Canonical JSON: numbers as ECMAScript writes them (jouleclaw-jcs), and the integer range (sandbox-attest) #313

Merged

dcharlot merged 4 commits from jouleclaw/jcs-ecmascript-numbers into main 2026-10-18 15:46:46 -04:00

Conversation 0 Commits 4 Files changed 7 +3574 -40

dcharlot commented 2026-10-18 02:37:22 -04:00

Owner

Copy link

Two canonical-JSON (RFC 8785) writers in the repository could produce text that other implementations do not.

jouleclaw-jcs laid numbers out as ryu does, not as ECMA-262's Number::toString, which RFC 8785 requires:

value	before	RFC 8785
0.000001	1e-6	0.000001
1.5e-5	1.5e-5	0.000015
1e16	1e16	10000000000000000
2^53 + 1 (u64)	9007199254740993	no canonical form

Every number is now the double it reads as, written with the fewest digits that read back as it (the nearest of those, a tie going to the even, as the standard says: RFC 8785's 1424953923781206.25 is 1424953923781206.2), in plain decimals from 10⁻⁶ up to 10²¹ and in exponent form outside. An integer past ±(2⁵³ − 1), which a double cannot hold exactly, is UnrepresentableNumber. Checked against RFC 8785's appendix B and 3,385 vectors written by a JavaScript engine (testdata/es-number-tostring.txt). The crate is not published and nothing in the repository depends on it. ryu is no longer a dependency.

sandbox-attest wrote integers verbatim at any size. Past 2⁵³ that is not what an implementation holding numbers as doubles writes, so the attestation's JCS form was not byte-identical across platforms, as SPEC §4 requires. SPEC §4 now states the range, and the writer refuses a core outside it. No current core holds such an integer (times are microseconds, energies microjoules). The sandbox workspace's tests, conformance vectors included, all pass; a manual is_multiple_of that clippy flags is fixed too.

The same writer, verified the same way (2 million random doubles against a JavaScript engine), now signs the quantum job service's receipts.

🤖 Generated with Claude Code

Two canonical-JSON (RFC 8785) writers in the repository could produce text that other implementations do not. **jouleclaw-jcs** laid numbers out as ryu does, not as ECMA-262's Number::toString, which RFC 8785 requires: | value | before | RFC 8785 | |---|---|---| | 0.000001 | `1e-6` | `0.000001` | | 1.5e-5 | `1.5e-5` | `0.000015` | | 1e16 | `1e16` | `10000000000000000` | | 2^53 + 1 (u64) | `9007199254740993` | no canonical form | Every number is now the double it reads as, written with the fewest digits that read back as it (the nearest of those, a tie going to the even, as the standard says: RFC 8785's 1424953923781206.25 is `1424953923781206.2`), in plain decimals from 10⁻⁶ up to 10²¹ and in exponent form outside. An integer past ±(2⁵³ − 1), which a double cannot hold exactly, is `UnrepresentableNumber`. Checked against RFC 8785's appendix B and 3,385 vectors written by a JavaScript engine (`testdata/es-number-tostring.txt`). The crate is not published and nothing in the repository depends on it. `ryu` is no longer a dependency. **sandbox-attest** wrote integers verbatim at any size. Past 2⁵³ that is not what an implementation holding numbers as doubles writes, so the attestation's JCS form was not byte-identical across platforms, as SPEC §4 requires. SPEC §4 now states the range, and the writer refuses a core outside it. No current core holds such an integer (times are microseconds, energies microjoules). The sandbox workspace's tests, conformance vectors included, all pass; a manual `is_multiple_of` that clippy flags is fixed too. The same writer, verified the same way (2 million random doubles against a JavaScript engine), now signs the quantum job service's receipts. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

dcharlot added 2 commits 2026-10-18 02:37:22 -04:00

jouleclaw-jcs: numbers as ECMAScript writes them, as RFC 8785 requires ... 92f5ac4dc1

The writer laid numbers out as ryu does, which is not ECMA-262's
Number::toString: 0.000001 came out as 1e-6, 1e16 as 1e16, 1.5e-5 as
1.5e-5, and integers past 2^53 verbatim, so a JCS text from this crate
could not be reproduced by any other RFC 8785 implementation. Every number
is now the double it reads as, written with the fewest digits that read
back (the nearest of those, a tie going to the even: RFC 8785's
1424953923781206.25 is ...6.2), in plain decimals from 1e-6 up to 1e21 and
in exponent form outside. An integer past +-(2^53 - 1), which a double
cannot hold, is UnrepresentableNumber. Checked against RFC 8785's appendix
B and 3,385 vectors written by a JavaScript engine. ryu is no longer a
dependency.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

sandbox: the attestation's JCS form refuses integers past +-(2^53 - 1), as RFC 8785 and SPEC §4's byte-identical rule require ... 5e9f48c090

RFC 8785 holds numbers as IEEE-754 doubles (I-JSON, RFC 7493 §2.2). A
larger integer was written verbatim, which an implementation that reads it
as a double writes differently, so the encoding was not byte-identical
across platforms as SPEC §4 requires. SPEC §4 now says so, and the writer
refuses such a core (IntegerOutOfRange). No current core holds one: times
are microseconds and energies microjoules, both far below 2^53. Also a
manual is_multiple_of that clippy flags.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

dcharlot added 1 commit 2026-10-18 07:20:25 -04:00

ci: run the checks again (the jcp-rs job lost its connection while downloading the Rust installer, before building anything) ... 25d1987948

Co-Authored-By: Claude Opus 5.5 <[email protected]>

dcharlot added 1 commit 2026-10-18 10:36:01 -04:00

ci: run the checks again (the macOS energy job ran out of disk on the runner's volume, before building anything of this change) ... 4c41f7e2a7

Co-Authored-By: Claude Opus 5.5 <[email protected]>

dcharlot merged commit f5b1b4e4fa into main 2026-10-18 15:46:46 -04:00

dcharlot referenced this pull request from a commit 2026-10-18 15:46:47 -04:00

Merge pull request 'Canonical JSON: numbers as ECMAScript writes them (jouleclaw-jcs), and the integer range (sandbox-attest)' (#313) from jouleclaw/jcs-ecmascript-numbers 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

Transaction-Science/open-standards!313

No description provided.