from_res1d cannot reach EPANET link quantities — no breakpoint for single-gridpoint reaches

[jpalm3r]

Summary

from_res1d can't reach EPANET link quantities, because EPANET-style water distribution links never get a breakpoint to match against.

Root cause

Res1DReach.__init__ (src/modelskill/model/adapters/_res1d.py:77-109) only creates intermediate breakpoints when a reach has more than 2 gridpoints:

intermediate_gridpoints = (
    reach.gridpoints[1:-1] if len(reach.gridpoints) > 2 else []  # line 95-97
)

EPANET-style links typically have exactly 2 gridpoints (start and end only, no intermediate H-points), so intermediate_gridpoints is always empty and Res1DReach.breakpoints is always [] for these links. As a result, NetworkModelResult._extract_reach (src/modelskill/model/network.py:212-281) can never find data to match a ReachObservation against such a link, and always falls through to the ValueError at lines 277-281.

Request

Treat a 2-gridpoint reach as having one implicit mid-reach breakpoint. This fits the existing ReachBreakPoint/BasicReach data model without changing it. A proof of concept for this already exists.

Relevant extension points

• ReachBreakPoint (src/modelskill/network.py:81-138)
• BasicReach (network.py:270-323)

Found by a test user evaluating the network functionality ahead of a 1.4.0 release.

[jpalm3r]

Investigated and confirmed the root cause, summary below.

Confirmed. Res1DReach.__init__ (_res1d.py:95-97) builds break points only from interior gridpoints (reach.gridpoints[1:-1]). A 2-gridpoint reach yields an empty list, so breakpoints == [] and _extract_reach always falls through to the ValueError.

Data model. For a standard MIKE reach, reach.to_dataframe() gives one column per (quantity, chainage): water level at the endpoint H-points (which belong to the start/end nodes), discharge at the interior Q-point. The current code turns that interior Q-point into a break point. An EPANET-style link has only the two endpoint H-points, so the flow quantity has nowhere to attach — hence the bug.

Main blocker: no test fixture exists. The only .res1d fixture (tests/testdata/network.res1d) is a MIKE urban model with a gridpoint-count distribution of {3: 107, 5: 11} across 118 reaches — zero 2-gridpoint reaches. mikeio1d 1.2.1 ships no sample data either. So there is currently no way to reproduce this or write a regression test. A small EPANET-style .res1d is a prerequisite. Could you share the sample file behind your proof of concept? That file is also the ground truth for the one open question below.

Open question the fixture must answer. The fix synthesizes one mid-reach break point, but its data has to come from a real source. Two candidates:

1. reach.to_dataframe() exposes a Discharge:<link>:<chainage> column at an interior chainage → use that column and its real chainage.
2. The flow is a reach-level item with no interior chainage → place the break point at length / 2 and pull the column by quantity name.

I could not tell which holds without an EPANET file; the planned helper is written so this is the only thing left to confirm.

Planned fix. Confined to Res1DReach.__init__ plus a small extraction helper: when len(reach.gridpoints) == 2, build one GridPoint (existing ReachBreakPoint) carrying the link quantity, respecting the existing populate_gridpoints flag. No change to the abstract data model, _extract_reach, or graph wiring — a single-break-point reach already flows through both. Reaches with 3+ gridpoints are untouched, so existing behavior is unchanged.