Analyser: fixed an issue with unit cancellation.

[agarny]

Fixes #1485.

[Copilot]

Copilot review overview

🟡 Changes recommended

The cleanup discards small nonzero exponents before later powers can amplify them, producing incorrect dimensional results.

Review effort: Balanced
Findings: 1

Open (1)

• Preserve small unit exponents before POWER amplification · New

What changed in this PR

Addresses #1485 by removing cancelled unit entries so the analyser recognises dimensionless products regardless of operand order.

Changes:

• Centralises unit-exponent accumulation and cancellation.
• Adds regression coverage for three operand orderings.

File	Description
tests/​analyser/​analyserunits.cpp	Tests dimensionless products across operand orderings.
src/​analyser.cpp	Adds and uses the cancellation helper.
src/​analyser_p.h	Declares the private helper.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Copilot review overview

🔵 Needs a closer look

The cancellation check discards legitimate tiny accumulated exponents, causing incorrect units after power operations.

Review effort: Balanced
Findings: None

Resolved since last review (1)

• Preserve small unit exponents before POWER amplification

Previously missed (1)

In code that hasn't changed since last review

Preserve genuine tiny exponents instead of dropping dimensions

src/​analyser.cpp:1133

This check erases genuine tiny exponents even when nothing cancels. In tinyExponentAmplifiedByPower, adding <unit units="second" exponent="0"/> after the existing 1e-16 entry deletes the second dimension, so the later power appears dimensionless instead of having units of seconds. Reversing those entries preserves the dimension. Two 1e-16 entries also disappear instead of becoming second^2 after the power. Use a cancellation rule that preserves meaningful small accumulated exponents rather than treating every near-zero sum as cancellation, and extend the regression tests to cover zero and same-sign contributions in both orders.

[Copilot]

Copilot review overview

🟡 Changes recommended

Numerical edge cases can erase genuine dimensions and incorrectly accept dimensional expressions as dimensionless.

Review effort: Balanced
Findings: 3

Open (3)

• Prevent overflowed exponents from being canceled as infinity · New
• Preserve low-order exponent contributions before cancellation · New
• Preserve genuine nonzero residual exponents during cancellation · New

[Copilot]

Copilot review overview

🔵 Needs a closer look

The cancellation threshold can erase an exactly representable nonzero exponent and incorrectly classify units as dimensionless.

Review effort: Balanced
Findings: None

Resolved since last review (3)

• Preserve genuine nonzero residual exponents during cancellation
• Preserve low-order exponent contributions before cancellation
• Prevent overflowed exponents from being canceled as infinity

[agarny]

@FinbarArgus, this should fix it.

[agarny]

@FinbarArgus, this should fix it.

I mean, fix issue #1485.