[ftgonc]

Summary

collect_references skips local.*, so any dependency that goes through a local value is lost. A resource with compartment_id = local.compartment_id, where the local reads a data source, gets no edge to that data source. Configs that route values through locals (very common) end up with a mostly disconnected resource graph: oxid graph under-reports dependencies, and plan/apply ordering misses real ones.

• Keep local.<name> references, both in expressions and in ${…} interpolations inside nested blocks
• resolve_local_references: resolve each local to the resource and data source addresses it reaches, following locals that reference other locals. Cycles between locals are cut instead of looping
• Expand local references when adding edges for resources, data sources and outputs
• Also collect references from count and for_each, not only from attributes (e.g. for_each = local.endpoints)
• Keep resources whose count/for_each depends on locals, resources or data sources. The DAG builder evaluates these with variables only, so count = length(local.dbs) evaluated to 0 (and toset(local.ids) to an empty set) and the resource silently disappeared from the graph. Such resources are now a single unindexed node, as already happens when for_each evaluates to null; their dependencies are still tracked
• Fix data source labels in to_dot, which were prefixed twice (data.data.x.y) because the address already includes data.

var.*, each.*, count.*, path.*, terraform.* and self are still ignored, as before.

Known gaps (not in this PR)

• lifecycle { precondition / postcondition } blocks aren't parsed (LifecycleConfig has no field for them), so references inside them don't create edges. Terraform does count them
• module.* references are still opaque
• A resource whose count/for_each can't be evaluated at graph time is one unindexed node, so its instances aren't individually addressable in the graph. That's the existing behaviour for unresolvable for_each, now applied consistently; previously it had no node at all

Test plan

• New tests/resource_graph_test.rs: local → data source (direct attribute and map attribute), chained locals through a for expression, for_each = local.x, output referencing a local, cyclic locals (no hang, no edges), var through a local (still no edge), count = length(local.x) and for_each = toset(local.x) keep the resource, count = var.n still expands, single data. prefix in DOT labels
• Full test suite passes (129 tests, 0 failures)
• cargo fmt --all -- --check clean
• cargo clippy --all-targets -- -D warnings: no new findings. Current stable clippy reports 33 pre-existing double_must_use errors in src/state/backend.rs on main as well, untouched here
• Manually on a real config that uses locals heavily (no modules): oxid graph went from 5 to 41 edges, and 34 of the 41 dependencies Terraform records in state are now reachable in the graph (previously almost none). At least one of the remaining 7 comes from a lifecycle precondition. On another config, 5 resources with count = length(local.…) that were missing from the graph now appear. No cycles were introduced on any of the six configurations tried

[chaitanya-sistla]

@ftgonc approved workflows