REJECTED 2026-08-30 (frankP) — DUPLICATE, filed before I read the decided/ queue

Superseded by bug-p-object-value-types-standard-meaning [P p70, working/, owner frank-user], which the owner decided in decided/decide-revisit-object-types-rtl-generics-fired-the-trigger hours before I filed this. Same wall, same line, same message. Do not work this ticket; that one is a live lock.

I filed it off a measurement without checking decided/ or working/ for the same subject first — the wall reproduced, so the finding was real, but the ticket was not new. A duplicate is the invisible-work problem in the ticket system itself: two entries that each look individually reasonable, and the reader who greps for the wall now has to work out which is authoritative.

One fact from my measurement that the live ticket does not carry, offered here rather than edited into another lane's working/ ticket: the wall reproduces identically on pinned and on HEAD — same line, same message — so it is long-standing and NOT a regression from the generics work of 2026-08-29/30. That distinction is worth having before anyone bisects for it.

The reproduction command, for whoever wants it:

$ ./compiler/pascal26 -Fu<rtl-generics/src> coll.pas   # {$mode delphi}; uses Generics.Collections
$ stable_linux_amd64/default/pinned -Fu<same> coll.pas

Both: generics.collections.pas:146: generic templates must be class, record, interface, array or procedure declarations.

A generic template cannot be an `object` type

Repro

program g;
{$mode delphi}
type
  TBox<T> = object
    V: T;
  end;
var
  b: TBox<Integer>;
begin
  b.V := 7;
  writeln(b.V);
end.

pascal26:N: error: generic templates must be class, record, interface, array or procedure declarations

FPC compiles it. object is the old Turbo Pascal value-semantics class: a record with methods and inheritance, no reference semantics, no automatic lifetime. It is not deprecated in FPC and rtl-generics uses it deliberately, for exactly the reason it exists — a collection helper that must not allocate.

Where it was found, and the measurement that matters

The current stop for uses Generics.Collections:

pascal26:146: error: generic templates must be class, record, interface, array or procedure declarations
  in: .../rtl-generics/src/generics.collections.pas
  near:  T  PT   >>> object strict private

  TCustomPointersCollection<T, PT> = object
  strict private type
    TLocalEnumerable = TEnumerable<T>;
  protected
    function Enumerable: TLocalEnumerable; inline;

Measured on HEAD and on pinned, same command, same line, same message — so this is a long-standing gap and not a regression from any of the generics work of 2026-08-29/30. It is recorded that way deliberately: the nested-type ticket's summary carried a stale "the stop is line 120" figure that no longer reproduced against either binary, and an unverified corpus figure is worth less than none.

Shape of the fix, as far as it was looked at

The rejection is a kind-check on what follows = in a template declaration; the list already admits class / record / interface / array / procedure. object is closest to record in every respect that matters to the template machinery (value semantics, a body terminated by end, nested type sections), so the question is whether it can join the record arm outright or needs its own — which is a question about inheritance (object can have a parent, records cannot) and about the class-vs-record split in the specialization streamer, not about the kind check itself. Nobody has traced it; do not treat this paragraph as a design.

Note the second line of the corpus snippet: strict private type — a visibility-qualified nested type section inside an object. Whoever takes this should check that path is reachable before assuming the kind check is the whole job.

Gate