llvm-project/clang-tools-extra/clangd/Quality.h

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

227 lines
8.0 KiB
C
Raw Normal View History

//===--- Quality.h - Ranking alternatives for ambiguous queries --*- C++-*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
///
/// Some operations such as code completion produce a set of candidates.
/// Usually the user can choose between them, but we should put the best options
/// at the top (they're easier to select, and more likely to be seen).
///
/// This file defines building blocks for ranking candidates.
/// It's used by the features directly and also in the implementation of
/// indexes, as indexes also need to heuristically limit their results.
///
/// The facilities here are:
/// - retrieving scoring signals from e.g. indexes, AST, CodeCompletionString
/// These are structured in a way that they can be debugged, and are fairly
/// consistent regardless of the source.
/// - compute scores from scoring signals. These are suitable for sorting.
/// - sorting utilities like the TopN container.
/// These could be split up further to isolate dependencies if we care.
///
//===----------------------------------------------------------------------===//
#ifndef LLVM_CLANG_TOOLS_EXTRA_CLANGD_QUALITY_H
#define LLVM_CLANG_TOOLS_EXTRA_CLANGD_QUALITY_H
#include "ExpectedTypes.h"
#include "FileDistance.h"
#include "clang/Sema/CodeCompleteConsumer.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/StringRef.h"
#include "llvm/ADT/StringSet.h"
#include <algorithm>
#include <functional>
#include <vector>
namespace llvm {
class raw_ostream;
} // namespace llvm
namespace clang {
class CodeCompletionResult;
namespace clangd {
struct Symbol;
class URIDistance;
// Signals structs are designed to be aggregated from 0 or more sources.
// A default instance has neutral signals, and sources are merged into it.
// They can be dumped for debugging, and evaluate()d into a score.
/// Attributes of a symbol that affect how much we like it.
struct SymbolQualitySignals {
bool Deprecated = false;
bool ReservedName = false; // __foo, _Foo are usually implementation details.
// FIXME: make these findable once user types _.
bool ImplementationDetail = false;
unsigned References = 0;
enum SymbolCategory {
Unknown = 0,
Variable,
Macro,
Type,
Function,
[clangd] Tune down quality score for class constructors so that it's ranked after class types. Summary: Currently, class constructors have the same score as the class types, and they are often ranked before class types. This is often not desireable and can be annoying when snippet is enabled and constructor signatures are added. Metrics: ``` ================================================================================================== OVERALL ================================================================================================== Total measurements: 111117 (+0) All measurements: MRR: 64.06 (+0.20) Top-5: 75.73% (+0.14%) Top-100: 93.71% (+0.01%) Full identifiers: MRR: 98.25 (+0.55) Top-5: 99.04% (+0.03%) Top-100: 99.16% (+0.00%) Filter length 0-5: MRR: 15.23 (+0.02) 50.50 (-0.02) 65.04 (+0.11) 70.75 (+0.19) 74.37 (+0.25) 79.43 (+0.32) Top-5: 40.90% (+0.03%) 74.52% (+0.03%) 87.23% (+0.15%) 91.68% (+0.08%) 93.68% (+0.14%) 95.87% (+0.12%) Top-100: 68.21% (+0.02%) 96.28% (+0.07%) 98.43% (+0.00%) 98.72% (+0.00%) 98.74% (+0.01%) 98.81% (+0.00%) ================================================================================================== DEFAULT ================================================================================================== Total measurements: 57535 (+0) All measurements: MRR: 58.07 (+0.37) Top-5: 69.94% (+0.26%) Top-100: 90.14% (+0.03%) Full identifiers: MRR: 97.13 (+1.05) Top-5: 98.14% (+0.06%) Top-100: 98.34% (+0.00%) Filter length 0-5: MRR: 13.91 (+0.00) 38.53 (+0.01) 55.58 (+0.21) 63.63 (+0.30) 69.23 (+0.47) 72.87 (+0.60) Top-5: 24.99% (+0.00%) 62.70% (+0.06%) 82.80% (+0.30%) 88.66% (+0.16%) 92.02% (+0.27%) 93.53% (+0.21%) Top-100: 51.56% (+0.05%) 93.19% (+0.13%) 97.30% (+0.00%) 97.81% (+0.00%) 97.85% (+0.01%) 97.79% (+0.00%) ``` Remark: - The full-id completions have +1.05 MRR improvement. - There is no noticeable impact on EXPLICIT_MEMBER_ACCESS and WANT_LOCAL. Reviewers: sammccall Subscribers: ilya-biryukov, MaskRay, jkorous, arphaman, cfe-commits Differential Revision: https://reviews.llvm.org/D49667 llvm-svn: 337816
2018-07-24 16:51:52 +08:00
Constructor,
Destructor,
Namespace,
Keyword,
Operator,
} Category = Unknown;
void merge(const CodeCompletionResult &SemaCCResult);
void merge(const Symbol &IndexResult);
// Condense these signals down to a single number, higher is better.
