[jdhughes-dev]

MODFLOW 6.8.2 reads SFR packagedata through IDM and keeps HK, ROUGH, and USTRF only as deprecated API aliases, so every SFR model run through modflowapi writes three deprecation warnings to mfsim.lst. The packagedata hk, rough, and ustrf columns are now read from __INPUT__/<model>/<sfr>/BEDK, MANNING, and UPSTREAM_FRAC when libmf6 has those addresses, and from the legacy names otherwise; the column names are unchanged.

• adv_input_vars in datamodel.py maps a column to its input context name.
• test_sfr_input_context_names checks the addresses, values, and a write that persists into the next stress period, against a small SFR model whose input files the test writes as text; it passes against 6.8.1 and current MODFLOW 6 develop.

[jdhughes-dev]

Reviewer Summary

• Why the input context: on MODFLOW 6 develop, this%hk, this%rough, and this%ustrf are pointers to BEDK, MANNING, and UPSTREAM_FRAC in the input context (gwf-sfr.f90:567-569). The package path holds only the deprecated aliases from mem_alias_deprecated (gwf-sfr.f90:742-747), which point to the same memory.
• No change in results: packagedata, period data, and diversions values read through modflowapi match between 6.8.1 and develop for the SFR, LAK, MAW, and UZF examples in modflow6-examples. Final heads agree to about 1e-11, with and without writes to SFR hk and rough and UZF finf.
• Warnings removed: with this change, develop writes no API VARIABLE ... WAS DEPRECATED warnings to mfsim.lst; without it, every SFR model writes three.
• Column names unchanged: hk, rough, and ustrf stay the recarray column names for every libmf6 version, so existing scripts are unaffected.
• Address selection: the input context address is used when it is in get_input_var_names(), read once per SFR package, so a pre-6.8.2 nightly (6.9.0.dev0) falls back to the legacy names. If neither address exists, get_value_ptr raises XMIError (unknown variable); _set_data only catches InputError for unsupported types, so a missing variable is not dropped silently.
• Other packages: the LAK, MAW, and UZF names modflowapi reads are not aliased on develop. The transport advanced packages (SFT, LKT, MWT, UZT) are not mapped in modflowapi, so their aliases do not apply.
• Adding renames: a later rename is one entry in adv_input_vars in datamodel.py.

[mjreno]

Verified

• On MODFLOW 6 develop, this%hk, this%rough and this%ustrf point at BEDK, MANNING and UPSTREAM_FRAC in the input context, and HK, ROUGH and USTRF are deprecated aliases of the same memory, so the new addresses read and write exactly what the old ones did. These arrays are indexed by reach, so PACKAGEDATA row order doesn't matter. Using them rather than the PACKAGEDATA input arrays (RHK, MAN, USTRF) is correct.
• Behavior with MODFLOW-ORG/modflow6#3093 (in 6.8.2): a value written through modflowapi persists unless that reach's PACKAGEDATA value is linked to a time series, in which case the series value replaces it at the next time step.
• test_sfr_input_context_names passes against libmf6 6.8.1 and against current develop (which includes #3093).

Minor suggestions

1. libmf6_version(): re.match returns None for a version string it doesn't recognize, which surfaces as an AttributeError. A clear error message would help.
2. The test writes hk and immediately reads the same address back. Reading it again at the next time step, or comparing with the deprecated HK address (same memory), would show the write reaches the package's working array and persists, which the manual comparison in the summary already demonstrates.

[jdhughes-dev]

Thanks @mjreno.

1. libmf6_version() now raises ValueError: unrecognized libmf6 version: '<version>' when the version string doesn't match (ca2438b).
2. test_sfr_input_context_names now reads hk again at the start of the second stress period and checks the written values persisted (7be056a). I didn't compare against the deprecated HK address so the test doesn't trigger the deprecation warning.