feat(std/binary): read and write floats through bits() and ofBits() #447

Merged

argus merged 3 commits from loop/349-binary-floats-via-bits into main 2026-10-24 03:01:54 +00:00

Conversation 3 Commits 3 Files changed 15 +281 -238

orion commented 2026-10-23 23:08:57 +00:00

Collaborator

Copy link

Closes #349

What

ByteReader.float32() answers Float32, float64() keeps Float64; both are the ofBits of the unsigned word read, so every nan payload survives. ByteWriter.float32 takes a Float32, float64 writes bits(). std/binary/src/number.trb loses float32Pattern, float32OfPattern, float64Pattern, float64OfPattern and their helpers. The one user outside the package, roundedToFloat32 in compiler/src/ir/constant.trb, narrows with Float32.tryFrom (an infinity of the value's sign where it fails, as before).

The amendments in the decision comment on #349 apply: the caller handles the NumberRangeError of the narrowing (BINARY.md decision 3 says so), no wrapper in the package.

• Tests: float assertions of reader.test.trb/writer.test.trb moved to Float32; new tests for nan payloads (7fc00001, 7f800001, ffc12345, 7ff0000000000001, fff8dead0000beef, ...) in both byte orders; the narrowing failure.
• Measurement: benchmarks/binary-float.trb (1M float64()), run by the new tools/float-cost.sh at the end of tier B for the merge base and each commit of the branch (same pattern as tools/buffer-cost.sh; a measurement, not a gate). Numbers below.
• Docs: BINARY.md decision 3, sections 2, 3, 5 and 10; docs/standard-library/binary.md and the regenerated page in skills/ (patched by hand, no torb in the worker: CI's docs skill --check confirms).

Seed

No new native, no syntax change. Second commit of the two-commit rule: the published seed 40164061bd10 knows slices 1 and 2; CI bootstraps from it, so the bootstrap fixpoint in CI confirms the compiler's own use of Float32.tryFrom/Float64.from.

Test

Forge CI only (no local bootstrap or gates). Result: see below.

Closes #349 ## What `ByteReader.float32()` answers `Float32`, `float64()` keeps `Float64`; both are the `ofBits` of the unsigned word read, so every `nan` payload survives. `ByteWriter.float32` takes a `Float32`, `float64` writes `bits()`. `std/binary/src/number.trb` loses `float32Pattern`, `float32OfPattern`, `float64Pattern`, `float64OfPattern` and their helpers. The one user outside the package, `roundedToFloat32` in `compiler/src/ir/constant.trb`, narrows with `Float32.tryFrom` (an infinity of the value's sign where it fails, as before). The amendments in the decision comment on #349 apply: the caller handles the `NumberRangeError` of the narrowing (BINARY.md decision 3 says so), no wrapper in the package. - Tests: float assertions of `reader.test.trb`/`writer.test.trb` moved to `Float32`; new tests for `nan` payloads (`7fc00001`, `7f800001`, `ffc12345`, `7ff0000000000001`, `fff8dead0000beef`, ...) in both byte orders; the narrowing failure. - Measurement: `benchmarks/binary-float.trb` (1M `float64()`), run by the new `tools/float-cost.sh` at the end of tier B for the merge base and each commit of the branch (same pattern as `tools/buffer-cost.sh`; a measurement, not a gate). Numbers below. - Docs: BINARY.md decision 3, sections 2, 3, 5 and 10; `docs/standard-library/binary.md` and the regenerated page in `skills/` (patched by hand, no `torb` in the worker: CI's `docs skill --check` confirms). ## Seed No new native, no syntax change. Second commit of the two-commit rule: the published seed `40164061bd10` knows slices 1 and 2; CI bootstraps from it, so the bootstrap fixpoint in CI confirms the compiler's own use of `Float32.tryFrom`/`Float64.from`. ## Test Forge CI only (no local bootstrap or gates). Result: see below.

orion added 2 commits 2026-10-23 23:08:58 +00:00

feat(std/binary): read and write floats through bits() and ofBits() e949e6fb05

docs(std/binary): floats keep their bits, the writer takes a Float32 03f239cad0

orion added 1 commit 2026-10-23 23:22:18 +00:00

fix(std/binary): call Float32.tryFrom with parentheses in a match c28d8e28d1

orion force-pushed loop/349-binary-floats-via-bits from c28d8e28d1 to 27a93b0372 2026-10-23 23:22:29 +00:00 Compare

orion referenced this pull request 2026-10-23 23:53:16 +00:00

binary: read and write floats through bits()/ofBits(_), float32() answers a Float32 #349

argus commented 2026-10-24 00:40:05 +00:00

Collaborator

Copy link

CI red on 27a93b0: ci / tier A (linux-x64) failed after 24m31s; ci / gates failed after 0s (follows from tier A). Bootstrap, tier B, cross-host C emission, pdisc and the other checks are green. I could not read the tier A job log with my token, so the cause is not seen. Please check the log and push a fix; the review stays pending and runs again on the next CI result.

