refactor(checker,getter): declare Call and StringLit; fold the alterator and find/count consumers #1698

Activity

1. dekobon

   added

   refactorCode maintainability / tech-debt cleanup (no behaviour change)Code maintainability / tech-debt cleanup (no behaviour change)

   on Oct 23, 2026
2. dekobon

   added a parent issue on Oct 23, 2026

   • refactor(ast): one role declaration per language for Checker, Getter and metric dispatchers #1690
3. dekobon

   marked this as blocked on Oct 23, 2026

   • refactor(abc,cyclomatic): declare Comparison and ShortCircuit in Java, Python and C++ with every consumer — go/no-go #1694
4. dekobon

   marked this as blocking 2 issues on Oct 23, 2026

   • test(parity): derive cross-consumer parity from the role tables beside the hand-written suites #1700
   • docs(ast): document the role tables and close out #1701
5. dekobon commented on Oct 23, 2026

   

   dekobon

   on Oct 23, 2026

   OwnerAuthor

   More actions

   Tracking is ready

   The body now links the master issue and the work packages it references. Native parent and blocked-by relationships match the listed completion prerequisites. Implementation and acceptance checkboxes remain open; prerequisite closure and the R04: #1694 go decision govern execution order.
6. dekobon commented on Oct 25, 2026

   

   dekobon

   on Oct 25, 2026

   Last edited by dekobon

   OwnerAuthor

   More actions

   Done — Call and StringLit declared in every language; is_call / is_string, the getters' interpolation sets, the dump's string flattening and the ABC branch count read the rows

   Branch refactor/1696-1698-recognition-roles from 7425561f (main), commits c168daae (R08) and f58ade41 (the C alterator arm the scripted pass missed); ten R06 commits before them. Not pushed, no PR.

   Rows. Call and StringLit in all 25 languages (roles/<lang>.rs), with shared fragments c_family_calls! / c_family_strings!, js_family_strings!, tcl_family_strings! (+ tcl_family_string_gates!, emitted per dialect over its BracedWordKinds). The two gates the issue named are gated_by entries: Perl's CallExpressionWithBareword(gated_by bareword_is_call) (= perl_bareword_is_call, the outermost-dispatch and not-a-hash-key rule, derivation on the row) and the Tcl family's BracedWord(gated_by is_braced_literal) (= !Getter::is_braced_script_word, #1381/#1382). wraps holds each getter's former interpolation set — PHP's direct-child kinds and {, Ruby's array element kinds, Perl's list_item, Groovy's two GString forms, the Tcl family's substitutions. excludes: PHP String2, TS String2, TSX String3 (type keywords, #1261/#1474). Every call-site rationale from the six impl_simple_is_string! / impl_js_family_is_string! sites and the per-language is_call comments moved onto the rows (macro-comments rule).

   What the rows now record about the pinned grammars (each probed with bca dump, each a defensive owner the tests hold unseen): C/C++/Mozcpp/ObjC's unsuffixed CallExpression (#1254); TypeScript's and TSX's CallExpression2..4 — a decorator call, a typeof f() in a type and a generic call all arrive as the first id; C#'s InvocationExpression / InvocationExpression2 — bare, member, generic and nullsafe calls in a class, struct, record, interface and field initialiser all carry the third; Ruby's Call / Call2 / Call4 — ten call spellings all carry Call3, Call5 is the hidden _call; the JS family's named String — every literal, a module string, a property key and an import alias included, is the String2 alias; PHP's hidden String3; Bash's three heredoc aliases. The issue's "csharp_invocation_expr_kinds! folds in" is done by listing the three ids on the row; the macro's one remaining user is a comment.

   Consumers. Checker::is_call / is_string and the _with_code pair come from impl_call_row! / impl_string_lit_row! (two macros: the bypass gate's row-forwarding recogniser binds one kind to one row). The _with_code spellings are the rows under their gates; the byte-less ones are the row's kind table through a new lists_kind — not a known-empty chain, which here would read a Tcl body as a script, and the existing tcl_script_body_is_a_string_only_to_the_byteless_spelling pins that the two still disagree. Perl's is_call_with_code and the Tcl family's is_string_with_code overrides are gone. Every getter's string_operand_type(node, &[…]) reads string_lit::WRAPS; Bash's expansion read, PHP's direct-child set (the heredoc_body descent stays) and Kotlin's long form read it too (Kotlin's short $name has no node and stays a byte read). The dump's string flattening reads the row's owners through a shared Alterator::flatten_verbatim; the non-string flattening — char / rune literals, Groovy's fragments, PHP's string keyword alias with its #1474 / grammar-dispatch §6 rationale — stays each language's own arm beside it, and the Tcl family's keeps_children veto is untouched; tests/grammars/alterator_string_flattening.rs is unchanged and passes. The ABC branch count reads call::owns_kind in every language, keeping object construction as the walker's own (NewExpression, ObjectCreationExpression, ConstructorInitializer, ConstructorDelegationCall, Groovy's New, Java's New / ExplicitConstructorInvocation, Rust's TryExpression). Java's and Elixir's cognitive boundary / lambda arms read their rows as the other languages' do since R06.

