Fail loudly on graphs beyond the 32-bit node range

DuckDBSelfHits() coerced nnode (and supplied node ids) with as.integer(),
which silently produced NA above 2^31 and corrupted graph reconstruction at
read time. It now errors with an explicit message when nnode or a node id
exceeds the 32-bit integer range; graphs with more than ~2.1e9 nodes are not
yet supported (the node-id slots are 32-bit). Behavior within the 32-bit range
is unchanged.

This closes the read-side half of the > 2^31 hardening: the write path can now
emit a > 2^31-edge graph index (BiocDuckDB writeParquet index_max), so a huge
node count must fail at construction rather than silently become NA.

Bump version to 0.99.4; update NEWS.

Author: aboyounp-gne

DESCRIPTION

@@ -1,5 +1,5 @@
 Package: DuckDBDataFrame
-Version: 0.99.3
+Version: 0.99.4
 Date: 2026-07-22
 Title: DuckDB-Backed DataFrame and Table Structures
 Description:

NEWS.md

@@ -1,3 +1,12 @@
+# DuckDBDataFrame 0.99.4
+
+## Bug fixes
+
+- `DuckDBSelfHits()` now fails loudly when `nnode` (or a supplied node id)
+  exceeds the 32-bit integer range, instead of letting `as.integer()` silently
+  coerce it to `NA` (which corrupted graph reconstruction). Graphs with more
+  than ~2.1e9 nodes are not yet supported; the error says so explicitly.
+
 # DuckDBDataFrame 0.99.3
 
 ## Bug fixes

R/DuckDBSelfHits-class.R

@@ -448,6 +448,14 @@ function(conn, from, to, nnode, mcols = NULL, keycol = NULL, dimtbl = NULL, node
         stop("'nnode' must be a single non-negative integer")
     }
 
+    # Node ids are held in 32-bit integer slots. Fail loudly on a > 2^31-node
+    # graph rather than letting as.integer() silently coerce nnode to NA (which
+    # would corrupt reconstruction at read time).
+    if (nnode > .Machine$integer.max) {
+        stop("'nnode' (", format(nnode, scientific = FALSE), ") exceeds the ",
+             "32-bit integer range; graphs with more than ",
+             .Machine$integer.max, " nodes are not yet supported.")
+    }
     if (!is.integer(nnode)) {
         nnode <- as.integer(nnode)
     }
@@ -457,6 +465,10 @@ function(conn, from, to, nnode, mcols = NULL, keycol = NULL, dimtbl = NULL, node
     } else if (!is.numeric(nodes)) {
         stop("'nodes' must be a numeric vector")
     } else {
+        if (any(nodes > .Machine$integer.max, na.rm = TRUE)) {
+            stop("'nodes' contains ids beyond the 32-bit integer range; node ",
+                 "ids above ", .Machine$integer.max, " are not yet supported.")
+        }
         if (!is.integer(nodes)) {
             nodes <- as.integer(nodes)
         }

tests/testthat/test-DuckDBSelfHits.R

@@ -351,3 +351,13 @@ test_that("isolated node subset produces empty result for a DuckDBSelfHits", {
     expect_identical(nnode(sub), 1L)
     expect_identical(nrow(as.data.frame(sub)), 0L)
 })
+
+test_that("nnode / node ids beyond the 32-bit range fail loudly (not silent NA)", {
+    expect_error(
+        DuckDBSelfHits(selfhits_parquet, from = "from", to = "to", nnode = 3e9),
+        "32-bit")
+    expect_error(
+        DuckDBSelfHits(selfhits_parquet, from = "from", to = "to", nnode = 5L,
+                       nodes = c(1, 3e9)),
+        "32-bit")
+})