Scalar value properties that merely forwarded to a private backing field
(IsAsync, Operator and the operator-type enums, Format, ClassType, Variance,
ParameterModifier, the parameter bool flags, and similar) become
auto-properties, formatted on a single line. Behavior is identical;
PrimitiveExpression's constructor sets the property instead of the dropped
field.
Assisted-by: Claude:claude-opus-4-8:Claude Code
Each concrete node's constructors are now emitted by the source generator
from its members in source order: single and collection [Slot] children,
the [NameSlot] string, and settable enum scalars (Operator, FieldDirection,
...). It emits the empty ctor (for object initializers), a required-prefix
ctor, one ending at each collection, and the full ctor, with later ctors
forwarding to shorter ones via this(...) and a params[] overload alongside
each IEnumerable<T> one. The hand-written ctors the generator now produces
are removed; scalar/location/Identifier convenience overloads that it cannot
express are kept (e.g. AssignmentExpression(left, right), SimpleType(Identifier),
the string+TextLocation overloads).
Because ctor parameters follow source order, BinaryOperatorExpression and
AssignmentExpression declare Operator between Left and Right so the generated
ctor is the expected (left, op, right). Pure-scalar nodes whose state is not
in slots (e.g. PrimitiveExpression's literal value) are left untouched.
Assisted-by: Claude:claude-opus-4-8:Claude Code
TokenRole was a printer-side descriptor whose only jobs were holding a token's
text and giving the writers an identity to single out specific tokens. The text
becomes plain const strings on the nodes, and the few identity checks are
reexpressed as node-stack context: interpolation braces are recognized by an
Interpolation on the writer's stack, record class versus struct coloring keys
off TypeDeclaration.ClassType, and the accessor/this/base/override cases fall
out of the surrounding node. WriteKeyword/WriteToken drop the descriptor
parameter. The constants are named for what the token is: a keyword, a symbol
token, or a modifier.
Assisted-by: Claude:claude-opus-4-8:Claude Code
Slot identity already lived in node.Slot/CSharpSlotInfo/SlotKind after
the storage flip, so the parallel Role/Role<T> child model was redundant.
The Role-keyed mutation/query API is reexpressed over SlotKind, and the
role-index packing on AstNode flags is gone.
Because a node's Slot is now derived from its index in its parent rather
than stored on the child, the located-AST reattach in
InsertMissingTokensDecorator must capture the child's slot kind before
Remove() detaches it.
Assisted-by: Claude:claude-opus-4-8:Claude Code
With every optional slot nullable, the null-object pattern is dead. Generated
non-nullable getters return the backing field directly, which surfaced a last
tier of slots the decompiler legitimately leaves empty (omitted range operands,
an implicitly-typed array creation, unnamed parameters, an unbound generic
argument, and others) and flips them to nullable too. The machinery is then
removed entirely: the per-node null classes, the .Null statics and
VisitNullNode, AstNode.IsNull, the role null object, and Identifier.Null.
AcceptVisitor becomes unconditionally generated, and consumers move from
.IsNull to is null and from unconditional visits to ?.AcceptVisitor.
Assisted-by: Claude:claude-opus-4-8:Claude Code
Optional single-child slots return T? with a real null instead of a role
null-object, taking the C# grammar as the oracle for which slots are optional.
The generator emits the property type as T? and matches it with MatchOptional,
and consumers move from .IsNull to is null / ?.. This covers the optional
statement, member, try-catch, creation-initializer and pattern slots and the
optional NameSlot tokens. A few slots the grammar marks required but the
decompiler legitimately leaves empty (the implicit-element-access target, an
implicitly-typed lambda parameter's type) are flipped to nullable as well.
Assisted-by: Claude:claude-opus-4-8:Claude Code
A [NameSlot("role")] partial string property makes the generator own the
backing Identifier token slot, the string accessor, and the match term, so a
convenience name string and its hand-written token slot collapse to a single
declaration. A nullOnEmpty option stores a null token for an empty name, used
where the output visitor keys off an absent token. Apply it across every
name-token node.
Assisted-by: Claude:claude-opus-4-8:Claude Code
NodeType was NRefactory's coarse node category, but only three reads remained
here: two checks now expressed as "is not Trivia" and one debug assert for the
pattern category. Remove the enum, the abstract property, every per-node
override, and the generator's emission, preserving the pattern-placeholder case
through an IPatternPlaceholder marker interface the output-visitor assert
checks. Also remove the unused PrimitiveExpression.AdvanceLocation helper.
