It is only safe to perform a destroy_value on an alloc_box that contains an
initialized value. We preserve the original cleanups for the value we are
pushing through the scope implying that the box will not contain an initialized
value when its lifetime ends (just like an alloc_stack). Thus we must use
dealloc_box here.
The surprising thing about tracking down this error is that I was not hitting
any memory issues at -Onone. Instead what was happening was that at -O, we were
miscompiling (creating non-dominating uses of SSA values) in the face of an
address being taken twice.
This does not seem to hit any SILGen tests today (I hit this when testing a
patch I am trying to land today).
rdar://31521023
These names are not perfect, but they provide more descriptive background on
what the parameter actually does. Simply, these parameters say that the
underlying base address will last longer than the usage of the underlying value,
so a begin_borrow/load_borrow could be used to access the underlying value from
the base. If the base is destroyed before the borrow finishes, we have an
ownership violation. =><=.
I also copy-edited/added doxygen comments above some of these methods as well.
The specific exposed problem had to do with my using the same emission routine
for both lvalues using delegating init self (where we want formal accesses) and
for routines that wanted normal access to self. By splitting them the issue is
resolved.
As a benefit, I added a small peephole that I needed to add for my own purposes
(i.e. to maintain invariants), but that also incidentally improve codegen in
other places!
rdar://31521023
Today, SILGenFunction::emitRValue assumes the caller will create any cleanup
scopes that are needed to cleanup side-effects relating to the rvalue
evaluation. The API also provides the ability for the caller to specify that a
+0 rvalues is an "ok" result. The API then tries to produce a +0 rvalue and
returns a +1 rvalue otherwise. These two properties create conflicting
requirements on the caller since the caller does not know whether or not it
should create a scope (if a +1 rvalue will be returned) or not (if a +0 rvalue
would be returned).
The key issue here is the optionality of returning a +0 rvalue. This change
begins to resolve this difference by creating two separate APIs that guarantee
to the caller whether or not a +0 or a +1 rvalue is returned and also creates
local scopes for the caller as appropriate. So by using these APIs, the caller
knows that the +0 or +1 rvalue that is returned has properly been put into the
caller scope. So the caller no longer needs to create its own scopes anymore.
emitPlusOneRValue is emitRValue except that it scopes the rvalue emission and
then *pushes* the produced rvalue through the scope. emitPlusZeroRValue is
currently a stub implementation that just calls emitPlusOneRValue and then
borrows the resulting +1 RValue in the outer scope, creating the +0 RValue that
was requested by the caller.
rdar://33358110
just for pointer identity.
The current technique for deciding whether that's the case is *extremely*
hacky and need to be replaced with an attribute, but I'm reluctant to
take that on so late in the schedule. The hack is terrible but not too
hard to back out in the future. Anyone who names a method like this just
to get the magic behavior knows well that they are not on the side of
righteousness.
rdar://33265254
conversions that reverse an implicit conversion done to align
foreign declarations with their imported types.
For example, consider an Objective-C method that returns an NSString*:
- (nonnull NSString*) foo;
This will be imported into Swift as a method returning a String:
func foo() -> String
A call to this method will implicitly convert the result to String
behind the scenes. If the user then casts the result back to NSString*,
that would normally be compiled as an additional conversion. The
compiler cannot simply eliminate the conversion because that is not
necessarily semantically equivalent.
This peephole recognizes as-casts that immediately reverse a bridging
conversion as a special case and gives them special power to eliminate
both conversions. For example, 'foo() as NSString' will simply return
the original return value. In addition to call results, this also
applies to call arguments, property accesses, and subscript accesses.
ground work for the syntactic bridging peephole.
- Pass source and dest formal types to the bridging routines in addition
to the dest lowered type. The dest lowered type is still necessary
in order to handle non-standard abstraction patterns for the dest type.
- Change bridging abstraction patterns to store bridged formal types
instead of the formal type.
- Improve how SIL type lowering deals with import-as-member patterns.
