Qt Trading Codegen Drift Remediation

All three history dialogs in ores.qt.trading use an old hand-struct pattern:

struct HistoryResult {
    bool success;
    std::string message;
    std::vector<T> versions;
};

The current template (cpp_qt_history_dialog.cpp.mustache) generates:

using HistoryResult = std::expected<SomeHistoryResponse, std::string>;

BookHistoryDialog.cpp additionally had the wrong response field name (->versions instead of ->books), causing a compile error fixed in-place as a minimal unblock. PortfolioHistoryDialog.cpp and BookStatusHistoryDialog.cpp compile correctly but diverge from the template.

ClientBookModel.cpp set request.offset and request.limit on get_books_request, which no longer has those fields. The template was already updated to gate these assignments on {{#domain_entity.qt.has_pagination}}. The book entity model has no has_pagination: true flag, so regeneration drops those lines. Fixed in-place as a minimal unblock.

In cpp_qt_history_dialog.cpp.mustache, the has_uuid_primary_key branch does:

request.id = id;

where id captures id_, which is boost::uuids::uuid, but request.id is std::string. This has never caused a build failure because no UUID-primary-key dialog has been regenerated with the new template. It will fail on next regeneration of BookHistoryDialog or PortfolioHistoryDialog.

Three trading domain entity models (lifecycle_event, trade_identifier, trade) were created without a cpp.table_display array. The template cpp_domain_type_table.cpp.mustache iterates over this array to emit column headers and row data. With the array absent, the template generated stub convert_to_table bodies — empty header row, empty data rows — causing every table-based unit test to fail with !table.empty().

Fixed in-place by adding cpp.table_display to all three models and updating the three generated *_table.cpp files to match. As part of the proper codegen remediation, regeneration must be verified to produce identical output to confirm the model entries are correct.

The generator template cpp_domain_type_generator.cpp.mustache emits:

r.tenant_id = utility::uuid::tenant_id::from_string(tid_str).value();

where tid_str defaults to "system" when no tenant_id key is present in the context. "system" is not a valid UUID, so from_string returns an unexpected value and .value() throws bad access to std::expected without expected value. Every unit test that calls a generator with a bare generation_context{} crashes.

Fixed in template and all 23 affected generated *_generator.cpp files:

r.tenant_id = utility::uuid::tenant_id::from_string(tid_str)
    .value_or(utility::uuid::tenant_id::system());

As part of the proper codegen remediation, all 23 generator files must be regenerated from the corrected template and verified to compile and pass their unit tests.

Regenerate all 23 *_generator.cpp files that had the from_string().value() call. Confirm they compile and that unit tests calling generators with a bare generation_context{} pass without exception.

Affected components: ores.reporting.api, ores.refdata.api (11 files), ores.trading.api (generators + generator directories, 11 files).

Regenerate:

• ores.trading.api/src/domain/lifecycle_event_table.cpp
• ores.trading.api/src/domain/trade_identifier_table.cpp
• ores.trading.api/src/domain/trade_table.cpp

Confirm output is identical to the in-place fix and that the table unit tests in ores.trading.api/tests/ pass.

The old wrapper-struct pattern likely exists in other Qt components that predate the template migration. Run:

grep -rln "struct HistoryResult" projects/ores.qt.*/src/

File any found as a follow-up story rather than blocking this work.

Other domain entity models may also be missing cpp.table_display, causing silent empty-table output. Run:

grep -rL "table_display" projects/ores.codegen/models/**/*_domain_entity.json

Cross-check against entities that have table unit tests to prioritise fixes.