Assisted-by: Claude:claude-opus-4-8:Claude Code
Delete source-fidelity state the decompiler never reads, since it generates
the AST and never parses: SyntaxTree.FileName/ConditionalSymbols/TopExpression,
Comment.StartsLine/IsDocumentation, the preprocessor line/file fields, and a
computed ComposedType flag. Hoist the duplicated start/end location storage
into the shared Trivia base, and drop the redundant per-instance Location field
on single-token leaf nodes in favor of the base's print-time location.
Generate DoMatch for the last nodes (Comment, SyntaxTree, the using and
namespace declarations, Identifier).
Assisted-by: Claude:claude-opus-4-8:Claude Code
Generate DoMatch across the remaining expression, statement, type-member,
type-reference and general-scope nodes, including the inherited
EntityDeclaration name/return-type/attribute match, and stop generating it for
abstract base nodes. Matching every structural member fixes real
under-matching bugs (PointerReferenceExpression ignored its Target;
ExtensionDeclaration matched anything) while computed or derived members are
excluded from matching. PrimitiveExpression and PreProcessorDirective stay
hand-written by design. Pretty output is byte-identical.
Assisted-by: Claude:claude-opus-4-8:Claude Code
The grammar-production doc comments were transcribed in ANTLR style and
often copied the C# spec verbatim, including sub-rules the AST does not
model. Rewrite them as W3C EBNF (::=, ?, *, +), unroll sub-productions
that are not themselves AST nodes (e.g. anonymous_function_modifier ->
'async'), and shape each production to the node's actual members. Render
top-level alternations of node productions as multi-line <code> blocks;
keep operator and keyword token lists inline.
Assisted-by: Claude:claude-opus-4-8:Claude Code
Replace the hand-written DoMatch overrides on the expression nodes with
generated ones. The generator matches every real child and structural member,
so several matchers become stricter than the hand-written versions that
under-matched (for example a previously-skipped child or an ignored flag),
while Pretty output is unaffected. Convenience-string identifier slots are
excluded so the name string is matched once rather than also via its token.
Assisted-by: Claude:claude-opus-4-8:Claude Code
Preparation for making optional single-value slots nullable: a role may omit
its null object and a slot may be cleared via SetChildByRole(role, null), and
the generated DoMatch emits MatchOptional for a nullable single child so an
absent child matches correctly. Both are inert until a slot is actually made
nullable.
Assisted-by: Claude:claude-opus-4-8:Claude Code
Each concrete syntax node now carries, in an XML-doc <remarks> block, the
matching production from the C# language specification grammar (ECMA/Microsoft,
ANTLR notation), quoted verbatim. Aggregate nodes (e.g. BinaryOperatorExpression,
ComposedType, TypeDeclaration) list every production they span; lexical/trivia
nodes (Comment, the preprocessor directives, Identifier) cite the lexical rule.
Nodes with no spec production -- ErrorExpression, UndocumentedExpression,
InvocationAstType, TypeReferenceExpression, NamedExpression, DocumentationReference,
and the C# 14 ExtensionDeclaration -- carry a hand-written EBNF plus a note
explaining why no official production exists.
Assisted-by: Claude:claude-opus-4-8:Claude Code
Turn on #nullable enable across the AST transform pipeline, ahead of
annotating the slot properties themselves. TransformContext now exposes the
nullable CurrentMember/CurrentTypeDefinition/CurrentModule contract already
declared by ITypeResolveContext, and the generated pattern-to-node conversion
returns a non-null node so patterns can be used in collection initializers
without warnings. No IL changes.
Assisted-by: Claude:claude-opus-4-8:Claude Code
The token writers and the UI syntax highlighter only need a token's identity (to
single out structural braces, the constructor this/base keyword, the override
modifier, and to colour keywords) -- not the AST child-role machinery. Make
TokenRole a standalone printer-side descriptor instead of a Role, turn the
modifier marker into a TokenRole, and change the WriteKeyword/WriteToken
signatures from Role to TokenRole across the writer hierarchy and the highlighter.
This lets the child Role hierarchy be removed without disturbing token output.
