Coverage Report

Created: 2022-01-18 06:27

/Users/buildslave/jenkins/workspace/coverage/llvm-project/clang/lib/Rewrite/DeltaTree.cpp
Line
Count
Source (jump to first uncovered line)
1
//===- DeltaTree.cpp - B-Tree for Rewrite Delta tracking ------------------===//
2
//
3
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4
// See https://llvm.org/LICENSE.txt for license information.
5
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6
//
7
//===----------------------------------------------------------------------===//
8
//
9
// This file implements the DeltaTree and related classes.
10
//
11
//===----------------------------------------------------------------------===//
12
13
#include "clang/Rewrite/Core/DeltaTree.h"
14
#include "clang/Basic/LLVM.h"
15
#include "llvm/Support/Casting.h"
16
#include <cassert>
17
#include <cstring>
18
19
using namespace clang;
20
21
/// The DeltaTree class is a multiway search tree (BTree) structure with some
22
/// fancy features.  B-Trees are generally more memory and cache efficient
23
/// than binary trees, because they store multiple keys/values in each node.
24
///
25
/// DeltaTree implements a key/value mapping from FileIndex to Delta, allowing
26
/// fast lookup by FileIndex.  However, an added (important) bonus is that it
27
/// can also efficiently tell us the full accumulated delta for a specific
28
/// file offset as well, without traversing the whole tree.
29
///
30
/// The nodes of the tree are made up of instances of two classes:
31
/// DeltaTreeNode and DeltaTreeInteriorNode.  The later subclasses the
32
/// former and adds children pointers.  Each node knows the full delta of all
33
/// entries (recursively) contained inside of it, which allows us to get the
34
/// full delta implied by a whole subtree in constant time.
35
36
namespace {
37
38
  /// SourceDelta - As code in the original input buffer is added and deleted,
39
  /// SourceDelta records are used to keep track of how the input SourceLocation
40
  /// object is mapped into the output buffer.
41
  struct SourceDelta {
42
    unsigned FileLoc;
43
    int Delta;
44
45
113k
    static SourceDelta get(unsigned Loc, int D) {
46
113k
      SourceDelta Delta;
47
113k
      Delta.FileLoc = Loc;
48
113k
      Delta.Delta = D;
49
113k
      return Delta;
50
113k
    }
51
  };
52
53
  /// DeltaTreeNode - The common part of all nodes.
54
  ///
55
  class DeltaTreeNode {
56
  public:
57
    struct InsertResult {
58
      DeltaTreeNode *LHS, *RHS;
59
      SourceDelta Split;
60
    };
61
62
  private:
63
    friend class DeltaTreeInteriorNode;
64
65
    /// WidthFactor - This controls the number of K/V slots held in the BTree:
66
    /// how wide it is.  Each level of the BTree is guaranteed to have at least
67
    /// WidthFactor-1 K/V pairs (except the root) and may have at most
68
    /// 2*WidthFactor-1 K/V pairs.
69
    enum { WidthFactor = 8 };
70
71
    /// Values - This tracks the SourceDelta's currently in this node.
72
    SourceDelta Values[2*WidthFactor-1];
73
74
    /// NumValuesUsed - This tracks the number of values this node currently
75
    /// holds.
76
    unsigned char NumValuesUsed = 0;
77
78
    /// IsLeaf - This is true if this is a leaf of the btree.  If false, this is
79
    /// an interior node, and is actually an instance of DeltaTreeInteriorNode.
80
    bool IsLeaf;
81
82
    /// FullDelta - This is the full delta of all the values in this node and
83
    /// all children nodes.
84
    int FullDelta = 0;
85
86
  public:
87
65.3k
    DeltaTreeNode(bool isLeaf = true) : IsLeaf(isLeaf) {}
88
89
956k
    bool isLeaf() const { return IsLeaf; }
90
826k
    int getFullDelta() const { return FullDelta; }
91
133k
