feat(language): a bound is null where it is open, never infinite, in the file and in the program

[FBumann]

Prompt: I think the default in upper and lower should be null instead of inf

Prompt: Shouldn't the engine (specsolve) handle this? We can drop inf fully…? Or why do we need to keep the constant inf in program?

Note

The following content was generated by AI.

A bound is a finite number, a parameter name, or null, and null is the default. An infinite bound is refused, and the refusal names null. The program holds an open side as None, not as Constant(±inf). The engine chooses what stands for it in a solve: fluxopt/specsolve#1737.

What this changes

• File. BoundsBlock.lower / upper are float | str | None = None. upper: null loads; it was refused. An infinity is refused: bounds.upper is inf, and a bound is finite. An open side is null: write upper: null, or leave it out.
• Program. VariableDeclaration.lower / upper are Expression | None = None. Lowering passes None through.
• Readers. The unboundedness advice, the typesetter's domain line and the separability walk ask whether a side is None, where they compared it with an infinite constant.
• Removed. sos.coefficients loses its infinity comparisons, _is_absent its infinite-bound branch, and boundedness its _OPEN table.
• Schema regenerated. declarations.md states the default and the refusal. Golden output unchanged.

Why

• One spelling of "absent". Every other field a file may leave open (where, description, dims on a named expression, activity, points) accepts null and defaults to it. Bounds were the one exception.
• Merge and override. The composition functions of feat(language): a file says what it reads, and fragments and patches compose into one model #691 lay a patch over a base, and feat(language): merge and override load every model they compose and return one #690 makes a null in a patch mean "absent". On a declaration it removes the declaration. On a field it drops the field, which then takes its default. With an infinite default, bounds: {upper: null} in a patch built a model the schema refused, so a patch could not open a bound. With null as the default, the value a patch writes and the default it falls back to are the same. null means absent in a file and in a patch alike.
• Infinity is a solver's spelling. The program states what the file means, and "no bound" is not a number. The readers here asked "is it the infinite constant?" to mean "is it open?", and is None says that directly. linopy wants ±inf, and a relational sink may want another value; that choice is the engine's.
• A bound no value meets now fails at load. lower: .inf with the upper side open loaded before, because only two literal bounds that cross were refused. Refusing every infinity closes that gap with no rule of its own.

Tests, guards, gates, what follows

Tests.

• test_an_open_bound_is_null_in_the_file_and_in_the_program covers upper omitted and null. Each case asserts the model holds None, the program holds None, and to_dict() writes no upper. On main, null was a SchemaError and the program held Constant(inf).
• test_an_infinite_bound_is_refused_with_the_null_that_opens_a_side covers .inf and -.inf on each side. On main all four loaded.

Guard. The infinity refusal was deleted and the suite run; all four refusal cases fail. Restored.

Coverage moved.

• infinite-bounds-that-cross in test_a_rule_decided_without_data is gone, because the refusal above now covers both of its infinities.
• test_advice.py states an open lower bound as None, where it wrote -inf and rewrote the JSON to -.inf.

Gates run. pixi run lint passes. pixi run test passes. mkdocs build shows no warnings except one: this sandbox cannot fetch the docs.python.org inventory, which fails --strict. render-tex rendered 31 models.

Not run. compile-tex: the tectonic bundle download is blocked here. CI runs it.

Breaks a consumer. VariableDeclaration.lower / upper may now be None, which a consumer reading them has to handle. fluxopt/specsolve#1737 does that for both of its lanes and passes on its current pin. It lands before the pin moves to a release that carries this.

[read-the-docs-community]

Documentation build overview

📚 math-spec | 🛠️ Build #34746946 | 📁 Comparing f0d4533 against latest (9c7c0a5)

  🔍 Preview build  

5 files changed · ± 5 modified

± Modified

• reference/language/declarations/index.html
• reference/math_spec/boundedness/index.html
• reference/math_spec/model/index.html
• reference/math_spec/program/index.html
• reference/math_spec/sos/index.html