84708da611
PolyglotSpecRunner, BehavioralParityChecker, ParityRegressionGuard, PolyglotSuiteOrchestrator — 26/26 tests pass. Integration step 1942: poly-sort Python+Rust+Go parity verified. Metrics: 7 whetstone calls, 8855 tokens. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
65 lines
2.1 KiB
C++
65 lines
2.1 KiB
C++
#pragma once
|
|
#include "PolyglotSpecRunner.h"
|
|
#include <map>
|
|
#include <string>
|
|
#include <utility>
|
|
#include <vector>
|
|
|
|
namespace whetstone {
|
|
|
|
struct GuardResult {
|
|
bool passed = true;
|
|
std::vector<std::string> regressions; // approved langs whose output changed
|
|
std::vector<std::string> newLanguages; // unapproved langs present in results
|
|
};
|
|
|
|
class ParityRegressionGuard {
|
|
public:
|
|
// Mark (astNodeId, language) as approved with an expected output.
|
|
void approve(const std::string& astNodeId, const std::string& language,
|
|
const std::string& expectedOutput) {
|
|
baselines_[{astNodeId, language}] = expectedOutput;
|
|
}
|
|
|
|
void updateBaseline(const std::string& astNodeId, const std::string& language,
|
|
const std::string& newOutput) {
|
|
baselines_[{astNodeId, language}] = newOutput;
|
|
}
|
|
|
|
bool isApproved(const std::string& astNodeId, const std::string& language) const {
|
|
return baselines_.count({astNodeId, language}) > 0;
|
|
}
|
|
|
|
int approvedCount(const std::string& astNodeId) const {
|
|
int n = 0;
|
|
for (const auto& [key, _] : baselines_)
|
|
if (key.first == astNodeId) ++n;
|
|
return n;
|
|
}
|
|
|
|
// Check results against baselines.
|
|
// regressions: approved language whose actual output != baseline.
|
|
// newLanguages: language in results that has no baseline.
|
|
GuardResult guard(const std::string& astNodeId,
|
|
const std::vector<RunResult>& results) const {
|
|
GuardResult gr;
|
|
for (const auto& r : results) {
|
|
if (!r.ran) continue;
|
|
auto key = std::make_pair(astNodeId, r.language);
|
|
auto it = baselines_.find(key);
|
|
if (it == baselines_.end()) {
|
|
gr.newLanguages.push_back(r.language);
|
|
} else if (r.output != it->second) {
|
|
gr.regressions.push_back(r.language);
|
|
}
|
|
}
|
|
gr.passed = gr.regressions.empty();
|
|
return gr;
|
|
}
|
|
|
|
private:
|
|
std::map<std::pair<std::string, std::string>, std::string> baselines_;
|
|
};
|
|
|
|
} // namespace whetstone
|