← board

The measurement

program p; {$mode delphi}
type TC = class
     private FF: Integer;
     published property F: Integer read FF write FF stored False;
     end;
begin end.



pascal26: error: expected ':' before 'False'

FPC accepts this. The clause also takes a field or a parameterless boolean method (stored FIsStored, stored GetStored), not only a constant.

Why it is ranked here rather than lower

stored is not decoration: it is what a streaming system asks before writing a property out, so it is load-bearing for anything that persists a component. We ship lib/pcl and test_pcl_lfm reads an .lfm. A form that round-trips through us today cannot express "do not persist this property".

How it was found, which is the part worth keeping

Dispositioning tclass14a, a %FAIL row whose own comment says "class properties are not for sreaming therefore 'stored' is not supported". pxx refuses the row, so it counted as a pass — for the wrong reason. The discriminator was probing the construct where it is LEGAL: an ordinary published property. That is refused too, so the row was never testing its own subject.

task-t-twelve-syntax-shaped-fail-rows-may-be-refused-by-a-parse-gap-rather-than-their-own-subject

Resolution

Landed with [[bug-p-a-property-default-value-clause-is-read-as-the-default-indexed-property-marker]] in one commit, because they are the same arm: stored, default <value>, default;, nodefault and the hint directives are now one ParsePropertyTailDirectives in compiler/pasparser_decl.inc, called from both the ordinary and the REDECLARATION branch of ParsePropertyDecl — which had written the default arm and the hint loop out twice and carried the same bug in both copies. Splitting default into its two clauses without also taking stored would have left a second procedure to write.

stored takes ONE operand — a constant, a field name, or a parameterless boolean method name — parsed and DISCARDED. pxx has no RTTI streaming, so it has no consumer today; what was load-bearing was that its presence made the whole declaration unparseable. stored with no operand (stored;) is refused with "stored" needs a constant, a field name or a parameterless boolean method rather than silently accepted, so the clause cannot become a no-op by accident.

Rows E..H of test/test_a_property_default_clause_is_two_clauses.pas cover stored False, stored <method>, and stored and default on one property, against fpc -Mdelphi 3.2.2's own output.

One row of the ORIGINAL probe was dropped rather than fixed

stored FF — an Integer FIELD — is accepted by pxx and refused by fpc ("The type of the storage symbol must be boolean"). Us accepting what fpc rejects is not a defect, so the row came out of the test rather than the compiler growing a type check for a value it discards.

The residual, and it has an owner

stored now parses on a CLASS property, where fpc refuses it. That is the same "we accept what fpc rejects" rule and would normally end here — except frankS found the sibling that makes it an INTERNAL inconsistency rather than an FPC-parity question: tclass14b is a published class property, and pxx already refuses that with a class property cannot be published (pasparser_decl.inc:6735). published and stored are two spellings of one rule — a class property has no instance, so nothing about streaming applies to it — and we implement it for one spelling only. Narrowing follows in its own commit, worded to match the existing arm, because a refusal's risk is refusing working code and it gets its own gate.

It retires a false pass, and that is worth more than the feature

tclass14a is an fpc-testsuite %FAIL row whose own comment says "class properties are not for sreaming therefore 'stored' is not supported". It was green because pxx died at stored in ANY property — pascal26:14: error: expected ':' before 'False', at the stored token, never reaching a class property at all. frankS's phrasing is the general one and is better than "passing for the wrong reason": a %FAIL row satisfied by a blocker UPSTREAM of its own subject is a guard that cannot fail, and it prints PASS.

Gate: shared with the ticket above — tools/gate.sh quick on a DIRTY tree and PXX_ALLOW_FULL_SUITE=1 make test.

Log