With this change and some other changes that I am committing in parallel, the
stdlib and all of the overlays send all proper pass manager notifications.
rdar://42301529
Previously SILModule contained two different pathways for the deserializer to
send notifications that it had created functions:
1. A list of function pointers that were called when a function's body was
deserialized. This was added recently so that access enforcement elimination is
run on newly deserialized SIL code if we have already eliminated access
enforcement from the module.
2. SILModule::SerializationCallback. This is an implementation of the full
callback interface and is used by the SILModule to update linkage and other
sorts of book keeping.
To fix the pass manager notification infrastructure, I need to be able to send
notifications to a SILPassManager when deserializing. I also need to be able to
eliminate these callbacks when a SILPassManager is destroyed. These requirements
are incompatible with the current two implementations since: (2) is an
implementation detail of SILModule and (1) only notifies on function bodies
being deserialized instead of the creation of new declarations (what the caller
analysis wants).
Rather than adding a third group of callbacks, this commit refactors the
infrastructure in such a way that all of these use cases can use one
implementation. This is done by:
1. Lifting the interface of SerializedSILLoader::Callback into a base
notification protocol for deserialization called
DeserializationNotificationHandlerBase and its base no-op implementation into an
implementation of the aforementioned protocol:
DeserializationNotificationHandler.
2. Changing SILModule::SerializationCallback to implement
DeserializationNotificationHandler.
3. Creating a class called FunctionBodyDeserializationNotificationHandler that
takes in a function pointer and uses that to just override the
didDeserializeFunctionBody. This eliminates the need for the specific function
body deserialization list.
4. Replacing the state associated with the two other pathways with a single
DeserializationNotificationHandlerSet class that contains a set of
DeserializationNotificationHandler and chains notifications to them. This set
implements DeserializationNotificationHandlerBase so we know that its
implementation will always be in sync with DeserializationNotificationHandler.
rdar://42301529
DiagnoseStaticExclusivity no longer asserts as follows when
non-escaping closures are passed to withoutActuallyEscaping:
Applied argument must be @noescape function type: ...
A partial_apply with @inout_aliasable may only be used as a @noescape
function type argument.
Subsequent commits will improve diagnostics to detect actual conflicts
in these situations.
Fixes <rdar://problem/43059088> Assertion in DiagnoseStaticExclusivity.
ConvertFunction and reabstraction thunks need this attribute. Otherwise,
there is no way to identify that withoutActuallyEscaping was used
to explicitly perform a conversion.
The destination of a [without_actually_escaping] conversion always has
an escaping function type. The source may have either an escaping or
@noescape function type. The conversion itself may be a nop, and there
is nothing distinctive about it. The thing that is special about these
conversions is that the source function type may have unboxed
captures. i.e. they have @inout_aliasable parameters. Exclusivity
requires that the compiler enforce a SIL data flow invariant that
nonescaping closures with unboxed captures can never be stored or
passed as an @escaping function argument. Adding this attribute allows
the compiler to enforce the invariant in general with an escape hatch
for withoutActuallyEscaping.
Certain uses of protocols only formally need the requirement
signature, not any of the method requirements. This results in IRGen
seeing a protocol where none of the members have been validated except
the associated types. Account for this by allowing ProtocolInfo to
only contain the layout for the base protocols and associated types,
if requested.
Note that this relies on the layout of a witness table always putting
the "requirement signature part" at the front, or at least at offsets
that aren't affected by function requirements.
rdar://problem/43260117
1) It's possible to materialize a tuple value with an @escaping or
@autoclosure element in it.
I don't think this causes any bad behavior in 4.2 because these
flags have no semantic effect after the type checker, but now
I'm adding an assertion that will fire when such types are
serialized, so let's make sure it doesn't happen by explicitly
clearing out these flags when lowering tuples types.
2) It's also possible to materialize a tuple with a single vararg
element. Again, this was not a problem in 4.2, but with the above
change to start clearing tuple flags, we now end up in a
situation where the lowered type is not a tuple, because
TupleType::get() returns a ParenType if the tuple has one
element that is not vararg (which it no longer is, because we
just cleared all the flags).
Fix the second problem by treating one-element vararg tuples just
like tuples with inout, __shared and __owned elements, that is,
by always exploding them when they appear at the top level of a
function parameter list, ensuring we never try to materialize
a value whose type is the entire tuple type.
These problems all stem from the fact that lowering a function type
with the opaque abstraction pattern treats the top level argument
list as a single tuple argument. Once that is fixed, much of the
above will simplify down to assertions.
These are trivial functions and should be inlined away; the only
tricky bit is they need to be defined after the iterator type.
This gives a slight speedup of stdlib compilation time (about 5%
of the time spent generating the swiftmodule).
This works around a potential circular dependence issue where TypeSubstCloner
needs access to SILOptFunctionBuilder but is in libswiftSIL.
rdar://42301529
Summary:
CodeView does not recognize zero as an artificial line location
and so a line location of zero causes unexpected behavior when
stepping through user code. If we find a line location of zero
and our scope has not changed, we use the most recent debug
location. That is expected to be the user code that most likely
relates to the current instruction and is similar to the behavior
in LLDB.
Test Plan:
test/DebugInfo/linetable-codeview.swift
This allowed me to fold all of the weird direct calls to createFunction into a
singular createFunctionForForwardReference. This is the only API that is needed
by the SILParser so only providing that gives us a significantly cleaner API.
rdar://42301529
This patch adds SIL-level debug info support for variables whose
static type is rewritten by an optimizer transformation. When a
function is (generic-)specialized or inlined, the static types of
inlined variables my change as they are remapped into the generic
environment of the inlined call site. With this patch all inlined
SILDebugScopes that point to functions with a generic signature are
recursively rewritten to point to clones of the original function with
new unique mangled names. The new mangled names consist of the old
mangled names plus the new substituions, similar (or exactly,
respectively) to how generic specialization is handled.
