttr, store the specialized generic signature. This
prevents clients from having to rebuild the same specialized generic
signature on every use.
This provides a singular instruction for convert an unmanaged value to a ref,
then strong_retain it. I expanded the definition of UNCHECKED_REF_STORAGE to
include these copy like instructions. This instruction is valid in all SIL.
The reason why I am adding this instruction is that currently when we emit an
access to an unowned (unsafe) ivar, we use an unmanaged_to_ref and a strong
retain. This can look to the optimizer like a strong retain that can potentially
be optimized. By combining the two together into a new instruction, we can avoid
this potential problem since the pattern matching will break.
We already do this for other ASTContext-allocated types (see
Decl.cpp). This will prevent the sort of mistakes in the previous two
commits.
Note that if any particular subclass of FileUnit wants to have its
destructor run, it can opt into that manually using
ASTContext::addDestructorCleanup. SourceFile and BuiltinUnit both do
this. But we generally don't /want/ to do this if we can avoid it
because it adds to compiler teardown time.
Implementing it in LoadedFile is nice in theory, but causes a leak in
practice because that type is ASTContext-allocated and usually never
destroyed.
https://bugs.swift.org/browse/SR-11366
...rather than the buffer, for a compiled module that came from a
module interface.
This was already happening at a higher level
(ModuleDecl::getModuleFilename) so pushing it down to the low-level
ModuleFile::getModuleFilename doesn't really change things much. The
important fix that goes with this is that SerializedASTFile no longer
leaks this name by storing it outside of ModuleFile.
https://bugs.swift.org/browse/SR-11365
The code tried extra hard to make sure this stayed in sync with the
returned ValidationInfo, but there's no real reason for that. Just
keep a single "HasError" bit for checking that the module's not being
obviously misused. (I snuck some groundwork for this into the previous
commit.)
`ModuleFile::error` was being used both for errors of initial parse
and configuration (non-fatal) and format errors during actual
deserialization (fatal, indicating a corrupted module). Split out the
latter to `ModuleFile::fatal()` (to go with the existing
`ModuleFile::fatal(llvm::Error)`) and be more consistent about
explicitly setting statuses for the former.
Since 'fatal()' is always fatal, this also allows deleting dummy
recovery code that would never be used in practice.
Previously, cycle-breaking logic would delay certain actions until all
re-entrant deserialization was complete for a particular module. That
hasn't been used in a while, though, now that the AST itself supports
more laziness, so let's take it out.
No functionality change; this really was unused.
Double-underscored names are reserved for the C++ "implementation"
(language and standard library). Even though "__Consuming" isn't
likely to be part of the C++ standard any time soon, we should follow
the rules.
Note that the API digester will continue to use the string
"__Consuming" for now, even though the underscores aren't really
significant, to avoid invalidating existing dumps.
No functionality change.
This means that we no longer have the invariant that the extendedType always
contains the generic parameters. So we need to fix the assertions/test cases
for it.
The only place this was used in Decl.h was the failability kind of a
constructor.
I decided to replace this with a boolean isFailable() bit. Now that
we have isImplicitlyUnwrappedOptional(), it seems to make more sense
to not have ConstructorDecl represent redundant information which
might not be internally consistent.
Most callers of getFailability() actually only care if the result is
failable or not; the few callers that care about it being IUO can
check isImplicitlyUnwrappedOptional() as well.
This flag is set by DefinitInitialization if the lifetime of the stored value is controlled dynamically.
If the flag is set, it's not (easily) possibly to statically calculate the lifetime of the stored value.
This indicates that the "self" argument to the current function is always dynamically of the exact
static base class type, allowing metadata accesses in IRGen to use the local self metadata to answer
metadata requests for the class type. Set this attribute on allocating entry points of designated
inits, which is one of the most common places where we emit redundant metadata accesses.
Implicit accessors are sometimes transparent for performance reasons.
Previously this was done in Sema by maybeMarkTransparent(), which would
add a TransparentAttr. Replace this with a request.
Once accessors are no longer listed as members of their parent context,
a failure to deserialize a VarDecl or SubscriptDecl needs to create a
MissingMemberDecl with the total number of vtable entries expected for
all of the accessors of the storage.
Note that until the accessor change actually lands, we always compute
the expected number of vtable entries as 0.
...which allows the AST printer to correctly choose whether to print
it, which means it can be printed in a module interface in a non-WMO
build, which is necessary for @objc enums to have a correct run-time
representation when clients use that interface.
rdar://problem/53469608
Previously we would copy this attribute from a superclass to the
subclass when validating a subclass. However, this is only correct
if the superclass is always guaranteed to have been validated
before the subclass.
Indeed, it appears this assumption is no longer true, so we would
sometimes lose track of the attribute, which would result in SILGen
failing to emit the ivar initializer entry point.
Instead, check for the attribute as part of the superclass walk
in checkAncestry(), ensuring the result is always up to date, correct,
and cached.
As a follow-up, we should also convert checkAncestry() into a
request, but I'm not doing that here to keep the fix short.
Fixes <rdar://problem/50845438>.
