administrator/blaise
1
0
Fork 0
Code Issues Pull requests Actions Packages Projects Releases Wiki Activity

docs: narrow the ordinal-set future-improvements entry to the unimplemented part

Browse source 
`set of byte` and `set of Boolean` are implemented (issue #105) — named ordinal
base types lower to a bit set (byte uses the 256-bit jumbo path), with working
membership/Include/Exclude and [lo..hi] range literals on both backends. The
section is rewritten to record that, leaving only the genuinely-remaining work:

* the anonymous integer-subrange base form `set of 0..255` / `set of 1..10`,
  still rejected at parse time ("Expected enum type or '(' after 'set of'");
* a minor `set of Boolean` follow-on where `True in s` is rejected by the `in`
  type-check (Ord(b) in s works).
This commit is contained in:
Graeme Geldenhuys 2026-06-18 13:00:04 +01:00
parent bc0485285f
commit f107b0dabc
1 changed files with 16 additions and 15 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

31
docs/future-improvements.adoc
View file

@ -34,24 +34,25 @@ prioritise them.
- reference to function/procedure (anonymous methods)
- cdecl/stdcall calling-convention markers on procedural types
=== Ordinal-based set types (`set of byte`, `set of 0..255`)
=== Set types over an integer subrange (`set of 0..255`)
Set types currently require an enumeration base type; `set of byte` (and,
more generally, a set over an integer subrange) is rejected with
"Set base type 'byte' must be an enumeration type". FPC and Delphi accept
ordinal base types, so `set of byte` is standard Pascal.
`set of byte` and `set of Boolean` are *implemented* (issue #105): the named
ordinal base types lower to a bit set — `set of byte` uses the 256-bit jumbo
path, membership/Include/Exclude and `[lo..hi]` range literals all work on both
backends. See `docs/language-rationale.adoc` and the `set of Byte` tests in
`cp.test.sets.pas` / `cp.test.e2e.sets.pas`.
Supporting it means teaching the set machinery to treat an ordinal range as
an implicit member set: the bit count is the range width (0..255 → 256 bits,
the jumbo-set path), element ordinals are the integer values themselves, and
membership/include/exclude lower exactly as they do for enum sets. The work
spans the semantic pass (base-type validation, bit-count derivation) and both
codegen backends, including the jumbo-set path for `set of byte`.
What remains is the *anonymous integer-subrange* base form,
`set of 0..255` / `set of 1..10`, which is still rejected at parse time
("Expected enum type or '(' after 'set of'"). Standard Pascal accepts a
subrange there; the bit count is the range width and element ordinals are the
integer values themselves — the same machinery `set of byte` already uses, so
this is mostly parser + base-type-resolution work, not new codegen.
This was surfaced by issue #105, whose example used `set of byte`. The set
range syntax `[lo..hi]` requested by that issue was implemented independently
(see `docs/language-rationale.adoc`) and is base-type-agnostic — ranges will
work for ordinal sets automatically once this lands.
Minor follow-on: for `set of Boolean`, a Boolean operand on the left of `in`
(e.g. `True in s`) is rejected by the `in` type-check ("operand must be
numeric"); `Ord(b) in s` works. The check should accept a Boolean (or any
ordinal) operand against a matching ordinal-base set.
=== Method level attributes