On libSwiftCore.dylib (x86_64), this yields a 17% increase in unique
source vars and a ~24% increase in variables with a debug location.
rdar://problem/28859432
rdar://problem/34526036
This commit does not modify those APIs or their usage. It just:
1. Moves the APIs onto SILFunctionBuilder and makes SILFunctionBuilder a friend
of SILModule.
2. Hides the APIs on SILModule so all users need to use SILFunctionBuilder to
create/destroy functions.
I am doing this in order to allow for adding/removing function notifications to
be enforced via the type system in the SILOptimizer. In the process of finishing
off CallerAnalysis for FSO, I discovered that we were not doing this everywhere
we need to. After considering various other options such as:
1. Verifying after all passes that the notifications were sent correctly and
asserting. Turned out to be expensive.
2. Putting a callback in SILModule. This would add an unnecessary virtual call.
I realized that by using a builder we can:
1. Enforce that users of SILFunctionBuilder can only construct composed function
builders by making the composed function builder's friends of
SILFunctionBuilder (notice I did not use the word subclass, I am talking
about a pure composition).
2. Refactor a huge amount of code in SILOpt/SILGen that involve function
creation onto a SILGenFunctionBuilder/SILOptFunctionBuilder struct. Many of
the SILFunction creation code in question are straight up copies of each
other with small variations. A builder would be a great way to simplify that
code.
3. Reduce the size of SILModule.cpp by 25% from ~30k -> ~23k making the whole
file easier to read.
NOTE: In this commit, I do not hide the constructor of SILFunctionBuilder since
I have not created the derived builder structs yet. Once I have created those in
a subsequent commit, I will hide that constructor.
rdar://42301529
Add support to static diagnostics for tracking noescape closures through block
arguments.
Improve the noescape closure SIL verification logic to match. Cleanup noescape
closure handling in both diagnostics and SIL verification to be more
robust--they must perfectly match each other.
Fixes <rdar://problem/42560459> [Exclusivity] Failure to statically diagnose a
conflict when passing conditional noescape closures.
Initially reported in [SR-8266] Compiler crash when checking exclusivity of
inout alias.
Example:
struct S {
var x: Int
mutating func takeNoescapeClosure(_ f: ()->()) { f() }
mutating func testNoescapePartialApplyPhiUse(z : Bool) {
func f1() {
x = 1 // expected-note {{conflicting access is here}}
}
func f2() {
x = 1 // expected-note {{conflicting access is here}}
}
takeNoescapeClosure(z ? f1 : f2)
// expected-error@-1 2 {{overlapping accesses to 'self', but modification requires exclusive access; consider copying to a local variable}}
}
}
Allows a SIL pass to follow a def-use chain through phis.
Other terminators can also propagate values through block arguments, but they
always need special handling.
At least most of these were latent bugs since the code was
unreachable in the PartialApply case. But that's no excuse to misuse
the API.
Also, whenever referring to an integer index, be explicit about
whether it is an applied argument or callee argument.
Summary:
CodeView does not recognize zero as an artificial line location
and so a line location of zero causes unexpected behavior when
stepping through user code. If we find a line location of zero
and our scope has not changed, we use the most recent debug
location. That is expected to be the user code that most likely
relates to the current instruction and is similar to the behavior
in LLDB.
Test Plan:
test/DebugInfo/linetable-codeview.swift
print and parse as a stable hexadecimal form that isn't interpreted as UTF8.
One use case is in representing serialized protobuf strings (as in the
tensorflow branch: f7ed452eba/lib/SILOptimizer/Mandatory/TFPartition.cpp (L3875)).
The original work was done by @lattner and merged into the tensorflow
branch. This PR is to upstream those changes.
For now, the accessors have been underscored as `_read` and `_modify`.
I'll prepare an evolution proposal for this feature which should allow
us to remove the underscores or, y'know, rename them to `purple` and
`lettuce`.
`_read` accessors do not make any effort yet to avoid copying the
value being yielded. I'll work on it in follow-up patches.
Opaque accesses to properties and subscripts defined with `_modify`
accessors will use an inefficient `materializeForSet` pattern that
materializes the value to a temporary instead of accessing it in-place.
That will be fixed by migrating to `modify` over `materializeForSet`,
which is next up after the `read` optimizations.
SIL ownership verification doesn't pass yet for the test cases here
because of a general fault in SILGen where borrows can outlive their
borrowed value due to being cleaned up on the general cleanup stack
when the borrowed value is cleaned up on the formal-access stack.
Michael, Andy, and I discussed various ways to fix this, but it seems
clear to me that it's not in any way specific to coroutine accesses.
rdar://35399664
Summary:
There are a few problems with how Swift currently emits location
information for CodeView.
1. WinDbg does not work well with column information so all column
locations must be set to zero.
2. Some instructions, e.g., ``a + b``, will emit ``@llvm.trap()``
and ``unreachable``. Those instructions should have artificial
locations, i.e., they should have a line location of zero.
3. Some instructions, e.g., ``a / b``, will emit ``unreachable``
sandwiched between other code for that instruction. This makes
WinDbg confused and it cannot decide which set of instructions
to break on. Those instructions should have the same line location
as the others.
4. There are several prologue instructions with artificial line
locations that create breaks in the linetables. Those instructions
should have valid line locations, usually at the start of the
function.
5. Case bodies have cleanup instructions with artificial line
locations unless it has a ``do`` block. Those locations should
be the last line in the case block.
Test Plan:
test/DebugInfo/basic.swift
test/DebugInfo/columns.swift
test/DebugInfo/linetable-codeview.swift
test/DebugInfo/line-directive-codeview.swift
