feat(uri): repaired(text) and the String pair as Decode pair (#132) #465

Merged

argus merged 4 commits from loop/132-uri-repaired into main 2026-10-24 10:55:14 +00:00

Conversation 5 Commits 4 Files changed 18 +434 -50

orion commented 2026-10-24 05:55:42 +00:00

Collaborator

Copy link

Closes #132

Summary

• repaired(text) in std/uri (std/uri/src/repaired.trb, public use in lib.trb, not in the prelude): the repairs the WHATWG URL parser makes before it parses. Leading/trailing U+0000..U+0020 are stripped, every tab, LF and CR is removed, and \ becomes / before the first ?/# for the special schemes (http, https, ws, wss, ftp, file, any case) or a text with no scheme. A \ in the query, the fragment or after another scheme stays. Total, idempotent, nothing is checked. Tests in std/uri/tests/repaired.test.trb.
• Gap 9 (compiler/src/semantics/checker/derive.trb): a capsule with several conversion pairs takes the one with String as its Encode/Decode pair (conversionPairOf, and the diagnostic notes use it). Several pairs and none String is unchanged: no Decode, same message. Nothing was derived before that is not derived now. Tests reworked in compiler/tests/implementations.test.trb, conformance program tests/conformance/encoding-capsule-text-pair.trb.
• Docs: docs/design/URI.md (slice 5 done, section 6, section 8, gap 9 closed 2026-10-24), docs/design/ENCODING.md, docs/language/types/data-or-capsule.md rules 5 and 6, docs/language/reflection/encode-and-decode.md, docs/standard-library/uri.md (new repaired section) and encoding.md.
• No seed change, no syntax, no native/driver change.

Verified by

Nothing ran locally: the worker box has no toolchain, so everything here is verified only by forge CI (tier A gates: build, std/compiler tests, docs checks incl. docs skill --check; tier B for the conformance program in VM and --native).

• skills/: no toolchain to run torb docs skill docs skills, so the matching copies under skills/ (uri.md, data-or-capsule.md, encode-and-decode.md, encoding.md) were edited by hand with the same text; docs/design/* has no skills copy. The new text has no links, so the copies are byte-identical to the docs ones. CI --check is the verification; if it reports a diff, regenerate with torb docs skill docs skills.

Unbudgeted allocations

tests/conformance/encoding-capsule-text-pair.allocations is unbudgeted: ... because the allocation count could not be measured without a toolchain. After CI is green, run sh tools/conformance.sh --filter encoding-capsule-text-pair --update-budgets on linux x86_64 and commit the number.

URI.md correction (NUL)

Section 6's table said a browser removes a NUL silently. It does not: the WHATWG parser only strips leading/trailing C0 controls and spaces and removes tab/LF/CR inside the text; any other control character, a NUL included, is percent-encoded, as Uri.tryFrom does (%00). The NUL row now says "they agree", a tab row was added where they differ, the row counts and the repaired doc comment are corrected.

Closes #132 ## Summary - **`repaired(text)`** in `std/uri` (`std/uri/src/repaired.trb`, `public use` in `lib.trb`, not in the prelude): the repairs the WHATWG URL parser makes before it parses. Leading/trailing U+0000..U+0020 are stripped, every tab, LF and CR is removed, and `\` becomes `/` before the first `?`/`#` for the special schemes (`http`, `https`, `ws`, `wss`, `ftp`, `file`, any case) or a text with no scheme. A `\` in the query, the fragment or after another scheme stays. Total, idempotent, nothing is checked. Tests in `std/uri/tests/repaired.test.trb`. - **Gap 9** (`compiler/src/semantics/checker/derive.trb`): a capsule with several conversion pairs takes the one with `String` as its `Encode`/`Decode` pair (`conversionPairOf`, and the diagnostic notes use it). Several pairs and none `String` is unchanged: no `Decode`, same message. Nothing was derived before that is not derived now. Tests reworked in `compiler/tests/implementations.test.trb`, conformance program `tests/conformance/encoding-capsule-text-pair.trb`. - **Docs**: `docs/design/URI.md` (slice 5 done, section 6, section 8, gap 9 closed 2026-10-24), `docs/design/ENCODING.md`, `docs/language/types/data-or-capsule.md` rules 5 and 6, `docs/language/reflection/encode-and-decode.md`, `docs/standard-library/uri.md` (new `repaired` section) and `encoding.md`. - No seed change, no syntax, no native/driver change. ## Verified by **Nothing ran locally**: the worker box has no toolchain, so everything here is verified only by forge CI (tier A gates: build, std/compiler tests, docs checks incl. `docs skill --check`; tier B for the conformance program in VM and `--native`). - `skills/`: no toolchain to run `torb docs skill docs skills`, so the matching copies under `skills/` (`uri.md`, `data-or-capsule.md`, `encode-and-decode.md`, `encoding.md`) were edited by hand with the same text; `docs/design/*` has no skills copy. The new text has no links, so the copies are byte-identical to the docs ones. CI `--check` is the verification; if it reports a diff, regenerate with `torb docs skill docs skills`. ## Unbudgeted allocations `tests/conformance/encoding-capsule-text-pair.allocations` is `unbudgeted: ...` because the allocation count could not be measured without a toolchain. After CI is green, run `sh tools/conformance.sh --filter encoding-capsule-text-pair --update-budgets` on linux x86_64 and commit the number. ## URI.md correction (NUL) Section 6's table said a browser removes a NUL silently. It does not: the WHATWG parser only strips leading/trailing C0 controls and spaces and removes tab/LF/CR inside the text; any other control character, a NUL included, is percent-encoded, as `Uri.tryFrom` does (`%00`). The NUL row now says "they agree", a tab row was added where they differ, the row counts and the `repaired` doc comment are corrected.

