Repository navigation
Add the note severity and a user-facing sentence for every diagnostic - #193
Merged
Merged
Conversation
The 97 switch-arm warning twins (PF-W1001..PF-W1097) carried their error twin's user_message word for word, so 29 of them told a user that a call PineForge had just accepted "is refused" (request.seed in a non-default switch arm is PF-W1027, a warning with the na fallback kept), and the nine "TradingView rejects" ones never said PineForge only warns. - catalog: rewrite user_message of all 97 twins. Ninety-one say what the limitation is and that "In a switch arm PineForge only warns and keeps the arm's lowering" (the catalog's own explanation of the twin); the six DIVERGENT_VARS warnings (PF-W1067..PF-W1072) are warnings everywhere and name no switch arm. PF-E6062 now says it is array and matrix history inside request.security (collection_history.py), not user-defined type or drawing references (that is PF-E3043). Only user_message text changes: entry and field order, message, hint, explanation, args, severity and the pin digests are byte for byte as before. - tests/test_diagnostic_notes.py: one case per twin (own sentence, no refusal wording, the switch-arm clause exactly where the twin's explanation has it), a no-refusal check over every nonfatal sentence with a detector table and mutation controls, a real accepted-arm interface case (request.seed -> PF-W1027, text constant -> PF-W1078: transpile succeeds, warning level, glue envelope carries the sentence), the same call outside an arm keeping the error sentence, and PF-E6062 on the request_security_size shape. - tests/test_diagnostic_codes.py: restore (severity == "error") == (code[3] == "E"). New tests are authored, not run (phase limit); the final Spot proof runs them. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…call-out PUBLIC_CONTRACT.md: replace the claim that visual-only and display-only diagnostics are informational with the exact set of eleven codes that migrate from warning to note, and name PF-W1502, PF-W1506, PF-W1507, PF-W1078 and PF-W1552 as warnings that remain. Replace the stale "does not return nonfatal warnings" statement with the error, warning and note levels, read from the declared severity and never from the PF-E/PF-W prefix. State that PF-W1505 stays a static note with no per-call condition or suppression and that no sidecar exists yet, and that the code pin does not cover user_message (only the eleven migrated templates are fixed, by delta.json). Mark the note severity and user_message additions "since 1.5.0". CHANGELOG.md: add an Unreleased 1.5.0 entry with the consumer migration call-out: note is nonfatal, user_message is a raw ICU template over args, the exact migrated-code set, the stable legacy fields, and the warmup sidecar as later work. No release date; VERSION, package manifests and dependencies are unchanged. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
README, CLAUDE.md and AGENTS.md still said nonfatal diagnostics are warnings only. Name the note level where they describe what transpile_full() and the JSON envelopes return, read severity from the declared level instead of the PF-E/PF-W prefix, show user_message in the catalog entry example, and call the ta.ema warmup caveat, the visual-only skip and the display-only request lowering notes. Only PF-W1505, PF-W1508, PF-W1509 and PF-W1525 to PF-W1532 are notes; the other visual-only and display-only diagnostics stay warnings. No runtime warmup sidecar or conditional suppression is claimed. Rename tests/test_external_requests.py::test_inert_requests_are_lowered_with_a_warning to test_inert_requests_are_lowered_with_a_note. Body, assertions and everything else in the file are unchanged. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The note for a call PineForge skips said "{name} is not drawn in
backtests.", which is wrong for alert() and alertcondition(): they draw
nothing on a chart. The sentence is now "{name}() does nothing in a
backtest." (PF-W1509: "{full}() does nothing in a backtest."), which holds
for every call the note covers. `name` is the call's bare or dotted name,
so the template's "()" is the only pair of parentheses a rendered
sentence has.
Only the two user_message templates change. Severity, message, hint, args,
explanation, codes and the pin are untouched, and the executable product
files are the parent's. The delta fixture, the receiver vector and the two
doc examples quote the same text. New tests drive alert, alertcondition,
plot and table.new through transpile_full and the glue, and compare the
generated C++ and every other diagnostic field with the parent head's.
