Files
review-bot/budget/budget_test.go
T
Rodin 67d835909f
CI / test (pull_request) Successful in 13s
CI / review (gpt-5, sonnet, SONNET_REVIEW_TOKEN) (pull_request) Successful in 1m30s
CI / review (gpt-5-mini, gpt, GPT_REVIEW_TOKEN) (pull_request) Successful in 2m29s
feat: add context budget system for LLM overflow (#19)
Adds a budget package that estimates token usage and progressively
trims context to fit within model-specific limits.

Trim order (least important first):
1. Language patterns
2. Repository conventions
3. Full file context
4. Diff (truncated as last resort)

When content is trimmed, a note is appended to the user prompt so
the LLM knows context was reduced.

- New budget package with Fit(), EstimateTokens(), LimitForModel()
- Model limit table (GPT-4.1: 128K, GPT-5: 200K, Claude: 200K)
- Refactored review/prompt.go: BuildSystemBase() and BuildUserMeta()
  extract non-trimmable content; old functions delegate to new ones
- main.go uses budget.Fit() instead of direct prompt assembly
- 7 unit tests covering all trim paths

Closes #19
2026-05-01 18:46:53 -07:00

159 lines
4.2 KiB
Go

package budget
import (
"strings"
"testing"
)
func TestEstimateTokens(t *testing.T) {
tests := []struct {
input string
want int
}{
{"", 0},
{"abcd", 1},
{"12345678", 2},
{strings.Repeat("x", 400), 100},
}
for _, tt := range tests {
got := EstimateTokens(tt.input)
if got != tt.want {
t.Errorf("EstimateTokens(%d chars) = %d, want %d", len(tt.input), got, tt.want)
}
}
}
func TestLimitForModel(t *testing.T) {
tests := []struct {
model string
want int
}{
{"gpt-4.1", 128_000},
{"gpt-5", 200_000},
{"gpt-5-mini", 200_000},
{"unknown-model", defaultLimit},
{"gpt-4.1-2026-01-01", 128_000}, // prefix match
}
for _, tt := range tests {
got := LimitForModel(tt.model)
if got != tt.want {
t.Errorf("LimitForModel(%q) = %d, want %d", tt.model, got, tt.want)
}
}
}
func TestFit_AllFits(t *testing.T) {
s := Sections{
SystemBase: "system instructions",
Patterns: "some patterns",
Conventions: "some conventions",
FileContext: "file content",
Diff: "diff content",
UserMeta: "PR: title\n",
}
result := Fit("gpt-5", s)
if len(result.Trimmed) != 0 {
t.Errorf("expected no trimming, got %v", result.Trimmed)
}
if !strings.Contains(result.SystemPrompt, "some patterns") {
t.Error("expected patterns in system prompt")
}
if !strings.Contains(result.SystemPrompt, "some conventions") {
t.Error("expected conventions in system prompt")
}
if !strings.Contains(result.UserPrompt, "file content") {
t.Error("expected file context in user prompt")
}
}
func TestFit_TrimsPatterns(t *testing.T) {
// Create content that exceeds 128K token budget for gpt-4.1
// Budget ≈ 128K - 4K reserve = 124K tokens = ~496K chars
// Fill patterns with enough to push over
bigPatterns := strings.Repeat("x", 500_000) // ~125K tokens
s := Sections{
SystemBase: "base",
Patterns: bigPatterns,
Conventions: "conventions",
FileContext: "files",
Diff: "diff",
UserMeta: "meta",
}
result := Fit("gpt-4.1", s)
if len(result.Trimmed) == 0 {
t.Fatal("expected trimming")
}
if !strings.Contains(result.Trimmed[0], "patterns") {
t.Errorf("expected patterns to be trimmed first, got %v", result.Trimmed)
}
if strings.Contains(result.SystemPrompt, bigPatterns[:100]) {
t.Error("expected patterns to be removed from output")
}
// Conventions should survive
if !strings.Contains(result.SystemPrompt, "conventions") {
t.Error("expected conventions to survive after patterns trimmed")
}
}
func TestFit_TrimsConventions(t *testing.T) {
// Patterns + conventions + diff all exceed budget even after patterns removed
big := strings.Repeat("y", 520_000) // ~130K tokens each (exceeds 124K budget even alone)
s := Sections{
SystemBase: "base",
Patterns: big,
Conventions: big,
FileContext: "files",
Diff: "diff",
UserMeta: "meta",
}
result := Fit("gpt-4.1", s)
if len(result.Trimmed) < 2 {
t.Fatalf("expected at least 2 trimmed, got %v", result.Trimmed)
}
if !strings.Contains(result.Trimmed[0], "patterns") {
t.Errorf("expected patterns trimmed first, got %s", result.Trimmed[0])
}
if !strings.Contains(result.Trimmed[1], "conventions") {
t.Errorf("expected conventions trimmed second, got %s", result.Trimmed[1])
}
}
func TestFit_TruncatesDiff(t *testing.T) {
// Only diff is huge, no patterns/conventions
hugeDiff := strings.Repeat("z", 600_000) // ~150K tokens > 128K limit
s := Sections{
SystemBase: "base",
Diff: hugeDiff,
UserMeta: "meta",
}
result := Fit("gpt-4.1", s)
if len(result.Trimmed) == 0 {
t.Fatal("expected diff truncation")
}
if !strings.Contains(result.Trimmed[len(result.Trimmed)-1], "diff truncated") {
t.Errorf("expected diff truncation note, got %v", result.Trimmed)
}
if !strings.Contains(result.UserPrompt, "[diff truncated due to context limit]") {
t.Error("expected truncation marker in user prompt")
}
}
func TestFit_PreservesNoteInOutput(t *testing.T) {
big := strings.Repeat("w", 500_000)
s := Sections{
SystemBase: "base",
Patterns: big,
Diff: "small diff",
UserMeta: "meta",
}
result := Fit("gpt-4.1", s)
if !strings.Contains(result.UserPrompt, "⚠️ Note: Context was trimmed") {
t.Error("expected trimming note in user prompt")
}
}