Refuse eleven function names the library does not have (#733)

[Rafael-SOWNet]

Closes the remaining defect half of #733. The feature half — actually having these functions — is not this.

What was wrong

A name the grammar does not know, followed by a bracket, is the implicit multiplication that lets a(b + c) mean a * (b + c). For a one-argument call that never fails, so every CAS spelling AngouriMath lacks came back as the product of an undeclared variable with its argument, silently:

"floor(x)".ToEntity()               // floor * x
"floor(x) - 3 = 0".Solve("x")       // { 3 / floor }  — a root of nothing

The same absence was already reported for two arguments, and only reported badly:

"min(x, y)".ToEntity()              // UnhandledParseException: no viable alternative at input '*('

So whether a missing function was invisible or merely cryptic depended on how many arguments the caller happened to pass, and neither told them the function does not exist.

What the fix does

Eleven names — floor, ceil, ceiling, round, trunc, min, max, gcd, lcm, erf, conjugate — are refused by name, exactly as arcsinh already is, with a message that says what happened:

there is no function floor: AngouriMath has no rounding functions. It is refused by name rather than read as the product of a variable named floor with its argument, which is what an unknown name followed by a bracket would otherwise mean

Nothing is being removed — none of these was ever there. This turns a silent wrong answer into a parse error, which is the whole of it.

Refusing every unknown name is not an option, since that rule is what a(b + c) is. So these are refused one at a time and nothing else moves:

written	before and after
min + 1, max * 2, floor + floor	ordinary variables, unchanged
minimum(x), rounded(x), maxx(y)	still the implicit products
a(b + c)	still a * (b + c)
sgn, sign, signum, abs, gamma, factorial, exp, log10, log2, arsinh, phi	unchanged

re and im are deliberately not refused. They are the sympy spellings for the real and imaginary parts and misparse identically, but two letters is short enough that a caller may reasonably have a variable of that name, and refusing would break their expression to fix nobody's. Pinned by a test so the choice is visible rather than an oversight.

Measured

• Suite 4979 → 5004 passed / 0 failed, F# 130/130.
• Corpus 112/117, 0 wrong / 0 error / 0 timeout, every verdict and answer byte-identical.
• work/rootcheck 595/596; its one incomplete case is A power of a polynomial is solved by inverting into itself, so the roots come back containing x #744, fixed in Only invert an equation whose variable occurs once (#744) #745, not here.

Grammar regenerated with Sources/Utils/antlr_rerun.bat, so the generated files under Core/Antlr are part of the diff.

Recorded in BREAKING-CHANGES.md — it is a loud break and that file's rule is that a call which used to succeed and now does not gets an entry — and in Docs/Usage/Syntax.md, which is a reading of the grammar.

🤖 Generated with Claude Code