Unqualified property access inside a method body fails

• Type: bug
• Status: backlog
• Owner: —
• Track: A (compiler — name resolution)
• Found / Opened: 2026-06-30, while adding TThread.Terminated (multithreading M3)

Symptom

Inside a method, a bare (unqualified) reference to one of the class's own properties fails with undefined variable (<name>). Bare field access in the same position works, and Self.<property> works — only the unqualified property read/write is unresolved.

type TC = class
  private FX: Boolean;
  public property Done: Boolean read FX;
  procedure Go;
end;
procedure TC.Go;
begin
  if FX        then ... ;   { ok — bare field }
  if Self.Done then ... ;   { ok — qualified property }
  if Done      then ... ;   { ERROR: undefined variable (Done) }
end;

This breaks idiomatic FPC code, most visibly while not Terminated do in a TThread.Execute override (worked around with Self.Terminated in lib/rtl/palthreadobj.pas + test/test_tthread_terminate.pas).

Root cause (located)

compiler/parser.inc ~1169–1184: unqualified identifier resolution in a method, when not a local/global, does FindUField(selfClassCi, name) and synthesises an implicit-Self AN_FIELD. There is no parallel FindUProp fallback, so a name that is a property (not a field) falls through to "undefined variable".

Fix sketch

After the FindUField miss in that block (and the WITH-stack block at ~1135), also try FindUProp(selfClassCi, name) (symtab.inc:417). On a hit, synthesise the same access the working Self.<prop> postfix path produces — i.e. an implicit-Self base node + the property's read accessor (field-backed: read field; method-backed: call read method). Must cover read (factor context) and write (assignment LHS), and not regress indexed/default properties. Gate: self-host byte-identical + a focused test (bare read, bare write, method-backed read, field-backed read).

Impact

Quality-of-life / FPC compatibility for all classes, not threading-specific. Workaround (Self.) is available everywhere, so not urgent — but it is a real papercut for ported FPC code.

RESOLVED 2026-06-30 (parser.inc ParseLValueAST)

Added a FindUProp fallback beside the bare-FindUField path in ParseLValueAST (used for both read and assignment-LHS contexts). Behaviour now:

• bare property READ (field-backed) -> implicit Self + AN_FIELD on the backing read field (source span = the backing field, so codegen loads the right width — an early cut pointed the span at the property name and produced wrong-width Boolean loads: not Done came out TRUE);
• bare property READ (method-backed) -> implicit-Self getter call (virtual-aware);
• bare property WRITE (field-backed) -> AN_FIELD on the backing WRITE field (uses the write accessor, not the read one);
• bare property WRITE (method-backed setter) -> CLEAN compile-error pointing to Self.<prop> := ... (building the setter call from the lvalue path is too entangled — left as the one residual case);
• indexed/default properties unchanged (still need Self.<prop>).

Self-host byte-identical. Test: test/test_bare_property.pas (in make test-core) + test/test_tthread_terminate.pas now uses the bare FPC idiom while not Terminated

• ReturnValue := 42. Moving to done/ (residual = method-backed bare write, by design).