lake_from_gdf crashes with IndexError on a string lakeout value instead of creating an out-of-model outlet

[bdestombe]

Got an extra set of eyes looking at the code, and it identified a few bugs. I'll leave it here for now, until I find time to pick up the implementation.

Where: nlmod/gwf/lake.py:210-219, in lake_from_gdf (and docstring at :70-71).

Problem. lake_from_gdf treats any string lakeout value as a lake boundname to resolve. An out-of-model outlet is documented to require the integer -1 (flopy zero-based). If the input GeoDataFrame stores the outlet as the string '-1' (or any string that is not an existing lake boundname), the boundname mask is empty and gdf.loc[mask, "lakeno"].iloc[0] raises an opaque IndexError instead of either creating the out-of-model outlet or raising a clear error.

Reproduction. A GeoDataFrame with lakeout='-1' (string) on a lake, passed to lake_from_gdf. (Encountered with an NHFLO lakes_pwn dataset; the data side is being fixed to use integer -1, but the function could be more robust.)

Suggested fix. Coerce numeric strings before the boundname lookup (so '-1' -> out-of-model outlet), and raise a clear KeyError/ValueError naming the unresolved boundname when a string lakeout matches no lake. Possibly related: #500 (representation of empty fields in lake package construction).

[rubencalje]

I think this is solved in PR #562

[bdestombe]

Thanks, the reported case (a numeric string like '-1') works on dev now. Closing. One related case still gives an unclear IndexError: a lakeout that names a lake which is not in the gdf (e.g. dropped because it covers too few cells). A clearer error message there would help; happy to open a small PR for it.