[matt-edmondson]

Fixes #134

Problem

TryFormat, which every ToString(format, provider) goes through, threw a bare FormatException for any format other than empty or G. So $"{x:F2}", string.Format("{0:N2}", x), logging templates, and generic INumber<T> code that formats values all crashed at runtime.

Change

The new PreciseNumber/PreciseNumber.Formatting.cs renders the standard formats from the exact significand and exponent. Nothing goes through double, and rounding is half away from zero, the same as Round.

Format	Behaviour	Example (1234.5678)
F<n>	Round(n), padded to exactly n fractional digits. Default n is the culture's NumberDecimalDigits	F2 → 1234.57
N<n>	As F, plus the culture's group sizes, group separator, and NumberNegativePattern	N2 → 1,234.57
E<n>	One digit, n fractional digits, a signed exponent of at least 3 digits. Default n is 6, and e gives a lowercase marker	E5 → 1.23457E+003
R	The exact general form, the same as G	1234.5678
G<n>	ReduceSignificance(n), then the general form. G0 keeps every digit	G5 → 1234.6

• Unsupported formats: custom formats such as 0.00 or #,##0, C, X, R with a precision, and precisions above 999,999,999 throw a FormatException. Its message names the format and lists the supported ones. The issue allowed either this or a fallback to G. I chose the exception because a silent G fallback would hand back unrounded digits to someone who asked for two decimals.
• Very small values: E writes 1.00E-1000000 directly, with no million-character fixed-point expansion.
• Buffers: TryFormat returns false when the destination is too small for a standard format. ToString takes these formats straight to the renderer, so it never comes back empty because its buffer guess was too small.
• Size limit: a rendering longer than a string can hold still throws OverflowException, as the existing G path does.

README.md documents the supported formats.

Tests

PreciseNumberTests gets new and updated cases:

• TestToStringStandardFormats has 25 data rows covering F/N/E/R/G<n>, default precisions, carries (999.995 → 1,000.00, 9.9999 → 1.00E+001), zero, small negatives, and lowercase e.
• An interpolated {x:F2} and string.Format("{0:N2}").
• Custom separators, group sizes [3, 2], and negative pattern 0. These use a cloned NumberFormatInfo, because the test project runs in invariant-globalization mode.
• E2 at an exponent of -1,000,000.
• TryFormat returning false when the destination is too small.
• Descriptive FormatExceptions for unsupported formats.
• TestToStringVariousFormats and TestTryFormatVariousFormats previously asserted the old throw. They now assert the rendered text.

Verification:

• With the library change reverted, 39 tests fail. With it applied, all 512 pass on net10.0.
• The library builds clean on net7.0 through net10.0.

🤖 Generated with Claude Code

https://claude.ai/code/session_01BPNa479nM4FXr2G22kBHmy

---

Generated by Claude Code

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
96.3% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud