● All three issues are fixed, tested, and verified end-to-end. Full workspace test #34

Merged

navicore merged 2 commits from issue-31-32-33 into main 2026-06-23 20:26:52 +00:00

Conversation 1 Commits 2 Files changed 8 +224 -21

navicore commented 2026-06-23 19:59:52 +00:00

Owner

Copy link

suite passes (24 test binaries, 0 failures) and clippy is clean.

#31
#32
#33

Summary

#31 — functor/3 construct aliased all arg slots (termops.rs)
The construct loop called m.new_var() (which pushes its own self-ref cell) and then
pushed the returned ref again, laying down two cells per slot so every later arg
aliased the first variable. Now it writes a self-referencing REF directly into each
contiguous arg cell.

• functor(T, point, 2) → T = point(_6, _7) (was point(_6, _6))
• functor(T, point, 2), T = point(3, 4) → succeeds (was false.)

#32 — whole floats lost .0 in write/1 (render.rs)
term_to_string (used by write/1 and the binding display) formatted floats with raw
{}; render.rs already had the correct fmt_float helper and term_to_json used it.
Routed term_to_string through fmt_float too.

• write(2.0) → 2.0, 2 ** 10 → 1024.0, 3.5 unchanged.

#33 — missing /1 evaluable and writeq/1

• Added unary bitwise-complement arm ("\", 1) + bit_not to the arith evaluator
  (arith.rs). X is \ 0 → -1, \ 5 → -6. The operator already parsed; only evaluation
  was missing.
• Added writeq/1: a quoting variant of the term writer (render.rs
  term_to_string_quoted + atom_is_unquoted/quote_atom), wired through the runtime
  impl (miscops.rs), the shared vocabulary (builtins.rs), the compiler symbol map
  (lower.rs), the runtime metacall dispatch (control.rs), and the reference doc.
  writeq(foo) → foo, writeq('hello world') → 'hello world'.

On ordering: #31 was fully independent; #32 landed before #33 so writeq inherited
correct float rendering through the shared writer. Each came with a regression test
pinning the issue's exact reproduction.

suite passes (24 test binaries, 0 failures) and clippy is clean. https://git.navicore.tech/navicore/patch-prolog/issues/31 https://git.navicore.tech/navicore/patch-prolog/issues/32 https://git.navicore.tech/navicore/patch-prolog/issues/33 Summary #31 — functor/3 construct aliased all arg slots (termops.rs) The construct loop called m.new_var() (which pushes its own self-ref cell) and then pushed the returned ref again, laying down two cells per slot so every later arg aliased the first variable. Now it writes a self-referencing REF directly into each contiguous arg cell. - functor(T, point, 2) → T = point(_6, _7) (was point(_6, _6)) - functor(T, point, 2), T = point(3, 4) → succeeds (was false.) #32 — whole floats lost .0 in write/1 (render.rs) term_to_string (used by write/1 and the binding display) formatted floats with raw {}; render.rs already had the correct fmt_float helper and term_to_json used it. Routed term_to_string through fmt_float too. - write(2.0) → 2.0, 2 ** 10 → 1024.0, 3.5 unchanged. #33 — missing \/1 evaluable and writeq/1 - Added unary bitwise-complement arm ("\\", 1) + bit_not to the arith evaluator (arith.rs). X is \ 0 → -1, \ 5 → -6. The operator already parsed; only evaluation was missing. - Added writeq/1: a quoting variant of the term writer (render.rs term_to_string_quoted + atom_is_unquoted/quote_atom), wired through the runtime impl (miscops.rs), the shared vocabulary (builtins.rs), the compiler symbol map (lower.rs), the runtime metacall dispatch (control.rs), and the reference doc. writeq(foo) → foo, writeq('hello world') → 'hello world'. On ordering: #31 was fully independent; #32 landed before #33 so writeq inherited correct float rendering through the shared writer. Each came with a regression test pinning the issue's exact reproduction.

navicore added 1 commit 2026-06-23 19:59:52 +00:00

● All three issues are fixed, tested, and verified end-to-end. Full workspace test ... fd4b8640a3

suite passes (24 test binaries, 0 failures) and clippy is clean.

  Summary

  #31 — functor/3 construct aliased all arg slots (termops.rs)
  The construct loop called m.new_var() (which pushes its own self-ref cell) and then
  pushed the returned ref again, laying down two cells per slot so every later arg
  aliased the first variable. Now it writes a self-referencing REF directly into each
  contiguous arg cell.
  - functor(T, point, 2) → T = point(_6, _7) (was point(_6, _6))
  - functor(T, point, 2), T = point(3, 4) → succeeds (was false.)

  #32 — whole floats lost .0 in write/1 (render.rs)
  term_to_string (used by write/1 and the binding display) formatted floats with raw
  {}; render.rs already had the correct fmt_float helper and term_to_json used it.
  Routed term_to_string through fmt_float too.
  - write(2.0) → 2.0, 2 ** 10 → 1024.0, 3.5 unchanged.

  #33 — missing \/1 evaluable and writeq/1
  - Added unary bitwise-complement arm ("\\", 1) + bit_not to the arith evaluator
  (arith.rs). X is \ 0 → -1, \ 5 → -6. The operator already parsed; only evaluation
  was missing.
  - Added writeq/1: a quoting variant of the term writer (render.rs
  term_to_string_quoted + atom_is_unquoted/quote_atom), wired through the runtime
  impl (miscops.rs), the shared vocabulary (builtins.rs), the compiler symbol map
  (lower.rs), the runtime metacall dispatch (control.rs), and the reference doc.
  writeq(foo) → foo, writeq('hello world') → 'hello world'.

  On ordering: #31 was fully independent; #32 landed before #33 so writeq inherited
  correct float rendering through the shared writer. Each came with a regression test
  pinning the issue's exact reproduction.

navicore referenced this pull request 2026-06-23 20:00:10 +00:00

functor/3 construct mode shares one variable across all argument slots #31

navicore referenced this pull request 2026-06-23 20:00:22 +00:00

write/1 prints whole-valued floats without a decimal point (non-ISO output) #32

navicore referenced this pull request 2026-06-23 20:00:33 +00:00

Missing builtins: evaluable \/1 (bitwise complement) and writeq/1 #33

navicore commented 2026-06-23 20:12:50 +00:00

Author

Owner

Copy link

Review: all three issues fixed and verified end-to-end ✅ (one minor follow-up)

Reviewed the full diff and verified every fix against the original issue
reproductions on a build of this branch (fd4b864).

#31 — functor/3 construct ✅

Correct root-cause fix: replaces new_var() (which double-pushed a cell and
aliased every argument slot to one variable) with a self-referencing REF
written directly into each contiguous arg cell. The regression test asserts both
distinct vars and the behavioral T = point(3, 4) binding.

?- functor(T, point, 2).            T = point(_6, _7)   % distinct
?- functor(T, point, 2), T = point(3, 4).   % succeeds

#33 — \/1 + writeq/1 ✅

\/1 is !n, integer-only (errors on float via int_args_required — correct,
ISO \/1 is integer-only). writeq/1 is wired through every registration point
consistently (codegen DET_BUILTINS, det_builtin dispatch, vocab_invariant,
shared roster 55→56 / Det 25→26), with thorough quoting logic and tests.

?- X is \ 0.    X = -1
?- X is \ 5.    X = -6
?- writeq('hello world'), nl.    'hello world'

#32 — write/1 whole-valued floats ✅ (with a gap, below)

Routes the term_to_string float branch through the existing fmt_float (which
already forced .0, and which term_to_json already used). Minimal and correct.

?- write(2.0).        2.0
?- X is 2 ** 10, write(X).    1024.0
?- X is 7 / 2, write(X).      3.5

Minor follow-up (non-blocking): #32 misses the error-message path

fmt_float now covers term_to_string and term_to_json, but format_term_v
(render.rs, the byte contract for error messages) still uses raw
f64::to_string(), which drops the .0. So a whole float embedded in an error
term reintroduces exactly the "indistinguishable from an integer" problem #32
fixes:

?- atom_length(2.0, X).
Error: ... error(type_error(atom, 2), ...)      % culprit is 2.0, prints as "2"
?- atom_concat(2.0, b, X).
Error: ... error(type_error(atom, 2), ...)