The dead OptionalComma/OptionalSemicolon bodies (no comma/semicolon/whitespace
children exist since the token drop) become no-ops, and the few sites that handed
the writer a child role for a keyword now pass none. Output is unchanged across
the decompiler suite, and the UI builds.
Assisted-by: Claude:claude-opus-4-8:Claude Code
Introduce the successor to node.Role for child-slot identity: the generator
emits a CSharpSlotInfo per [Slot], exposed as node.Slot, plus a shared SlotKind
enum for the polymorphic "is this node in an embedded-statement / condition /
base-type slot?" comparisons a per-node identity cannot express. Migrate the
printer and transform position checks from node.Role to node.Slot and
node.Slot.Kind, and read identifier children and role-keyed writes through the
typed properties. Role is still present and is removed later; output is
unchanged.
Assisted-by: Claude:claude-opus-4-8:Claude Code
The pattern matcher walked collections through INode.Role/FirstChild/
NextSibling, skipping siblings of a different role. Now that each
AstNodeCollection<T> is already the per-role child list, the engine matches two
collections by list index, and INode sheds Role/FirstChild/NextSibling
entirely. A collection exposes its IReadOnlyList<INode> view through a cached
adapter rather than implementing the interface directly, so a typed collection
does not become ambiguous for LINQ. Characterization tests pin the matcher's
behavior first.
Assisted-by: Claude:claude-opus-4-8:Claude Code
Children were kept in a per-node doubly-linked list with the slot accessors
layered over it as a view. Storage now is the slot model: each node stores its
children in generated backing fields, AstNodeCollection<T> is backed by a
List<T>, and the flattened child-index space is owned by generated
GetChildCount/GetChild/SetChild/GetChildSlot members, with sibling navigation,
the role API and Clone re-expressed over them and indices renumbered lazily. A
DEBUG CheckInvariant runs after each transform, the analog of the IL
pipeline's per-transform check, so a transform that corrupts the tree fails at
that transform. Output is unchanged.
Assisted-by: Claude:claude-opus-4-8:Claude Code
Comments and preprocessor directives were positional children interleaved
into the child list, and punctuation, keywords and operators were token-node
children. Add a leading/trailing trivia side-channel for comments and
directives, emit it from the output visitor, and re-home every comment
receiver onto it (including inside-block comments as comment-only empty
statements and undecodable attribute arguments as an ErrorExpression). With
locations and sequence points no longer sourced from token nodes, stop
reconstructing them on the locations path and delete CSharpTokenNode,
CSharpModifierToken and InsertSpecialsDecorator. The AST no longer carries
token children or positional comments; output is byte-identical.
Assisted-by: Claude:claude-opus-4-8:Claude Code
Source locations were virtual, computed by recursing to the first and last
child, whose leftmost and rightmost leaves are token nodes; sequence-point
coordinates likewise came from reconstructed token nodes. Store locations as
fields assigned while printing, and derive sequence-point coordinates from the
surrounding real nodes plus the decompiler's fixed formatting, so neither
depends on token children. The using/foreach await modifier becomes a plain
bool field. Characterization gates lock the emitted locations and PDB
coordinates, which are unchanged.
Assisted-by: Claude:claude-opus-4-8:Claude Code
Member and local modifiers were stored as modifier token children, and a
ComposedType's ref/readonly/nullable/pointer specifiers and an array rank as
token and comma children. The output visitor already derived all of these
from scalar accessors, so move them to plain enum/bool/int fields. This
removes another dependency on token children ahead of deleting the token
nodes; the emitted keyword and specifier sequences are unchanged.
Assisted-by: Claude:claude-opus-4-8:Claude Code
Each child of a C# AST node is declared as a [Slot] partial property, and the
source generator emits the accessor bodies and an ordered slot schema
(SlotCount/GetSlotRole/IsCollectionSlot) from them. Generating the schema
keeps slot order from being mis-stated by hand and lets a DEBUG invariant
check declared slot order against document order on every decompile. The node
hierarchy is converted family by family; the EntityDeclaration leaves flatten
their inherited Attributes/ReturnType/NameToken into each leaf's ordered slot
set. Storage stays the NRefactory linked list at this stage, so only the
declaration model changes and output is unchanged.
