Merge pull request #5 from madsuite-org/anvil/model-selection

Select a model by name in a library carrying several

Author: sshin23

README.md

@@ -100,6 +100,25 @@ compiled from a recipe names its fields `arg1`, `arg2`, ... for that reason.
 Each value is checked against the kind and type the schema declares for its
 slot; nothing is coerced across it.
 
+### Several models in one library
+
+One shared library may export any number of models, each under its own symbol
+prefix with its own schema and its own instances. A leading string argument
+selects one by name — the name **is** the prefix its ABI functions are
+exported under (unambiguous, since a model argument is never a string):
+
+```python
+m = cnlpmodels.CModel("@grid", "acopf", bus, 100.0)  # acopf_* inside libgrid.so
+d = cnlpmodels.CModel("@grid", "dcopf", bus)         # dcopf_* in the same file
+sch = cnlpmodels.schema(lib, "acopf")                # schemas are per model
+```
+
+A mistyped name is refused at selection, with the witness symbol named, rather
+than surfacing as a raw `undefined symbol` several calls later. Omitting the
+name keeps the single-model spelling, where the prefix falls back to the
+library name — a one-model library is unaffected. This is the same selection
+spelling as CNLPModels.jl's `CNLPModel(lib, :acopf, ...)`.
+
 ## Implementing a compatible library
 
 1. Export the functions above with C linkage under one prefix.

src/cnlpmodels/__init__.py

@@ -65,7 +65,12 @@ def _paths():
 def lib(name):
     """Resolve `lib<name>.so` against the search path — also accepting the
     `<dir>/<name>/lib/` and `<dir>/lib/` layouts `compile_library` produces —
-    load it, and cache the handle by name."""
+    load it, and cache the handle by name.
+
+    A leading `@` is accepted and ignored: this function only ever takes a
+    name, and the sigil spelling travels from `CModel`'s string argument."""
+    if name.startswith("@"):
+        name = name[1:]
     if name not in _LIBS:
         ext = {"win32": ".dll", "darwin": ".dylib"}.get(sys.platform, ".so")
         fname = f"lib{name}{ext}"
@@ -155,9 +160,33 @@ def _check(st, what):
         raise RuntimeError(f"{what} returned nonzero status {st}")
 
 
-def schema(lib, *, prefix="rec"):
-    """The library's data schema (ABI v2), as published by `<prefix>_schema`."""
+def _require_model(lib, model):
+    """A name this library does not carry is reported here, clearly.
+
+    `_nvar` is the witness symbol: the ABI requires it of every model however
+    the model is instantiated — unlike `_new` (absent from builder-only
+    models) or `_data_begin` (absent from one-knob ones). Without this check a
+    mistyped name surfaces as a raw ctypes `undefined symbol` error several
+    layers down."""
+    try:
+        getattr(lib, f"{model}_nvar")
+    except AttributeError:
+        where = getattr(lib, "_name", "this library")
+        raise ValueError(
+            f"{where} carries no model named {model!r} "
+            f"(it exports no {model}_nvar)"
+        ) from None
+
+
+def schema(lib, model=None, *, prefix="rec"):
+    """The library's data schema (ABI v2), as published by `<prefix>_schema`.
+
+    In a library carrying several models the schema is per model — name the
+    one you want, `schema(lib, "acopf")`, exactly as in `CModel`."""
     import json
+    if model is not None:
+        _require_model(lib, model)
+        prefix = model
     fn = getattr(lib, f"{prefix}_schema")
     fn.restype = _c_int
     fn.argtypes = [ctypes.POINTER(ctypes.c_uint8), _c_int]
@@ -338,6 +367,16 @@ class CModel:
         m = cnlpmodels.CModel("rosen", 1000)                   # ./rosen (file or bundle dir)
         m = cnlpmodels.CModel("/opt/models/rosen", 1000)       # full path
 