float evaluateHeuristics() const;
};
llvm::raw_ostream &operator<<(llvm::raw_ostream &,
const SymbolQualitySignals &);
/// Attributes of a symbol-query pair that affect how much we like it.
struct SymbolRelevanceSignals {
/// The name of the symbol (for ContextWords). Must be explicitly assigned.
llvm::StringRef Name;
/// 0-1+ fuzzy-match score for unqualified name. Must be explicitly assigned.
float NameMatch = 1;
/// Lowercase words relevant to the context (e.g. near the completion point).
llvm::StringSet<>* ContextWords = nullptr;
bool Forbidden = false; // Unavailable (e.g const) or inaccessible (private).
/// Whether fixits needs to be applied for that completion or not.
bool NeedsFixIts = false;
bool InBaseClass = false; // A member from base class of the accessed class.
URIDistance *FileProximityMatch = nullptr;
/// These are used to calculate proximity between the index symbol and the
/// query.
llvm::StringRef SymbolURI;
/// FIXME: unify with index proximity score - signals should be
/// source-independent.
/// Proximity between best declaration and the query. [0-1], 1 is closest.
float SemaFileProximityScore = 0;
// Scope proximity is only considered (both index and sema) when this is set.
ScopeDistance *ScopeProximityMatch = nullptr;
llvm::Optional<llvm::StringRef> SymbolScope;
// A symbol from sema should be accessible from the current scope.
bool SemaSaysInScope = false;
// An approximate measure of where we expect the symbol to be used.
enum AccessibleScope {
FunctionScope,
ClassScope,
FileScope,
GlobalScope,
} Scope = GlobalScope;
enum QueryType {
CodeComplete,
Generic,
} Query = Generic;
CodeCompletionContext::Kind Context = CodeCompletionContext::CCC_Other;
// Whether symbol is an instance member of a class.
bool IsInstanceMember = false;
// Whether clang provided a preferred type in the completion context.
bool HadContextType = false;
// Whether a source completion item or a symbol had a type information.
bool HadSymbolType = false;
// Whether the item matches the type expected in the completion context.
bool TypeMatchesPreferred = false;
[clangd] Use Decision Forest to score code completions. By default clangd will score a code completion item using heuristics model. Scoring can be done by Decision Forest model by passing `--ranking_model=decision_forest` to clangd. Features omitted from the model: - `NameMatch` is excluded because the final score must be multiplicative in `NameMatch` to allow rescoring by the editor. - `NeedsFixIts` is excluded because the generating dataset that needs 'fixits' is non-trivial. There are multiple ways (heuristics) to combine the above two features with the prediction of the DF: - `NeedsFixIts` is used as is with a penalty of `0.5`. Various alternatives of combining NameMatch `N` and Decision forest Prediction `P` - N * scale(P, 0, 1): Linearly scale the output of model to range [0, 1] - N * a^P: - More natural: Prediction of each Decision Tree can be considered as a multiplicative boost (like NameMatch) - Ordering is independent of the absolute value of P. Order of two items is proportional to `a^{difference in model prediction score}`. Higher `a` gives higher weightage to model output as compared to NameMatch score. Baseline MRR = 0.619 MRR for various combinations: N * P = 0.6346, advantage%=2.5768 N * 1.1^P = 0.6600, advantage%=6.6853 N * **1.2**^P = 0.6669, advantage%=**7.8005** N * **1.3**^P = 0.6668, advantage%=**7.7795** N * **1.4**^P = 0.6659, advantage%=**7.6270** N * 1.5^P = 0.6646, advantage%=7.4200 N * 1.6^P = 0.6636, advantage%=7.2671 N * 1.7^P = 0.6629, advantage%=7.1450 N * 2^P = 0.6612, advantage%=6.8673 N * 2.5^P = 0.6598, advantage%=6.6491 N * 3^P = 0.6590, advantage%=6.5242 N * scaled[0, 1] = 0.6465, advantage%=4.5054 Differential Revision: https://reviews.llvm.org/D88281
2020-09-22 13:56:08 +08:00
/// Length of the unqualified partial name of Symbol typed in
/// CompletionPrefix.
unsigned FilterLength = 0;
/// Set of derived signals computed by calculateDerivedSignals(). Must not be
/// set explicitly.
struct DerivedSignals {
/// Whether Name contains some word from context.
bool NameMatchesContext = false;
/// Min distance between SymbolURI and all the headers included by the TU.
unsigned FileProximityDistance = FileDistance::Unreachable;
/// Min distance between SymbolScope and all the available scopes.