   Exemptions, each naming the role and the question it is not: the operand classification of string kinds in every getter (a projection the row does not answer), PHP's { / variable-name interpolation reads, get_operator_id_as_str's {} spelling, the exit-name reads on call kinds in nexits (Go, Lua, Bash, iRules, Perl, Elixir), Elixir's cognitive / cyclomatic keyword dispatch on call, Perl ABC's argument-shape helpers, Go's boolean-slot leaf, Ruby's symbol-name readers, Java ABC's return / lambda slots, Python's yield operator. Bypass baseline 40 → 10, all Elixir's documented byte-less kind table (#275) and impl_exit_match_kinds! (R07's Exit). Census 314 → 274, coverage −140, every --update remove-only.

   Tests. roles/call_string_tests.rs: per language, (is_call_with_code, is_string_with_code) of every call / string node by text in document order — every other node asserted neither — with every owner covered and the defensive aliases asserted unseen. Fixtures spell the gated cases: Perl's nested callees and the name => / -key hash keys (the latter the file-test misparse of #1545/#1572), the Tcl family's proc / if / when bodies against braced values, PHP's string keyword and (string) cast, Bash's $"…" in argument and assignment position (#1358), Ruby's safe-navigation call and %w / %i elements. elixir_anonymous_function_contributes_lambda_nesting and c_string_and_concatenated_string_flattened were added where a perturbation found a consumer unguarded.

   Evidence. Zero drift against the base binary: 25 populations, CLI ops / metrics / dump / find -t call|string|function / count plus the library-level verbatim record (raw bytes, Alterator export, find, count): 0 of 26,992 outputs differ. Recovery sweep over every fixture (each line, 16 offsets, after every operator punctuation, every word boundary; CLI + verbatim): 129,042 outputs, 0 differing, 0 failed executions. Perturbations, each failing at least one named test: 25 row edits (a gate dropped, an emitted alias dropped, a defensive alias dropped, a wraps entry dropped, PHP's keyword moved into owns — which the DSL refuses to build) and 12 consumers severed. make pre-commit: BCA_GATE: pass; make chain-audit passes; make bench-scaling 87 of 87; single-feature builds (rust / go / elixir / tcl / kotlin lib, rust / tcl / perl ast) green; check-test-lang-gates-compare COMPARE_REF=origin/main loses no build; both snapshot forms clean; the output submodule untouched.

   Review (astra-review, fresh context, checkpoint 4 on R08 and on the combined 7425561f..HEAD diff): no actionable findings; the reviewer independently ran both library suites (4,411 passed), the role tests and the three role gates. One thing the review did not catch and the issue's own checklist did: the C alterator arm was left unmigrated by the scripted pass (it had no test of its own — the C++ tests cover the C++ arm), found when ticking "the alterator reads the row for its string half"; migrated, tested, re-measured (0 drift on the C-family populations).

   Limitations. The byte-less is_call / is_string are the kind tables and say so; the three gated rows (Perl bareword, Tcl / iRules braced word) are the only places the two spellings differ, each pinned. Kotlin's short-form interpolation cannot be a wraps entry (no node). PHP's String2 is excluded by the row and listed by the dump arm, for the reasons on both.
7. dekobon

   closed this as completedon Oct 25, 2026
8. dekobon commented on Oct 26, 2026

   

   dekobon

   on Oct 26, 2026

   OwnerAuthor

   More actions

   Post-landing review pass (fixes in 18ee3c98)

   Two corrections to what this issue's close-out claimed, found by the acceptance-criteria audit:

   • "Declared in all twenty-five languages" was two short. Ccomment's and Preproc's StringLit rows were still pending R08 and their checkers still read impl_simple_is_string!. Both now declare StringLit { owns [StringLiteral, RawStringLiteral] } — the literals the pass must not strip — and read impl_row_predicates!; impl_simple_is_string! is gone, and the census / coverage baselines drop their last R08 cells and the four ids. find -t string on both helper grammars is byte-identical to before.
   • Perl's bareword call is gated_by bareword_is_call, not text_gated_by as step 1 says: the gate reads the parent off the chain and no bytes, so gated_by is the correct keyword. The byte-less is_call is the row's kind table either way.

   Also from the same pass: every language's is_primitive now reads the Primitive row's kind table, with the text-key selection for C# modifiers and Kotlin's reified moved into keys_operator_by_text overrides (they were still answered by is_primitive under the precondition #1599 deleted). Evidence on HEAD: 0 differing of 27,008 corpus records compared, 0 of 124,488 recovery records; the counts in the earlier comments ("26,992") were a miscount of the same 27,024-record runs.
9. dekobon

   added 2 commits that reference this issue on Oct 26, 2026

   refactor(checker,getter,alterator,abc): declare Call and StringLit by…

   ...

   b4f0cfe

   refactor(roles,metrics,checker): take the post-landing review passes …

   ...

   ff1047a