+    One library may carry **several models**, each under its own symbol
+    prefix with its own schema and instances. A leading string argument
+    selects one by name — the name is the prefix, mirroring CNLPModels.jl's
+    `CNLPModel(lib, :acopf, ...)`; unambiguous, since a model argument is
+    never a string. A mistyped name is refused at selection, not as a raw
+    `undefined symbol` several calls later:
+
+        m = cnlpmodels.CModel("@grid", "acopf", bus, 100.0)  # acopf_* in libgrid.so
+        d = cnlpmodels.CModel("@grid", "dcopf", bus)         # dcopf_*, same file
+
     The arguments are the values the model is instantiated with — one per field
     of the library's schema, positionally, in the order the library publishes
     them, which is the same spelling the producer side uses
@@ -354,10 +393,26 @@ class CModel:
     """
 
     def __init__(self, lib, *args, prefix=None):
+        # A leading string argument names a MODEL in a library carrying
+        # several — the name is the symbol prefix its ABI functions are
+        # exported under, mirroring CNLPModels.jl's
+        # `CNLPModel(lib, :acopf, ...)`. Unambiguous: a model argument is
+        # never a string.
+        model = None
+        if args and isinstance(args[0], str):
+            model, args = args[0], args[1:]
+            if prefix is not None and prefix != model:
+                raise TypeError(
+                    f"both a model name ({model!r}) and prefix= ({prefix!r}) "
+                    "were given; they mean the same thing — give one"
+                )
+            prefix = model
         if isinstance(lib, str):
             prefix = prefix if prefix is not None else _default_prefix(lib)
             lib = _resolve_spec(lib)
         prefix = prefix if prefix is not None else "rec"
+        if model is not None:
+            _require_model(lib, model)
         # Instantiate before resolving the evaluation table, so a failure to
         # build the model surfaces as what it is — not as a missing evaluation
         # symbol on a library that never got that far. Same order as the Julia

tests/test_cnlpmodels.py

@@ -292,3 +292,55 @@ def test_evaluation_shape_guards(lib):
         m.obj(np.zeros(3))
     with pytest.raises(ValueError, match=r"y must have shape \(1,\)"):
         m.hess(np.zeros(4), np.zeros(2))
+
+
+# ── Selecting a model by name in a library carrying several ──────────────────
+# The fixture carries several models in ONE shared library; a leading string
+# argument names one, and the name is the symbol prefix — the same selection
+# spelling as CNLPModels.jl's `CNLPModel(lib, :tq, ...)`.
+
+
+def test_model_selection_by_name(lib):
+    x = np.array([0.5, 0.25, 2.0, -1.0])
+    n, s, w = 4, 2.0, np.array([1.0, 2.0, 3.0, 4.0])
+
+    m = cnlpmodels.CModel(lib, "tq", 4)
+    assert m.nvar == 4
+    ms = cnlpmodels.CModel(lib, "sq", n, s, w)   # builder-only sibling
+    assert m.obj(x) == ((x - 1.0) ** 2).sum()
+    assert ms.obj(x) == (w * (x - s) ** 2).sum()
+
+    # Instances of DIFFERENT models coexist as freely as instances of one.
+    m6 = cnlpmodels.CModel(lib, "tq", 6)
+    assert m6.nvar == 6
+    assert m.obj(x) == ((x - 1.0) ** 2).sum()
+    assert ms.obj(x) == (w * (x - s) ** 2).sum()
+
+
+def test_unknown_model_name_is_refused_clearly(lib):
+    # A mistyped name is reported at selection, with the witness spelled out —
+    # not as a raw ctypes `undefined symbol` several calls later.
+    with pytest.raises(ValueError, match=r"carries no model named 'nosuch'"):
+        cnlpmodels.CModel(lib, "nosuch", 4)
+    with pytest.raises(ValueError, match=r"carries no model named"):
+        cnlpmodels.schema(lib, "nosuch")
+
+
+def test_model_name_and_prefix_must_agree(lib):
+    with pytest.raises(TypeError, match=r"give one"):
+        cnlpmodels.CModel(lib, "tq", 4, prefix="sq")
+    assert cnlpmodels.CModel(lib, "tq", 4, prefix="tq").nvar == 4
+
+
+def test_schema_by_model_name(lib):
+    sch = cnlpmodels.schema(lib, "sq")
+    assert [f["name"] for f in sch["fields"]] == ["n", "s", "w"]
+
+
+def test_at_sigil_is_accepted_by_lib(lib, tmp_path):
+    import shutil
+    shutil.copy(pathlib.Path(lib._name), tmp_path / "libtoy9.so")
+    cnlpmodels.set_path(tmp_path)
+    assert cnlpmodels.lib("@toy9") is cnlpmodels.lib("toy9")     # one cache entry
+    m = cnlpmodels.CModel("@toy9", "tq", 4)                      # with model selection
+    assert m.nvar == 4