Summary

Add semver comparison operators for local feature flag evaluation.

Problem

PostHog supports semver-based feature flag targeting in the UI (e.g., show a feature only to users on app version >=2.0.0), but the Rust SDK could not evaluate these flags locally. This forced all semver-targeted flags to fall back to server-side evaluation, adding latency and network dependency.

Related implementations:

• Python SDK: feat: add semver targeting support to local flag evaluation posthog-python#447
• PostHog server: feat(flags): support semver comparisons in user_blast_radius posthog#44596

Changes

• Add parse_semver() function to parse version strings into (major, minor, patch) tuples
  • Handles: v-prefix stripping, whitespace trimming, pre-release/build metadata stripping
  • Supports partial versions (defaults missing components to 0)
  • Ignores extra components beyond major.minor.patch
• Add 6 basic comparison operators: semver_eq, semver_neq, semver_gt, semver_gte, semver_lt, semver_lte
• Add 3 range operators:
  • semver_tilde: patch-level range (~1.2.3 means >=1.2.3 <1.3.0)
  • semver_caret: compatible-with range per semver spec (^1.2.3 means >=1.2.3 <2.0.0, with special handling for 0.x versions)
  • semver_wildcard: wildcard patterns (1.2.* means >=1.2.0 <1.3.0)
• Implement inline (~100 lines) rather than adding the semver crate to minimize dependency footprint for SDK consumers

How did you test this code?

Added 38 new tests covering:

Parsing tests:

• Basic parsing (1.2.3, 0.0.0, 10.20.30)
• v-prefix handling (v1.2.3, V1.2.3)
• Whitespace trimming
• Pre-release/build metadata stripping (1.2.3-alpha, 1.2.3+build)
• Partial versions (1.2 → 1.2.0, 1 → 1.0.0)
• Extra components ignored (1.2.3.4 → 1.2.3)
• Leading zeros (01.02.03 → 1.2.3)
• Invalid inputs (empty, non-numeric, leading dot)

Operator tests:

• All 6 comparison operators with true/false cases
• Tilde bounds including zero-version edge cases
• Caret bounds for all 3 semver-spec behaviors (X>0, 0.Y>0, 0.0.Z)
• Wildcard patterns (X.* and X.Y.*)
• Numeric ordering verification (9.0.0 < 10.0.0, not string comparison)

Error handling tests:

• Invalid property values return InconclusiveMatchError
• Invalid target values return InconclusiveMatchError
• Missing properties handled correctly
• Null values handled correctly

cargo test --all-features
# 85 passed; 0 failed