unsigned ScopeProximityDistance = FileDistance::Unreachable;
};
DerivedSignals calculateDerivedSignals() const;
void merge(const CodeCompletionResult &SemaResult);
void merge(const Symbol &IndexResult);
// Condense these signals down to a single number, higher is better.
float evaluateHeuristics() const;
};
llvm::raw_ostream &operator<<(llvm::raw_ostream &,
const SymbolRelevanceSignals &);
/// Combine symbol quality and relevance into a single score.
float evaluateSymbolAndRelevance(float SymbolQuality, float SymbolRelevance);
[clangd] Use Decision Forest to score code completions. By default clangd will score a code completion item using heuristics model. Scoring can be done by Decision Forest model by passing `--ranking_model=decision_forest` to clangd. Features omitted from the model: - `NameMatch` is excluded because the final score must be multiplicative in `NameMatch` to allow rescoring by the editor. - `NeedsFixIts` is excluded because the generating dataset that needs 'fixits' is non-trivial. There are multiple ways (heuristics) to combine the above two features with the prediction of the DF: - `NeedsFixIts` is used as is with a penalty of `0.5`. Various alternatives of combining NameMatch `N` and Decision forest Prediction `P` - N * scale(P, 0, 1): Linearly scale the output of model to range [0, 1] - N * a^P: - More natural: Prediction of each Decision Tree can be considered as a multiplicative boost (like NameMatch) - Ordering is independent of the absolute value of P. Order of two items is proportional to `a^{difference in model prediction score}`. Higher `a` gives higher weightage to model output as compared to NameMatch score. Baseline MRR = 0.619 MRR for various combinations: N * P = 0.6346, advantage%=2.5768 N * 1.1^P = 0.6600, advantage%=6.6853 N * **1.2**^P = 0.6669, advantage%=**7.8005** N * **1.3**^P = 0.6668, advantage%=**7.7795** N * **1.4**^P = 0.6659, advantage%=**7.6270** N * 1.5^P = 0.6646, advantage%=7.4200 N * 1.6^P = 0.6636, advantage%=7.2671 N * 1.7^P = 0.6629, advantage%=7.1450 N * 2^P = 0.6612, advantage%=6.8673 N * 2.5^P = 0.6598, advantage%=6.6491 N * 3^P = 0.6590, advantage%=6.5242 N * scaled[0, 1] = 0.6465, advantage%=4.5054 Differential Revision: https://reviews.llvm.org/D88281
2020-09-22 13:56:08 +08:00
float evaluateDecisionForest(const SymbolQualitySignals &Quality,
const SymbolRelevanceSignals &Relevance);
/// TopN<T> is a lossy container that preserves only the "best" N elements.
template <typename T, typename Compare = std::greater<T>> class TopN {
public:
using value_type = T;
TopN(size_t N, Compare Greater = Compare())
: N(N), Greater(std::move(Greater)) {}
// Adds a candidate to the set.
// Returns true if a candidate was dropped to get back under N.
bool push(value_type &&V) {
bool Dropped = false;
if (Heap.size() >= N) {
Dropped = true;
if (N > 0 && Greater(V, Heap.front())) {
std::pop_heap(Heap.begin(), Heap.end(), Greater);
Heap.back() = std::move(V);
std::push_heap(Heap.begin(), Heap.end(), Greater);
}
} else {
Heap.push_back(std::move(V));
std::push_heap(Heap.begin(), Heap.end(), Greater);
}
assert(Heap.size() <= N);
assert(std::is_heap(Heap.begin(), Heap.end(), Greater));
return Dropped;
}
// Returns candidates from best to worst.
std::vector<value_type> items() && {
std::sort_heap(Heap.begin(), Heap.end(), Greater);
assert(Heap.size() <= N);
return std::move(Heap);
}
private:
const size_t N;
std::vector<value_type> Heap; // Min-heap, comparator is Greater.
Compare Greater;
};
/// Returns a string that sorts in the same order as (-Score, Tiebreak), for
/// LSP. (The highest score compares smallest so it sorts at the top).
std::string sortText(float Score, llvm::StringRef Tiebreak = "");
struct SignatureQualitySignals {
uint32_t NumberOfParameters = 0;
uint32_t NumberOfOptionalParameters = 0;
CodeCompleteConsumer::OverloadCandidate::CandidateKind Kind =
CodeCompleteConsumer::OverloadCandidate::CandidateKind::CK_Function;
};
llvm::raw_ostream &operator<<(llvm::raw_ostream &,
const SignatureQualitySignals &);
} // namespace clangd
} // namespace clang
#endif // LLVM_CLANG_TOOLS_EXTRA_CLANGD_QUALITY_H