Assisted-by: Claude:claude-opus-4-8:Claude Code
The generator emits the IAstVisitor interface, the AcceptVisitor overloads,
and the null-node and pattern-placeholder nodes from [DecompilerAstNode]
declarations, so drop the hand-written equivalents across the C# AST: per-node
AcceptVisitor/DoMatch, the #region Null / #region PatternPlaceholder blocks,
IAstVisitor.cs, and now-dead usings. Also adds AccessorKind and moves
IdentifierExpressionBackreference into the PatternMatching folder.
Assisted-by: Claude:claude-opus-4-8:Claude Code
Introduce a Roslyn source generator that emits the visitor boilerplate for
the C# AST from [DecompilerAstNode]-tagged node declarations: the
IAstVisitor interface, the AcceptVisitor overloads, the pattern-placeholder
nodes, and the initial DoMatch support. AccessorKind lets an accessor's
keyword be chosen independently of its role, an early step toward shedding
the NRefactory role model.
The C# AST inherited NRefactory's freezable model (IFreezable, Freeze,
IsFrozen, a frozen flag bit, and ThrowIfFrozen guards on every mutator),
but the decompiler never uses it: nothing calls Freeze(), not even the
generated null-node singletons, so every IsFrozen guard only ever
evaluated false. The decompiler is single-threaded and never shares or
freezes nodes. Remove the whole apparatus as preparation for the
slot-based AST rewrite, which has no place for it. Roles are untouched
here, so the flags word keeps its role index; only the freed frozen bit
goes away.
The "IL with C#" view decompiles each method body as a bare handle, so a
static constructor is decompiled without its type's field declarations in
the syntax tree. MoveFieldInitializersToDeclarations then could not find a
declaration to move the static-field-initializer statement onto, asserted
(kind was Static, not Primary) and dropped the statement -- crashing Debug
builds and silently losing the assignment in Release.
Dropping the statement is only correct for the primary-constructor case,
where the assignment's backing member is synthesized and has no separate
declaration. For static/instance initializers a missing declaration just
means the member is not part of this partial syntax tree, so the
assignment must remain in the constructor body.
Assisted-by: Claude:claude-opus-4-8:Claude Code
Runtime async is a compiler feature that emits ordinary async/await (a
C# 5 construct), so reconstructing it should not require selecting C# 15.
The dedicated RuntimeAsync setting was also redundant: AsyncAwaitDecompiler
already runs the runtime-async transforms only when AsyncAwait is enabled.
Fold the behavior into the AsyncAwait setting and drop the separate toggle.
Assisted-by: Claude:claude-opus-4-8:Claude Code
On FIPS-mode systems the platform crypto provider refuses to create
SHA-1 instances (OpenSSL: error:03000098 invalid digest), so merely
displaying a strong-named assembly's identity failed. The public-key
token is a non-secret identity hash whose algorithm is fixed by
ECMA-335, so the two token sites now use dotnet/runtime's managed
Sha1ForNonSecretPurposes, vendored with its license header intact and
shielded from the repo formatter via generated_code in .editorconfig
so future upstream syncs diff cleanly. IncrementalHash was considered
and rejected: like SHA1.Create(), it resolves the digest through the
host crypto policy, and Roslyn's equivalent token code also relies on
the platform SHA-1, so it offers no precedent for FIPS safety.
Assisted-by: Claude:claude-fable-5:Claude Code
Windows maps CON, PRN, AUX, NUL, COM1-9 and LPT1-9 to devices -- on many
builds even with an extension appended, so a type named Con made both
whole-project export and the save dialog fail with IOException '\\.\Con'.
CleanUpName only checked for reserved names after re-appending the file
extension, where they never match, and the save-dialog default-name
helpers did not check them at all. The escape appends the underscore to
the base name (con_.txt, not con.txt_) because device-name parsing
ignores everything after the first dot, and is applied per path segment
so reserved directory names produced by namespaces are covered too. The
ILSpy.Tests.Windows fixture verifies on a real Windows filesystem that
the escaped names are creatable.
Assisted-by: Claude:claude-fable-5:Claude Code
The WPF app showed the disassembled IL header when hovering a member
reference in IL view, and the opcode's XML documentation when hovering
an instruction. The Avalonia port lost both: ILLanguage fell back to
the base ambience one-liner and the opcode tooltip carried only the
name and encoding. Opcode docs additionally need a modern-.NET source:
the WPF code read the .NET Framework reference-assembly docs, which do
not exist on modern .NET, so MscorlibDocumentation now falls back to
the ref pack parallel to the hosting runtime.
