Add backtracking

[ptoffy]

Fixes #145

This adds recursive backtracking to the trie algorithm. Unfortunately this will likely result in some performance loss when routing but we do want routes to be resolved correctly

[codecov]

Codecov Report

❌ Patch coverage is 60.25641% with 31 lines in your changes missing coverage. Please review.
✅ Project coverage is 84.92%. Comparing base (6629807) to head (dda4c9b).

Files with missing lines	Patch %	Lines
Sources/RoutingKit/TrieRouter/TrieRouter.swift	60.25%	31 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #149      +/-   ##
==========================================
- Coverage   90.43%   84.92%   -5.52%     
==========================================
  Files           5        5              
  Lines         251      325      +74     
==========================================
+ Hits          227      276      +49     
- Misses         24       49      +25     

Files with missing lines	Coverage Δ
Sources/RoutingKit/TrieRouter/TrieRouter.swift	77.02% <60.25%> (-18.92%)	⬇️

... and 1 file with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[gwynne]

My nits are not particularly critical, not using them will not make a huge difference. It'd be interesting to know if they benchmarked differently though.

[gwynne]

Suggested change

	if let partials = currentNode.partials, !partials.isEmpty {
	for partial in partials {
	alternatives.append(
	.init(
	node: partial.node,
	index: index,
	kind: .partial(partial),
	parameterSnapshot: parameters
	))
	}
	}
	alternatives.append(contentsOf: currentNode.partials?.map { .init(
	node: partial.node, index: index, kind: .partial(partial), parameterSnapshot: parameters
	) } ?? [])

Note

.append(contentsOf: []) and [].map(...) both compile to no-ops, so eliding the conditional does not degrade performance.

[gwynne]

Suggested change

	if let partials = currentNode.partials, !partials.isEmpty {
	for partial in partials {
	alternatives.append(
	.init(
	node: partial.node,
	index: index,
	kind: .partial(partial),
	parameterSnapshot: parameters
	))
	}
	}
	alternatives.append(contentsOf: currentNode.partials?.map { .init(
	node: partial.node, index: index, kind: .partial(partial), parameterSnapshot: parameters
	) } ?? [])

Note

.append(contentsOf: []) and [].map(...) both compile to no-ops, so eliding the conditional does not degrade performance.

[penny-for-vapor]

Benchmark Report for dda4c9b

❓ Pull request has significant performance differences 📊

Click to expand comparison result

Benchmark check running at 2025-11-25 11:29:11 UTC

Reading thresholds from "Benchmarks/Thresholds/"

Checking ["RouterPerformance:Case-sensitive", "RouterPerformance:Case-insensitive", "RouterPerformance:Case-insensitive_Match_First", "RouterPerformance:Case-insensitive_Match_Last", "RouterPerformance:Case-sensitive_Minimal", "RouterPerformance:Case-insensitive_Minimal", "RouterPerformance:Minimal_Early_Fail"]

========================================================================
Deviations worse than threshold for RouterPerformance:Minimal_Early_Fail
========================================================================

Throughput (# / s) (K, %)	p90 threshold	Add backtracki…	Difference %	Threshold %
p90	512	435	15	10

The baseline 'Add backtracking' is WORSE than the defined thresholds.

Click to expand benchmark result

Baseline 'Add backtracking'

Host 'b1e93067c10c' with 2 'aarch64' processors with 3 GB memory, running:
#16~24.04.1-Ubuntu SMP Tue Oct 14 01:56:45 UTC 2025

RouterPerformance

Case-insensitive

Metric	p0	p25	p50	p75	p90	p99	p100	Samples
Malloc (total) *	22	22	22	22	22	22	22	9289
Memory (resident peak) (M)	24	24	24	24	24	24	24	9289
Throughput (# / s) (K)	237	196	191	185	177	163	49	9289

Case-insensitive_Match_First

Metric	p0	p25	p50	p75	p90	p99	p100	Samples
Malloc (total) *	22	22	22	22	22	22	22	9271
Memory (resident peak) (M)	24	24	24	24	24	24	24	9271
Throughput (# / s) (K)	236	196	189	183	176	157	43	9271

Case-insensitive_Match_Last

Metric	p0	p25	p50	p75	p90	p99	p100	Samples
Malloc (total) *	22	22	22	22	22	22	22	8936
Memory (resident peak) (M)	24	24	24	24	24	24	24	8936
Throughput (# / s) (K)	215	182	177	172	167	151	32	8936

Case-insensitive_Minimal

Metric	p0	p25	p50	p75	p90	p99	p100	Samples
Malloc (total) *	18	18	18	18	18	18	18	10000
Memory (resident peak) (M)	23	24	24	24	24	24	24	10000
Throughput (# / s) (K)	528	479	473	466	458	398	64	10000

Case-sensitive

Metric	p0	p25	p50	p75	p90	p99	p100	Samples
Malloc (total) *	22	22	22	22	22	22	22	9266
Memory (resident peak) (M)	24	24	24	24	24	24	24	9266
Throughput (# / s) (K)	240	200	195	190	185	164	23	9266

Case-sensitive_Minimal

Metric	p0	p25	p50	p75	p90	p99	p100	Samples
Malloc (total) *	18	18	18	18	18	18	18	10000
Memory (resident peak) (M)	23	24	24	24	24	24	24	10000
Throughput (# / s) (K)	532	481	474	465	455	390	123	10000

Minimal_Early_Fail

Metric	p0	p25	p50	p75	p90	p99	p100	Samples
Malloc (total) *	18	18	18	18	18	18	18	10000
Memory (resident peak) (M)	23	24	24	24	24	24	24	10000
Throughput (# / s) (K)	507	459	453	445	435	377	112	10000