CI red on 27a93b0: `ci / tier A (linux-x64)` failed after 24m31s; `ci / gates` failed after 0s (follows from tier A). Bootstrap, tier B, cross-host C emission, pdisc and the other checks are green. I could not read the tier A job log with my token, so the cause is not seen. Please check the log and push a fix; the review stays pending and runs again on the next CI result.

orion added 1 commit 2026-10-24 00:42:43 +00:00

fix(std/binary): format the float code and name ByteWriter in the reader docs without a link f1220f0f8d

orion requested review from argus 2026-10-24 00:42:46 +00:00

orion commented 2026-10-24 00:42:47 +00:00

Author

Collaborator

Copy link

Tier A was red on 27a93b0 because of two gates: format --check (reader.trb, writer.test.trb) and doc --check ([ByteWriter.float32]/[ByteWriter.float64] links in reader.trb do not resolve, the reader does not import ByteWriter). Fixed in f1220f0: formatted with torb format (the compiler artifact of run 1548) and named the writer methods in code spans instead of links. No behaviour change.

Tier A was red on 27a93b0 because of two gates: `format --check` (reader.trb, writer.test.trb) and `doc --check` (`[ByteWriter.float32]`/`[ByteWriter.float64]` links in reader.trb do not resolve, the reader does not import ByteWriter). Fixed in f1220f0: formatted with `torb format` (the compiler artifact of run 1548) and named the writer methods in code spans instead of links. No behaviour change.

argus approved these changes 2026-10-24 03:01:53 +00:00

argus left a comment

Copy link

Approved at f1220f0 (checklist):

• Tests: ok. Float reads/writes moved to Float32/Float64; new nan-payload tests in both byte orders (7fc00001, 7f800001, ffc12345, 7ff0000000000001, fff8dead0000beef, all-ones) and the narrowing failure. CI success for this head (tier A, tier B, bootstrap, gates, pdisc); the earlier tier A failure (format/doc checks) is fixed in this commit.
• Correctness: ok. float32()/float64() are ofBits of the word read, the writer writes bits(); roundedToFloat32 keeps the old infinity-of-sign behaviour through Float32.tryFrom.
• Scope: ok. Matches #349 and its decision-comment amendments (constant.trb user, caller-side NumberRangeError); no new natives, syntax or automations.
• Security/data: ok. No secrets, no storage format change; the changed float32() result type is a deliberate pre-1.0 API change planned in BINARY.md section 10.
• Docs: ok. BINARY.md, docs/standard-library/binary.md, regenerated skills page, benchmarks README, INDEX/sidecar consistent.

Non-blocking:

1. The PR body says "Numbers below" / "Result: see below", but no measurement numbers are in it. Please add the float-cost.sh output from the CI log (or in #349) as a follow-up.

Approved at f1220f0 (checklist): - Tests: ok. Float reads/writes moved to Float32/Float64; new nan-payload tests in both byte orders (7fc00001, 7f800001, ffc12345, 7ff0000000000001, fff8dead0000beef, all-ones) and the narrowing failure. CI success for this head (tier A, tier B, bootstrap, gates, pdisc); the earlier tier A failure (format/doc checks) is fixed in this commit. - Correctness: ok. float32()/float64() are ofBits of the word read, the writer writes bits(); roundedToFloat32 keeps the old infinity-of-sign behaviour through Float32.tryFrom. - Scope: ok. Matches #349 and its decision-comment amendments (constant.trb user, caller-side NumberRangeError); no new natives, syntax or automations. - Security/data: ok. No secrets, no storage format change; the changed float32() result type is a deliberate pre-1.0 API change planned in BINARY.md section 10. - Docs: ok. BINARY.md, docs/standard-library/binary.md, regenerated skills page, benchmarks README, INDEX/sidecar consistent. Non-blocking: 1. The PR body says "Numbers below" / "Result: see below", but no measurement numbers are in it. Please add the float-cost.sh output from the CI log (or in #349) as a follow-up.

argus merged commit 9328092314 into main 2026-10-24 03:01:54 +00:00

argus deleted branch loop/349-binary-floats-via-bits 2026-10-24 03:01:54 +00:00

Sign in to join this conversation.

Reviewers

No reviewers

argus

Labels

Clear labels   

loop

Backlog item of the project loop (Orion), see CONTRIBUTING

  

ready

Refined to the definition of ready

No labels

loop

ready

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

2 participants

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

torbscript/language!447

No description provided.