Assisted-by: Claude:claude-fable-5:Claude Code
Hosts without a .NET Framework installation (e.g. Linux and macOS) have
no GAC; the only system-wide assembly store there is the shared-framework
directory of the runtime executing the decompiler, and
UniversalAssemblyResolver only consulted it through the version <= 4.0
legacy fallback. This made e.g. the type-forwards of a netstandard facade
(pointing to a versioned System.Runtime) unresolvable, which left
well-known types like Nullable<T> without a definition and among other
things misaligned nullability decoding (Nullable<T> occupies no slot in
the nullable metadata, so it must be recognized).
On Windows nothing the GAC answered changes; the new fallback only adds
resolutions that previously failed outright.
Assisted-by: Claude:claude-fable-5:Claude Code
The decompiled view only offered navigation on member names; finding
what an override actually overrides required opening the analyzer.
Attaching a reference to the modifier token gives the same
go-to-definition affordance Visual Studio has on 'override'. The
reference resolves via InheritanceHelper.GetBaseMember, so it targets
the nearest overridden member and degrades to plain text when the base
member cannot be resolved.
Assisted-by: Claude:claude-fable-5:Claude Code
When enabled, switch sections are ordered by their case label value
instead of by the underlying branch's IL offset. Default is false to
keep existing output unchanged. Useful when diffing decompiler output
across rebuilds of obfuscated assemblies, where IL block layout is
unstable but the case-to-value mapping is not.
Includes an ILPretty test that exercises a hand-written switch whose
table targets are placed at non-monotonic IL offsets (simulating
obfuscator block shuffling) and verifies the cases come out in
label-value order with the setting enabled. Also adds the
Resources.resx / Resources.Designer.cs entry so the WPF settings UI
shows a proper label instead of the raw key.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Fix anonymous-type lambda early-return emitting unresolvable cast
When a lambda's inferred return type contains an anonymous type and one
branch returns null, the decompiler emitted an explicit cast such as
`return (IEnumerable<<>f__AnonymousType0<int>>)null;`, which is invalid C#.
Skip the cast in IsPossibleLossOfTypeInformation for null literals whenever
the expected type contains an anonymous type:
null is implicitly convertible to any reference type, so no cast is needed,
and the anonymous type has no nameable form to cast to anyway.
Fixes#3751
ICSharpCode.Decompiler / ILSpyX / BamlDecompiler set RestoreLockedMode but
declared no RuntimeIdentifiers, so their lock files were RID-agnostic. A
RID-specific 'dotnet publish -r <rid>' of ILSpy (which declares all four
distribution RIDs) propagated the RID to these locked project references and
failed restore with NU1004 unless --no-restore or --force-evaluate was used.
Declare the same win-x64;win-arm64;linux-x64;osx-arm64 set on the three
libraries and regenerate the locks so a locked-mode publish works for every
platform out of the box.
Assisted-by: Claude:claude-opus-4-8:Claude Code
Replace the Windows-only .bat helpers (clean / debugbuild / releasebuild /
restore / updatedeps and BuildTools/format) with cross-platform pwsh
scripts at the repo root: restore.ps1, build.ps1 (-Configuration), clean.ps1,
updatedeps.ps1 and BuildTools/format.ps1, alongside the existing publish.ps1.
Enable a packages.lock.json for every project by hoisting
RestorePackagesWithLockFile into the root Directory.Build.props (the four
core libraries set it individually before) and commit the generated locks,
so restores are repeatable and CI can cache packages off them.
Cache the NuGet packages folder in the three setup-dotnet workflows
(build-ilspy, build-frontends, codeql-analysis), keyed on the lock files
per the setup-dotnet caching guidance.
Scope the Debug "Verify package contents" check to the *.filelist outputs
it actually generates. A project's packages.lock.json is keyed only by
(framework, RID), with no host-OS axis, so a lock produced on Linux
legitimately differs from one produced on Windows whenever an OS-conditional
PackageReference applies (Debug+Windows pulls Microsoft.DiaSymReader*). The
Windows restore then rewrites those locks; that churn must not fail a step
whose job is to police the VSIX/MSI file lists.