Cosmetic (error messages only), so not a merge blocker — but routing that last
TAG_FLT branch through fmt_float would make all float rendering consistent.

Everything else — code quality, regression test coverage, doc updates
(builtin-reference.md, roster counts) — is clean. Nice work.

## Review: all three issues fixed and verified end-to-end ✅ (one minor follow-up) Reviewed the full diff and verified every fix against the original issue reproductions on a build of this branch (`fd4b864`). ### #31 — `functor/3` construct ✅ Correct root-cause fix: replaces `new_var()` (which double-pushed a cell and aliased every argument slot to one variable) with a self-referencing `REF` written directly into each contiguous arg cell. The regression test asserts both distinct vars **and** the behavioral `T = point(3, 4)` binding. ``` ?- functor(T, point, 2). T = point(_6, _7) % distinct ?- functor(T, point, 2), T = point(3, 4). % succeeds ``` ### #33 — `\/1` + `writeq/1` ✅ `\/1` is `!n`, integer-only (errors on float via `int_args_required` — correct, ISO `\/1` is integer-only). `writeq/1` is wired through every registration point consistently (codegen `DET_BUILTINS`, `det_builtin` dispatch, `vocab_invariant`, shared roster 55→56 / Det 25→26), with thorough quoting logic and tests. ``` ?- X is \ 0. X = -1 ?- X is \ 5. X = -6 ?- writeq('hello world'), nl. 'hello world' ``` ### #32 — `write/1` whole-valued floats ✅ (with a gap, below) Routes the `term_to_string` float branch through the existing `fmt_float` (which already forced `.0`, and which `term_to_json` already used). Minimal and correct. ``` ?- write(2.0). 2.0 ?- X is 2 ** 10, write(X). 1024.0 ?- X is 7 / 2, write(X). 3.5 ``` ### Minor follow-up (non-blocking): #32 misses the error-message path `fmt_float` now covers `term_to_string` and `term_to_json`, but **`format_term_v`** (`render.rs`, the byte contract for error messages) still uses raw `f64::to_string()`, which drops the `.0`. So a whole float embedded in an error term reintroduces exactly the "indistinguishable from an integer" problem #32 fixes: ``` ?- atom_length(2.0, X). Error: ... error(type_error(atom, 2), ...) % culprit is 2.0, prints as "2" ?- atom_concat(2.0, b, X). Error: ... error(type_error(atom, 2), ...) ``` Cosmetic (error messages only), so not a merge blocker — but routing that last `TAG_FLT` branch through `fmt_float` would make all float rendering consistent. Everything else — code quality, regression test coverage, doc updates (`builtin-reference.md`, roster counts) — is clean. Nice work.

navicore added 1 commit 2026-06-23 20:24:56 +00:00

Done. Addressed the one non-blocking follow-up from PR 34. ... 8236ed83d6

format_term_v now routes through fmt_float (render.rs:272) — the last of
the three TAG_FLT rendering paths. All float rendering (JSON, text
bindings/write, and error-message terms) is now consistent. A
whole-valued float embedded in an error culprit keeps its .0:

?- atom_length(2.0, X).
Error: ... error(type_error(atom, 2.0), ...)      % was "2"
?- atom_concat(2.0, b, X).
Error: ... error(type_error(atom, 2.0), ...)

I extended the existing #32 regression test to also assert format_term
keeps the .0, locking the error-path behavior in. Full workspace suite
(24 binaries) passes and clippy is clean — no oracle/error-message test
had pinned the old 2 rendering, confirming the change is safe.

navicore merged commit e525d8ab8a into main 2026-06-23 20:26:52 +00:00

navicore deleted branch issue-31-32-33 2026-06-23 20:26:53 +00:00

navicore referenced this pull request from a commit 2026-06-23 20:26:53 +00:00

Merge pull request '● All three issues are fixed, tested, and verified end-to-end. Full workspace test' (#34) from issue-31-32-33 into main

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels

No items

No labels

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

navicore/patch-prolog!34

No description provided.