- Fix some AST bugs where inadequate information was being stored in
various expressions.
- Introduce the idea of a converting SGFContext and use it to regularize
the existing id-as-Any conversion peephole.
- Improve various places in SILGen to emit directly into contexts.
The main thing here is that we create a cleanup that inserts the
end_borrow_argument. Once I merge end_borrow with end_borrow_argument, there
will only be one cleanup.
rdar://31880847
through a few places.
This patch should be NFC for existing patterns, but it's preparing for
using SILGen's built-in bridging capabilities for more things.
SILBuilder now tracks data dependencies between instructions
that open existentials and uses of the opened type, so
SILGen's mechanism for this is no longer needed.
In particular, this simplifies ArchetypeCalleeBuilder.
This is an LValue component whose value is the class
reference inside of a class existential.
Unlike OpenOpaqueExistentialComponent, this is a logical
component, with a "writeback" consisting of wrapping the
new class reference in a class existential having the
same conformances as the original.
This is slightly awkward, but adding "by-address" operations
on class existentials, and projecting the payload out is
a big change and might not make sense for other reasons.
When lowering a LoadExpr, SILGen constructs an LValue
and loads from it to produce an RValue.
If a LoadExpr contains another LoadExpr, the innermost
LoadExpr builds its own LValue, which is then loaded
to an RValue, and turned back into an LValue by creating
a single ValueComponent.
When evaluating an OpenExistentialExpr inside an LValue,
we record the base expression and evaluate it as an LValue
later when we encounter the corresponding OpaqueValueExpr.
The problem is when this is combined with a nested
LoadExpr, we might be inside of a different LValue than
the original LValue that contained the OpenExistentialExpr.
This would trigger an assertion, because the mapping from
OpaqueValueExprs to their base expressions was per-LValue;
instead, it needs to be per-SILGenFunction.
conversions and extend lifetimes over the call.
Apply this logic to string-to-pointer conversions as well as
array-to-pointer conversions.
Fix the AST verifier to not blow up on optional pointer conversions,
and make sure we SILGen them correctly. There's still an AST bug
here, but I'll fix that in a follow-up patch.
This enables me to refactor some code in SILGenConstructor. Specifically this
code passes in a false for uninitialized and then creates the mark uninitialized
for itself.
NFC by default.
I had a much more comprehensive version of this patch that I had to
back out in order to just get something in to unblock progress.
There's a lot of technical debt in SILGen around l-values.
Enabling this will almost certainly not get past the SIL passes.
The reason I am doing this is that I am going to in the next couple of commits
change enum element dispatch to with or without semantic SIL use proper
ownership.
In this commit, I just did the copy and eliminated any parts of the code that
were predicated on having an address.
(This re-applies #7736 with an update to the
tsan-inout.swift execution test to handle configurations where
TSan's ignore_interceptors_accesses is enabled by default.)
Add SILGen instrumentation to treat inout accesses as Thread Sanitizer writes.
The goal is to catch races on inout accesses even when there is a not an
llvm-level read/write to a particular address. Ultimately
this will enable TSan to, for example, report racy writes to distinct
stored properties of a common struct as a data race.
This instrumentation is off by default. It can be enabled with the
'enable-experimental-tsan-inout-instrumentation' frontend flag.
The high-level approach is to add a SIL-level builtin that represents a call
to a TSan routine in compiler-rt. Then, when emitting an address for an LValue
as part of an inout expression, we call this builtin for each path component
that represents an LValue. I've added an 'isRValue()' method to PathComponent
that tracks whether a component represents an RValue or an LValue. Right the
only PathComponent that sometimes returns 'true' is ValueComponent().
For now, we're instrumenting only InoutExprs, but in the future it probably
makes sense to instrument all LValue accesses. In this patch I've
added a 'TSanKind' parameter to SILGenFunction::emitAddressOfLValue() and
its helpers to limit instrumentation to inout accesses. I envision that this
parameter will eventually go away.