Also drop the dead ILSpy.BamlDecompiler publish line from
publishlocaldev.ps1, mirroring the earlier publish.ps1 fix.
Assisted-by: Claude:claude-opus-4-8:Claude Code
XmlDocLoader's modern-.NET fallback built the ref-pack path from the EXACT runtime
patch version (shared/Microsoft.NETCore.App/10.0.8 -> packs/Microsoft.NETCore.App.Ref/
10.0.8/ref), but the targeting (ref) pack version usually differs from the installed
runtime patch -- e.g. pack 10.0.0 -- and a relocated CI install (DOTNET_INSTALL_DIR)
may carry only a different feature band. The exact-version lookup then missed and the
provider came back null, leaving CoreLib hovers undocumented. Broaden the search:
prefer an exact match, then the newest pack sharing the runtime's major.minor, then
the newest pack present.
Assisted-by: Claude:claude-opus-4-8:Claude Code
The portable-PDB writer carried its knobs (no-logo, pdb id, progress,
progress title) as a growing list of optional WritePdb parameters.
Turn the type into a configured instance whose options are properties,
and add EmbedSourceFiles (default true): when a PDB is generated next
to a project export whose .cs are already on disk, embedding the source
again is redundant, so the caller can turn it off. The per-document
checksum/hash is computed either way, so documents still resolve.
Assisted-by: Claude:claude-opus-4-8:Claude Code
DecompilerVersionInfo.template.cs is the source-of-truth for the
runtime version constants -- update-assemblyinfo.ps1 rewrites the
generated DecompilerVersionInfo.cs from it on every build, picking
up the live git revision and commit hash for the $INSERT...$ slots.
Assisted-by: Claude:claude-opus-4-7:Claude Code
Ports the WPF assemblyList_CollectionChanged history-prune from
ILSpy/AssemblyTree/AssemblyTreeModel.cs to the Avalonia
DockWorkspace.OnAssemblyListChanged handler. NavigationHistory<T>
already exposed the RemoveAll(Predicate<T>) primitive; the caller
wiring was the missing piece. Without it, a tree-row click after
removing an assembly walked through NavigationEntry.DisplayText
on a stale TreeNodeEntry, which hit MemberReferenceTreeNode.Signature
-> Language.EntityToString -> ILAmbience.ConvertSymbol and NRE'd on
a now-null ParentModule.
Assisted-by: Claude:claude-opus-4-7:Claude Code
Promotes the modern .NET XML-doc lookup from the Avalonia port into the
shared ICSharpCode.Decompiler library so every host (WPF, Avalonia, any
third-party consumer of XmlDocLoader) gets hover/tooltip documentation
for system entities without per-host fallback wiring.
After the cluster-1/3/4 fixes converged every caller on the same matching
shape (match the slot/kind/type of a reference ILVariable, then check the
init value), the `Predicate<StLoc>` parameter was just a hole through which
each caller restated that logic verbatim. Fold the slot/kind/type check
into the helper and have callers pass just the reference variable and a
value matcher.
The multi-handler matcher only recognized a switch-instruction dispatch — but
when a try-catch has just two handlers (or a handful with non-consecutive K
values), Roslyn emits an if-chain instead:
if (num == K_1) br case_K_1; br nextBlock
; nextBlock { if (num == K_2) br case_K_2; <leave outer | br end> }
Add a parallel matcher that walks the if-chain and collects (K, case-block)
pairs the same way MatchSwitchDispatch does, plus the terminating leave/branch
as the default exit. Call it as a fallback when the switch matcher rejects.
Also clone the default-exit before re-adding it to the continuation block —
in the if-chain shape it's a child of a *different* block (a later step in
the chain), not the now-cleared switch instruction, so the in-place re-add
relied on the switch's release cascade and didn't generalize.
Closes Cluster 2 from #3745.
The flag-based early-return rewriter was tied to one specific lowered shape:
the try body's flag-setter had to be exactly `stloc flag(K); leave try`, the
post-try check had to be a `br checkBlock` (not an inline `IfInstruction`), and
the early path had to be a direct Leave or a forward to a one-instruction
leave-block whose target was the function body. None of those hold for
`try { try { return X; } finally { await ... } } finally { await ... }`:
- The inner flag-setter has a leading capture-forwarding store
(`stloc capture(X); stloc innerFlag(K); leave inner-try`).
