C# 12 allows both on the explicitly typed parameter list of a lambda, and
nowhere else: an anonymous method cannot declare either, and neither can a
lambda whose parameter list is about to be dropped. Guarded by a setting so
the output stays valid for earlier language versions.
Only what the anonymous function's own metadata declares is written. A lambda
may state a default the delegate does not have, a different one, or none where
the delegate has one, and reflection over the lambda's method reports what the
lambda declared - so filling either in from the delegate's Invoke would make
the recompiled assembly describe itself differently from the original. Call
sites are unaffected either way, because they bind against the delegate, which
still declares both.
Roslyn writes ParamArrayAttribute on the anonymous function's own method only
from version 5 on; before that it stands on the delegate type alone, where it
is not the lambda's to restate, so the fixture guards those cases on ROSLYN5.
The correctness test reads the metadata back through reflection, which is the
only place the difference is observable.
Assisted-by: Claude:claude-opus-5[1m]:Claude Code
This was due to StackType.O doing double-duty as `object` and `other`.
While ExpressionBuilder would often improve the type of such locals, the `object` nevertheless ended up used in a couple of places, e.g. via the `typeHint`. This could result in value types being boxed even though the original IL didn't contain any `box` instruction.
This is an attempt to use better types for stack slot variables created by ILReader. The idea is: there aren't many IL instructions that produce "other" value types, and `InferType()` already handles pretty much all of them, so we can use that to assign types to our stack slots.
It's a bit more tricky if the stack is pushed to on multiple branches that join together before the value is used: here the variable type must be suitable for both assignments. In this case, we go back to the previously-used stacktype.
Shortening default(T) is the same problem as removing the redundant cast
around a lambda whose delegate type the context already fixes, so it uses
the same mechanism: ConvertTo makes the explicit type implicit when the
conversion is an identity conversion and the caller allows an implicit
one. The literal keeps the type it was shortened from, so any later
conversion to a different type - or any context that requires an explicit
type, such as an overload resolution recheck falling back to CastArguments
- can spell default(T) out again. That keeps the value intact where the
bare literal would change it, e.g. "object o = default(SomeStruct)", which
boxes a non-null struct while "default" would be null.
Because the shortened literal resolves to DefaultLiteralResolveResult,
CallBuilder's existing overload resolution recheck sees a real default
literal and rejects ambiguous calls on its own; no separate bookkeeping
about which arguments may stay untyped is needed. Only the contexts that
supply no target type at all restore the explicit form: an awaited
expression, and arguments of operator methods, which later become operator
or cast syntax rather than calls.
Assisted-by: Claude:claude-opus-5[1m]:Claude Code
The await surface had almost no fixture coverage beyond Task/ValueTask: every
GetAwaiter in the corpus was an instance method on the awaited type itself, so
the conversion VisitAwait applies to the operand was never exercised for an
inherited, interface-typed or extension-method awaiter. Probing that surface
turned up eight defects, all of which produce C# that does not compile.
AsyncAwaitPatterns pins the shapes that do round-trip, along the three axes the
translation actually depends on: the GetAwaiter receiver, the operand
expression, and the context the await sits in. Its Correctness twin pins what
Pretty cannot see - copy semantics of struct awaitables and the evaluation
order around the suspension point.
AsyncAwaitPatternsBugs is the spec for the defects, written as the C# that
ought to come out, with the current wrong output named per member. It fails
today; that is the point, and fixing a defect is meant to delete a comment
rather than edit an expectation.
Assisted-by: Claude:claude-opus-5[1m]:Claude Code
A nested designation over tuples, var (x, (a, b)) = t;, is lowered to one
temporary per nested designation - parents before children - followed by the
element reads in depth-first leaf order, and decompiled as a flat
deconstruction plus separate element statements.
The temporaries are now consumed into a tree of tuple nodes before the
conversions and assignments are matched, and the leaves get the same flat
depth-first indices the Deconstruct-call chain hands out, so conversion and
assignment matching runs unchanged. Two properties of the lowered IL shape
the matcher to it: earlier transforms rewrite non-escaping element reads
from ldloca to ldloc, and the temporaries are stack slots whose type is
imprecise, so the container's element type is authoritative and the match
variable is retyped to keep the tuple pattern's invariant.
An element that escapes the deconstruction - used after the statement, so
the pattern cannot consume all its reads - demotes back to a designator leaf
and the match is retried, which restores the flat deconstruction the
escaping read needs. The guard against consuming a pattern piecemeal extends
to the new shape: an element read whose container is stored by an earlier
element read defers to the match starting at that store.
Assisted-by: Claude:claude-opus-5:Claude Code
Element index resolution serves both pattern roots: a registered result of a
Deconstruct call, or an element read of a tuple, which it discovers on first
sight and then owns. In an attempt rooted in a Deconstruct call the tuple
branch must not engage - it overwrites the call's result bookkeeping and
rewires the element read to a fresh variable that the pattern never defines.
The shape that reaches it is a tuple whose element is custom-deconstructed
with discarded leaves, followed by an unrelated assignment: the tuple-rooted
attempt fails, the call-rooted one runs at the element's position, and, now
that an unrelated assignment ends a call pattern instead of rejecting it, the
mixed match is no longer rejected on the way out.
Assisted-by: Claude:claude-opus-5:Claude Code
A nested designation rebinds Deconstruct on the element's static type when
the output is recompiled, while the explicit call it replaces is bound at
the call site. Where a derived element type declares a Deconstruct of the
same arity as the called method, and the source deconstructs through a
base-typed view, the two bindings differ, so the sugared output calls the
wrong method - a divergence the runtime fixture demonstrates on optimized
builds, where copy propagation elides the view.