Add SILGen instrumentation to treat inout accesses as Thread Sanitizer writes.
The goal is to catch races on inout accesses even when there is a not an
llvm-level read/write to a particular address. Ultimately
this will enable TSan to, for example, report racy writes to distinct
stored properties of a common struct as a data race.
This instrumentation is off by default. It can be enabled with the
'enable-experimental-tsan-inout-instrumentation' frontend flag.
The high-level approach is to add a SIL-level builtin that represents a call
to a TSan routine in compiler-rt. Then, when emitting an address for an LValue
as part of an inout expression, we call this builtin for each path component
that represents an LValue. I've added an 'isRValue()' method to PathComponent
that tracks whether a component represents an RValue or an LValue. Right the
only PathComponent that sometimes returns 'true' is ValueComponent().
For now, we're instrumenting only InoutExprs, but in the future it probably
makes sense to instrument all LValue accesses. In this patch I've
added a 'TSanKind' parameter to SILGenFunction::emitAddressOfLValue() and
its helpers to limit instrumentation to inout accesses. I envision that this
parameter will eventually go away.
This structure contains all of the type information that we use to build a
ResultPlanPtr. In a subsequent commit, I am going to move ResultPlanPtr creation
out of emitApply and place it before the creation of arguments. This is to
enable indirect result initializations to belong to the scope outside of any
argument based scopes.
Without this the lifetimes of the indirect result initializations and arguments
can not be separated without using hacks. There is no reason why we can't emit
the indirect result temporaries before we emit any arguments since they will
remain dormant until after the final apply/any future committed argument scopes
being popped.
rdar://30955427
Change emitApplyOfLibraryIntrinsic() to take a SubstitutionMap,
and use the correct abstractions to build the map.
This gets rid of the last remaining uses of gatherAllSubstitutions()
in SIL.
Previously, we would put a destroy_value directly on the value that we tried to
cast. Since checked_cast_br is consuming, this would cause the destroy_value on
the failure path to be flagged as a double consume.
This commit causes SILGen to emit the value consumed by checked_cast_br as an
@owned argument to the failure BB, allowing semantic arc rules to be respected.
As an additional benefit, I also upgraded the ownership_model_eliminator test to
use semantic sil verification.
One issue that did come up though is that I was unable to use the new code in
all locations in the compiler. Specifically, there is one location in
SILGenPattern that uses argument unforwarding. I am going to need to undo
argument unforwarding in SILGenPattern in order to completely eliminate the old
code path.
This is in preparation for removing the +0 self hack.
This commit in more detail does the following:
1. It adds Formal Evaluation Scopes to certain places where the scopes were
missing. Specifically:
a. The SILGenLValue cases are places where we are invoking accessors. In each
one of these cases, we had a formal evaluation scope in the accessor
itself, but we did not have a scope that closed over the base access and
the accessor access. The base access is a formal evaluation in the sense
that just like with inout bases, we must create a new reference to the
base and re-destroy the base in a chain of accesses. This is to ensure
that we do not extend the lifetime of the base inappropriately.
b. The SILGenPoly case is a place where we have never properly placed a
Formal Evaluation Scope and have completely been relying on the +0 self
hack to make sure that archetype callees are properly destroyed
immediately after a +0 call.
2. It changes all of the places in SILGen that emit self to using formal access
cleanups instead of normal cleanups.
rdar://29791263
Eventually I want to move all of these emit* APIs to SILGenBuilder and make them
private details of SILGenBuilder. But that is a refactoring for another time.
rdar://29791263
Previously, we were emitting these cleanups at the end of the lexical scope
instead of at the end of the formal evaluation scope. This change ensures that
we always emit the cleanup immediately at the end of the formal evaluation
scope.
Previously in most cases we got away with this due to the +0 self
hack. Basically we would emit a get for a self parameter and then immediately
use that self parameter as a guaranteed parameter. Then the hack would insert
the destroy value forwarding the lexical scope level cleanup at the same time.
rdar://29791263
