Fix S4 method dispatch on imported S3 generics (e.g. setMethod("as.data.table", "Basket", ...))

[Copilot]

setMethod() on an imported S3 generic (like data.table::as.data.table) creates an implicit S4 generic in ns_env. The pkg_env population step was overwriting that implicit generic with the original S3 function from impenv, silently breaking search-path dispatch while internal namespace dispatch worked fine. Additionally, exportMethods()/exportClasses() directives in NAMESPACE were ignored, so exported generics never appeared in getNamespaceExports().

Fixes in R/loadfast.R

• pkg_env copy order reversed (both full and incremental paths): impenv is now the base layer, ns_env applied second so any shadows created during sourcing (implicit generics, deliberate overrides) win over raw imports
• exportMethods/exportClasses now processed: namespace export block extended to include nsInfo$exportMethods and nsInfo$exportClasses alongside nsInfo$exports

# Before — impenv overwrote the implicit generic back to the original S3 fn:
list2env(as.list(ns_env, all.names = FALSE), envir = pkg_env)
list2env(as.list(impenv, all.names = TRUE),  envir = pkg_env)  # ← clobbered

# After — ns_env wins on conflicts:
list2env(as.list(impenv, all.names = TRUE),  envir = pkg_env)
list2env(as.list(ns_env, all.names = FALSE), envir = pkg_env)

Test coverage

• Added Basket S4 class + setMethod("as.data.table", "Basket", ...) to devpackage fixture; as.data.table is imported but intentionally not exported (exportMethods omitted from NAMESPACE)
• Stage 1 checks (full load): method registration via existsMethod(), implicit generic is a genericFunction in both ns_env and pkg_env (not overwritten by the raw imported S3 fn), dispatch via get(), direct calls via get(..., pos = "package:devpackage") and evalq() in pkg_env scope, single-element and empty Basket edge cases, non-Basket data.frame dispatch regression, confirms as.data.table is not in getNamespaceExports()
• Stage 4b checks (incremental reload): same battery of direct-call and edge-case checks after an incremental reload triggered by mutating base.R, plus explicit checks that as.data.table remains a genericFunction in both ns_env and pkg_env after the reload (not reverted to the plain S3 function), confirming the method remains registered and dispatches correctly through all call paths

[felix-andreas-finccam]

@copilot can we also test if normal call works? also for incmrental test. are these new tests exhaustive? can you think of more

basket <- new("Basket", ...)
as.data.table(new()

[Copilot]

Added in e1d67be. Both Stage 1 (full load) and Stage 4b (incremental) now cover:

• existsMethod("as.data.table", "Basket") — method is registered
• Direct call via getExportedValue("devpackage", "as.data.table")(basket) — no get() wrapper
• Direct call via evalq(as.data.table(basket), envir = as.environment("package:devpackage")) — natural call syntax in pkg_env scope
• Single-element Basket
• Empty Basket → 0-row DT with item column
• Non-Basket data.frame dispatch still works (regression guard)

[felix-andreas-finccam]

@copilot remove this. we shouldn't export this method. we only import it. we shouldn't rely on an export

[Copilot]

Done in dcc3a12. Removed exportMethods(as.data.table) from devpackage/NAMESPACE. The export test now asserts that as.data.table is not in getNamespaceExports("devpackage"), and all direct-call tests were updated from getExportedValue() (which requires export) to get("as.data.table", pos = "package:devpackage") (search-path lookup).

[felix-andreas-finccam]

@copilot can we also add a test for implicit generics? (normal an incremental?)

[Copilot]

@copilot can we also add a test for implicit generics? (normal an incremental?)

Added in 387678b. Stage 1 already had the two implicit generic checks (ns_env + pkg_env). Stage 4b now has the same two checks after the incremental reload — verifying as.data.table remains a genericFunction in both ns4b (namespace env) and package:devpackage (search-path pkg env), not reverted to the raw imported S3 function.