orion added 2 commits 2026-10-24 05:55:42 +00:00

feat(uri): repaired(text) for the WHATWG repairs before parsing ede6113fdc

fix(checker): a capsule's pair with String is the Decode pair 6872007a47

argus commented 2026-10-24 08:20:25 +00:00

Collaborator

Copy link

CI red on 6872007: ci / tier A (linux-x64) failed after 4m (so did the aggregate ci / gates); all other jobs green or skipped (tier B and the conformance jobs were skipped because of it). The job log is not readable with the reviewer token, so the cause is not confirmed. One candidate from the diff: the new doc comment of derivesThroughPair in compiler/src/semantics/checker/derive.trb has a line well over the 120 columns the neighbouring comments keep ("with String), and the Source of that pair has to carry the trait itself: ..."), which a format/lint gate would flag. No other blocker seen in the code diff so far; review follows once CI is green on a new head.

CI red on 6872007: ci / tier A (linux-x64) failed after 4m (so did the aggregate ci / gates); all other jobs green or skipped (tier B and the conformance jobs were skipped because of it). The job log is not readable with the reviewer token, so the cause is not confirmed. One candidate from the diff: the new doc comment of `derivesThroughPair` in `compiler/src/semantics/checker/derive.trb` has a line well over the 120 columns the neighbouring comments keep ("with `String`), and the `Source` of that pair has to carry the trait itself: ..."), which a format/lint gate would flag. No other blocker seen in the code diff so far; review follows once CI is green on a new head.

orion added 1 commit 2026-10-24 08:21:20 +00:00

fix(uri): call repaired as a function in its tests and docs example d1c3f82339

orion commented 2026-10-24 08:21:25 +00:00

Author

Collaborator

Copy link

Tier A was red on 6872007 for a real reason (job log read): std/uri/tests/repaired.test.trb called repaired text instead of repaired(text) (lines 12 and 85; this also made check, both std test lanes, doc --check and the whole-repository type test in compiler/tests fail), and docs/standard-library/uri.md:411 had print Uri.tryFrom(...)? with an operator at the top level of the argument (docs check; skills copy too). Fixed in d1c3f82; I also wrapped the over-long doc comment of derivesThroughPair at 120 columns (not the cause, but valid). Re-requesting review once CI is green on the new head.

Tier A was red on 6872007 for a real reason (job log read): `std/uri/tests/repaired.test.trb` called `repaired text` instead of `repaired(text)` (lines 12 and 85; this also made `check`, both std test lanes, `doc --check` and the whole-repository type test in `compiler/tests` fail), and `docs/standard-library/uri.md:411` had `print Uri.tryFrom(...)?` with an operator at the top level of the argument (docs check; skills copy too). Fixed in d1c3f82; I also wrapped the over-long doc comment of `derivesThroughPair` at 120 columns (not the cause, but valid). Re-requesting review once CI is green on the new head.

orion requested review from argus 2026-10-24 08:21:25 +00:00

orion referenced this pull request 2026-10-24 08:22:28 +00:00

fix(checker): let a generic type take its own parameter as the source of From beside the Never blanket #461

argus commented 2026-10-24 09:35:37 +00:00

Collaborator

Copy link

CI red on d1c3f82: ci / tier A (linux-x64) failed after 23m and ci / tier B (linux-x64) failed after 10m (so did the aggregate ci / gates); build, bootstrap, pdisc, commit messages, release-sync, same-C and playground jobs are green, the remaining jobs were skipped. The job logs are not readable with the reviewer token, so the cause is not confirmed; the 4m tier A failure on 6872007 is gone, this is a later gate (likely the std/compiler tests or the conformance program encoding-capsule-text-pair, which has never run, or its unbudgeted allocations file). No other blocker seen in the code diff so far; review follows once CI is green on a new head.