The new tests have not been run; they run on Spot.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The four cases of test_the_sentence_is_the_only_change_to_the_parent_heads_output transpiled alert, alertcondition, plot and table.new with commit 5a252e7 and compared the C++ and every diagnostic field with that head's. The commit is an intermediate one that a squash landing does not keep, so reference_codegen() would have nothing to archive and the cases would fail on main. They go with NO_EFFECT_PARENT, the _but helper and the _e2e import, which only they used. The four real-call tests, the alias test and the comparison of the interface cases with the pre-migration output stay. The notes tests collect 153 cases instead of 157, so the whole collection is expected to be 9419 instead of 9423 with the same ten known skips; those counts are static expectations, not a run. PUBLIC_CONTRACT.md and CHANGELOG.md said that the template of every code but the eleven migrated ones is checked for form, not for wording. The tests check more, and both now say what: delta.json fixes the eleven strings, every template is checked for form, each of the 97 warning twins has a sentence of its own that says PineForge warns (the 91 that warn only in a switch arm also say the arm keeps its lowering), and no warning or note template says PineForge refuses or rejects something, though one may say what TradingView does. Other exact wording is not pinned. README.md gets the contract's caveat: text that says "since 1.5.0" describes the source tree on main, 1.5.0 is the planned next release, and the statements about 1.4.0 keep their historical meaning. Only these four files change; the catalog, the product code, the fixtures, the pins, VERSION and the package inputs are the parent's. No test was run for this commit. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
luisleo526
added a commit
that referenced
this pull request
Oct 9, 2026
* docs: state the report-ownership contract for the paired engine fix Add "A report belongs to the call that returned it" to the public contract, a pointer in the README's run section and an Unreleased changelog note. A call the engine refuses without beginning gets the empty report, never an earlier run's rows; a run that began and then failed keeps its partial rows; the explicit stream snapshot keeps the owner's rows; recorded outputs keep their own record of the last run that began. Callers read the error text, the error code and the run status, because a refusal before the engine admits the begin leaves the status at its earlier value. A batch or reentrant begin over a live stream latches Failed and a repeated stream begin does not; a configure time calendar failure leaves the run-number high-water where a failure inside the begin has already moved it. The text is framed as the contract of an engine fix that is not yet released: it names no release number and states that engine v1.4.0 does not carry it. Markdown only; no source, fixture, version, lock or workflow file changes. Rebased onto main 6c8eadb (#193), which added the 1.5.0 diagnostic notes. The changelog note now sits under its own "Planned 1.4.0-line patch — Unreleased" heading after main's unchanged "1.5.0 — Unreleased" section, and its opening is scoped to this documentation change: it adds no Python, JSON, diagnostic or emitted-C++ change, keeps the 1.5.0 diagnostics separate and names no tag, date or version. The contract's availability paragraph now says the report fix itself changes neither the generated C++ nor the transpiler's Python and JSON contract, so it does not contradict the "since 1.5.0" text beside it. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * docs: move the planned patch notes under 1.5.0 The report-ownership notes were drafted as a planned 1.4.0-line patch. The paired engine's report fix is planned for 1.5.0, so the notes now sit in a Documentation subsection of the 1.5.0 entry, and the entry's opening names that scope. The codegen change stays diagnostics only. Only CHANGELOG.md changes: README.md and docs/PUBLIC_CONTRACT.md are byte-identical to the reviewed commit 531755a. Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What this adds
Diagnostics only. No analyzer or code-generation file changes, and the C++ the transpiler emits is byte-identical for every corpus script.
note, a third diagnostic severity. A diagnostic is now anerror, awarningor anote. A note is nonfatal like a warning: a script that produces only notes still returnsok: trueand its C++.user_messageon every catalog entry (629 codes). A short, self-contained sentence a product can show to an end user, written as a template with named arguments. The existingmessage,explanationandhintfields are unchanged.docs/PUBLIC_CONTRACT.md), with a migration call-out in the changelog.Compatibility
A consumer that treats every severity other than
warningas an error will show the eleven notes as errors until it handlesnote; the changelog's migration section says so. The catalog's set of codes and their order are unchanged. Package version and dependencies are unchanged; the changelog entry is filed under the planned 1.5.0, still unreleased.Evidence
🤖 Generated with Claude Code
https://claude.ai/code/session_01Ko17oHtRco8iZ9vJptBAj8