Nesting is therefore only applied when the method the call binds to is the
one a designation would rebind to; otherwise the call stays explicit, where
its receiver cast preserves the binding.
Assisted-by: Claude:claude-opus-5:Claude Code
A nested designation, var (x, (a, b)) = o;, is lowered to a chain of
Deconstruct calls - the inner call taking the outer call's out-argument as
its target, through a defensive copy where the element is a struct - and
decompiled as a flat deconstruction followed by an explicit Deconstruct
call. The IL pattern node, its invariants and the C# builders already
support nested patterns; only the transform never built them.
MatchDeconstruction now consumes the chain into a tree of match patterns.
The leaves get flat indices in depth-first order, which is the order in
which StatementBuilder and ExpressionBuilder pair pattern variables with
assignments, so the conversion and assignment matching runs unchanged on
top of a nested pattern.
Two matching rules follow from the chain being consumed: a call pattern no
longer needs a matched assignment, because single-use leaves are covered by
the forwarding fixup in MatchAssignments; and a pattern is not rooted on an
element of an enclosing deconstruction, because blocks are processed back to
front, so the inner call is visited first and would otherwise consume the
pattern piecemeal, starving the outer call. That guard runs the enclosing
match as a dry run, which is precise: a barrier statement between the calls
or an element with further uses makes it fail, and the inner deconstruction
is then still transformed on its own.
Assisted-by: Claude:claude-opus-5:Claude Code
Optimized code stores no temporary for a deconstruction element that is
used only once after the deconstruction. MatchAssignments handled that
for trailing elements, but a nested deconstruction copies the inner
element to a temporary, so the elements preceding it are also left
without an assignment; their external load then violated the
DeconstructInstruction invariant that all pattern variable loads are
descendants of the instruction. The forwarding fixup now covers all
unassigned elements and inserts in pattern order, because the statement
and expression builders pair pattern variables with assignments
positionally. This also fixes the nested tuple deconstruction crash
reported in #3388.
Also unwrap the address of the tested operand in
VisitDeconstructInstruction: deconstructing a struct passes the
receiver by reference, which was emitted as an invalid cast,
'var (x, y) = (S)(ref s);', even without nesting.
Fixes#3388.
Assisted-by: Claude:claude-fable-5:Claude Code
The correctness DynamicTests only ran binary + - * / on a dynamic operand, so
no test recompiled and executed decompiled output for any unary operator. That
gap is why ~ on a dynamic value (issue #3820) shipped uncompilable output
undetected: a correctness case round-trips the decompilation through the
compiler, so it fails the moment the decompiler emits something that does not
recompile.
Add ~, -, +, and ! cases. They pin the runtime semantics of the dynamic unary
path and would have caught the OnesComplement regression directly.
Assisted-by: Claude:claude-opus-4-8:Claude Code
After inlining turns a runtime accumulator (e.g. a hand-written GetHashCode prime
chain `h = h * -1521134295 + ...`) into one expression, the leading `SEED * PRIME`
becomes a compile-time constant subexpression. C# always evaluates constant
subexpressions in a checked context, so emitting it bare fails to compile with
CS0220, even though the surrounding context is unchecked.
HandleBinaryNumeric now annotates such an overflowing constant binary operation
with ExplicitUncheckedAnnotation (instead of the implicit UncheckedAnnotation), so
AddCheckedBlocks wraps it in an explicit unchecked(...) - mirroring the existing
handling of overflowing constant n(u)int casts.
Assisted-by: Copilot:claude-opus-4.8:GitHub Copilot CLI
NullPropagationTransform rewrote `c != null ? c.AccessChain : default` to
`c?.AccessChain ?? default` whenever the access-chain result was a non-nullable
value type. For a by-ref-like type (a ref struct such as Span<T>) that form does
not compile: a ref struct cannot be wrapped in Nullable<T> (CS8978). Exclude
by-ref-like return types from the null-coalescing rewrite.
Assisted-by: Copilot:claude-opus-4.8:GitHub Copilot CLI
Compilation uses the .NET builds of the Roslyn toolsets (tasks/netcore*,
bincore csc.dll/vbc.dll launched through the dotnet host). ilasm/ildasm
options use the '-' prefix, which all platforms accept. The dotnet-hosted
compilers have no implicit references or SDK path: net40 compiles pass
mscorlib explicitly, and vbc gets -sdkpath, _MYTYPE=Empty and
-vbruntime:Microsoft.VisualBasic.Core.dll (the facade in the ref packs is
not followed for runtime helpers). The TestRunner gets a self-contained
build for the host platform.
Configurations depending on Windows-only tools or runtimes (legacy
csc/vbc, Roslyn 1.x/2.x, mcs, Force32Bit, executing net40 binaries) are
filtered from the matrix off-Windows via Tester.SupportedOnCurrentPlatform
or gated with [Platform("Win")]. PdbGen comparisons normalize document
name separators, and Correctness/Async uses Console.IsInputRedirected
instead of the Windows-only Console.CapsLock.
Assisted-by: Claude:claude-fable-5:Claude Code
This makes our logic more similar to that used by the dotnet runtime. This lets us infer correct stack types in edge cases such as #2401. It also improves support for obfuscated control flow such as #2878.
Siegfried Pammer
Daniel Grunwald
Christoph Wille
Sebastien Lebreton
Peter Crabtree
ElektroKill
Zachary Northrup
dotnet format