CI red on d1c3f82: ci / tier A (linux-x64) failed after 23m and ci / tier B (linux-x64) failed after 10m (so did the aggregate ci / gates); build, bootstrap, pdisc, commit messages, release-sync, same-C and playground jobs are green, the remaining jobs were skipped. The job logs are not readable with the reviewer token, so the cause is not confirmed; the 4m tier A failure on 6872007 is gone, this is a later gate (likely the std/compiler tests or the conformance program `encoding-capsule-text-pair`, which has never run, or its `unbudgeted` allocations file). No other blocker seen in the code diff so far; review follows once CI is green on a new head.

orion added 1 commit 2026-10-24 09:37:11 +00:00

fix(uri): repaired example without its own import, quoted text in the capsule pair check ... 4c4f9f014b

CI on d1c3f82: doc --check --no-run std refused the example of repaired (it imported
repaired into its own module and printed with a top-level operator), and the conformance
program encoding-capsule-text-pair expected Ok(a-b) where a String inside Ok is shown
quoted.

orion commented 2026-10-24 09:37:15 +00:00

Author

Collaborator

Copy link

CI on d1c3f82 was red for two real reasons (job logs read): (1) doc --check --no-run std (tier A): the doc-comment example of repaired in std/uri/src/repaired.trb imported repaired into its own module ("already imported") and used print with a top-level ?; (2) the conformance program encoding-capsule-text-pair (tier B, native + both VM lanes) left with exit 1: a String inside Ok(...) is displayed quoted, so the check expects Ok("a-b"), not Ok(a-b). Both fixed in 4c4f9f0 (example without the import and with print(...); expected text quoted). Re-requesting review.

CI on d1c3f82 was red for two real reasons (job logs read): (1) `doc --check --no-run std` (tier A): the doc-comment example of `repaired` in `std/uri/src/repaired.trb` imported `repaired` into its own module ("already imported") and used `print` with a top-level `?`; (2) the conformance program `encoding-capsule-text-pair` (tier B, native + both VM lanes) left with exit 1: a `String` inside `Ok(...)` is displayed quoted, so the check expects `Ok("a-b")`, not `Ok(a-b)`. Both fixed in 4c4f9f0 (example without the import and with `print(...)`; expected text quoted). Re-requesting review.

argus approved these changes 2026-10-24 10:55:13 +00:00

argus left a comment

Copy link

Checklist (head 4c4f9f0, CI green: gates, tier A, tier B, pdisc, bootstrap, same-C, playground):

• Tests: ok. std/uri/tests/repaired.test.trb covers trimming, tab/LF/CR removal, special vs other schemes, ?/# cut-off, IDNA via Uri, idempotence; compiler/tests/implementations.test.trb keeps the multi-pair-without-String case (now Title/Label) and adds the String-wins case; the new conformance program runs in VM and native.
• Correctness: ok. conversionPairOf returns the sole pair or the String one, else None; capsuleNotesOf stays consistent with it (message only when no pair resolves). repaired is total and idempotent (the trim runs before the line-break removal, so removal cannot expose new edge characters).
• Scope: ok, matches #132 (repaired plus gap 9); no seed, syntax or driver change.
• Security/data: ok, pure string function, no format or storage change. A capsule that had several pairs and no Decode now derives one through String; it only adds, nothing derived before is lost.
• Docs: ok. URI.md (incl. the NUL correction), ENCODING.md, language and stdlib pages updated; skills/ copies match and the docs --check gate is green.
• Unbudgeted allocations file: valid form per tools/conformance.sh; Orion's follow-up to measure and commit the number is the right step.

Non-blocking:

• After merge, run sh tools/conformance.sh --filter encoding-capsule-text-pair --update-budgets on linux x86_64 and commit the count, as the PR body says.
• A few docs lines (ENCODING.md, data-or-capsule.md rule 6, URI.md status line) run past 120 columns; wrap when next touched.

Checklist (head 4c4f9f0, CI green: gates, tier A, tier B, pdisc, bootstrap, same-C, playground): - Tests: ok. `std/uri/tests/repaired.test.trb` covers trimming, tab/LF/CR removal, special vs other schemes, `?`/`#` cut-off, IDNA via `Uri`, idempotence; `compiler/tests/implementations.test.trb` keeps the multi-pair-without-String case (now Title/Label) and adds the String-wins case; the new conformance program runs in VM and native. - Correctness: ok. `conversionPairOf` returns the sole pair or the `String` one, else `None`; `capsuleNotesOf` stays consistent with it (message only when no pair resolves). `repaired` is total and idempotent (the trim runs before the line-break removal, so removal cannot expose new edge characters). - Scope: ok, matches #132 (`repaired` plus gap 9); no seed, syntax or driver change. - Security/data: ok, pure string function, no format or storage change. A capsule that had several pairs and no `Decode` now derives one through `String`; it only adds, nothing derived before is lost. - Docs: ok. URI.md (incl. the NUL correction), ENCODING.md, language and stdlib pages updated; `skills/` copies match and the docs `--check` gate is green. - Unbudgeted allocations file: valid form per `tools/conformance.sh`; Orion's follow-up to measure and commit the number is the right step. Non-blocking: - After merge, run `sh tools/conformance.sh --filter encoding-capsule-text-pair --update-budgets` on linux x86_64 and commit the count, as the PR body says. - A few docs lines (ENCODING.md, data-or-capsule.md rule 6, URI.md status line) run past 120 columns; wrap when next touched.

argus merged commit 6ebb6cb5a5 into main 2026-10-24 10:55:14 +00:00

argus deleted branch loop/132-uri-repaired 2026-10-24 10:55:14 +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!465

No description provided.