- The inner check-block's early path branches to a multi-instruction helper
that sets the *outer* flag and leaves the outer try, instead of being a
direct return.
- SplitVariables hands out a separate ILVariable for the pre-init flag store
when the in-handler store is in a disjoint dataflow region.
Rebuild the matcher around the idea of a "template" — the chain of stores
the early path performs before its terminating Leave. Each flag-setter then
becomes its own prefix stores + a clone of the template, which collapses the
inner-then-outer flag chain in two passes (inner first, outer second, because
descendant order visits the inner TryFinally first). Also extend the
flag-setter scan to walk the whole try-block's descendants — after the inner
rewrite, the inner's spliced flag-setter lives inside the inner-try container
but still leaves outwards to the outer try, so it's an outer flag-setter from
the outer's perspective.
Add a `RUNTIMEASYNC` preprocessor symbol (defined when `EnableRuntimeAsync`
is set) and gate the new return-from-try-finally fixtures on it — the
state-machine async pipeline doesn't recover this shape, so it would expand
the same source into the `int result; try { ...; result = X; } finally { ... }
return result;` verbose form and the Async (state-machine) pretty test would
regress.
Closes Cluster 1 (1.1, 1.3) from #3745. Cluster 1.2 (void `return;` at the
end of a try-finally body) and 1.4 (break/continue across a try-finally) are
left for a follow-up: both round-trip semantically equivalently but the AST
emitter drops a trailing void `return;` and the break/continue lowering uses
a switch dispatch that the current single-K matcher can't recognize.
`try { throw new ...(); } finally { await ... }` lowers to a try whose only
exit is the throw (handled by the synthetic catch). The existing matcher
required at least one outward Branch to the continuation, which is too strict
— a throw-only try body produces zero outward branches but is still a valid
lowered shape. Two follow-on fixes were also needed:
- The pre-init's ILVariable diverges from the in-handler store after
SplitVariables when the try body has no path that reaches the dispatch's
load without going through the catch; match the flag init by slot/kind/type
instead of identity (same workaround the multi-handler matcher uses).
- With a throw-only try body the new TryFinally has unreachable endpoint,
so appending the no-exception successor after it would put a non-final
unreachable-endpoint instruction in the parent block. Skip the append in
that case — the parent block's endpoint is already correctly unreachable.
Closes Cluster 4 from #3745.
The single-handler try-catch matcher was tied to the top-level shape: it
required the try-catch be the last instruction in its parent block, that the
post-catch "no exception" path be a direct Leave that exits the function, and
that the flag-init's ILVariable be identical to the in-handler flag store.
None of those hold for an inner try-catch sitting inside an outer try-finally
where both await — the inner is followed by a `br continuation`, the no-exception
path leaves the outer try-block (not the function), and SplitVariables hands
out a separate ILVariable for the pre-init store.
Drop the "must be last instruction" gate, accept Leave-to-any-ancestor and
cross-container Branch as the no-exception exit (extracted into a new
`IsContainerExit` helper), and match the flag-init by slot/kind/type the same
way the multi-handler matcher already does.
Closes Cluster 3 from #3745.
When a return crosses an enclosing try-finally with await, runtime-async lowers it as: capture the return value, set an int flag to a unique non-zero value, leave the try block normally so the finally runs, then post-finally check "if (flag == K) return capture;". Detect that pattern after my outer try-finally rewrite (or, in optimized builds, the compiler-emitted TryFinally directly) and replace each capture-flag-and-leave site with a direct "leave outer (capture)" — the leave still passes through the TryFinally, so the user's finally body executes before the function returns, which matches the source-level semantics.
Handles both the "if (flag == K)" and "if (flag != K)" check forms (the optimizer emits the latter). Closes the last gap in Issue2436 — RuntimeAsync now passes both Optimize and non-Optimize modes; the full RuntimeAsync* sweep is 12/12 green.
Also remap reads of the captured-obj local inside the cleaned filter so optimized builds (where Roslyn inlines the typed-cast directly into the user filter expression instead of stashing it in a local) render against the catch variable rather than against "((T)obj)".
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Siegfried Pammer
danbopes
Jakob Hellermann
Christoph Wille