    bool isFull() const { return NumValuesUsed == 2*WidthFactor-1; }
92
93
1.76M
    unsigned getNumValuesUsed() const { return NumValuesUsed; }
94
95
1.71M
    const SourceDelta &getValue(unsigned i) const {
96
1.71M
      assert(i < NumValuesUsed && "Invalid value #");
97
0
      return Values[i];
98
1.71M
    }
99
100
1.77M
    SourceDelta &getValue(unsigned i) {
101
1.77M
      assert(i < NumValuesUsed && "Invalid value #");
102
0
      return Values[i];
103
1.77M
    }
104
105
    /// DoInsertion - Do an insertion of the specified FileIndex/Delta pair into
106
    /// this node.  If insertion is easy, do it and return false.  Otherwise,
107
    /// split the node, populate InsertRes with info about the split, and return
108
    /// true.
109
    bool DoInsertion(unsigned FileIndex, int Delta, InsertResult *InsertRes);
110
111
    void DoSplit(InsertResult &InsertRes);
112
113
114
    /// RecomputeFullDeltaLocally - Recompute the FullDelta field by doing a
115
    /// local walk over our contained deltas.
116
    void RecomputeFullDeltaLocally();
117
118
    void Destroy();
119
  };
120
121
  /// DeltaTreeInteriorNode - When isLeaf = false, a node has child pointers.
122
  /// This class tracks them.
123
  class DeltaTreeInteriorNode : public DeltaTreeNode {
124
    friend class DeltaTreeNode;
125
126
    DeltaTreeNode *Children[2*WidthFactor];
127
128
1.01k
    ~DeltaTreeInteriorNode() {
129
8.51k
      for (unsigned i = 0, e = NumValuesUsed+1; i != e; 
++i7.50k
)
130
7.50k
        Children[i]->Destroy();
131
1.01k
    }
132
133
  public:
134
558
    DeltaTreeInteriorNode() : DeltaTreeNode(false /*nonleaf*/) {}
135
136
    DeltaTreeInteriorNode(const InsertResult &IR)
137
455
        : DeltaTreeNode(false /*nonleaf*/) {
138
455
      Children[0] = IR.LHS;
139
455
      Children[1] = IR.RHS;
140
455
      Values[0] = IR.Split;
141
455
      FullDelta = IR.LHS->getFullDelta()+IR.RHS->getFullDelta()+IR.Split.Delta;
142
455
      NumValuesUsed = 1;
143
455
    }
144
145
966k
    const DeltaTreeNode *getChild(unsigned i) const {
146
966k
      assert(i < getNumValuesUsed()+1 && "Invalid child");
147
0
      return Children[i];
148
966k
    }
149
150
8.92k
    DeltaTreeNode *getChild(unsigned i) {
151
8.92k
      assert(i < getNumValuesUsed()+1 && "Invalid child");
152
0
      return Children[i];
153
8.92k
    }
154
155
624k
    static bool classof(const DeltaTreeNode *N) { return !N->isLeaf(); }
156
  };
157
158
} // namespace
159
160
/// Destroy - A 'virtual' destructor.
161
65.3k
void DeltaTreeNode::Destroy() {
162
65.3k
  if (isLeaf())
163
64.2k
    delete this;
164
1.01k
  else
165
1.01k
    delete cast<DeltaTreeInteriorNode>(this);
166
65.3k
}
167
168
/// RecomputeFullDeltaLocally - Recompute the FullDelta field by doing a
169
/// local walk over our contained deltas.
170
14.1k
void DeltaTreeNode::RecomputeFullDeltaLocally() {
171
14.1k
  int NewFullDelta = 0;
172
112k
  for (unsigned i = 0, e = getNumValuesUsed(); i != e; 
++i98.7k
)
173
98.7k
    NewFullDelta += Values[i].Delta;
174
14.1k
  if (auto *IN = dyn_cast<DeltaTreeInteriorNode>(this))
175
10.0k
    
for (unsigned i = 0, e = getNumValuesUsed()+1; 1.11k
i != e;
++i8.92k
)
176
8.92k
      NewFullDelta += IN->getChild(i)->getFullDelta();
177
14.1k
  FullDelta = NewFullDelta;
178
14.1k
}
179
180
/// DoInsertion - Do an insertion of the specified FileIndex/Delta pair into
181
/// this node.  If insertion is easy, do it and return false.  Otherwise,
182
/// split the node, populate InsertRes with info about the split, and return
183
/// true.
184
bool DeltaTreeNode::DoInsertion(unsigned FileIndex, int Delta,
185
282k
                                InsertResult *InsertRes) {
186
  // Maintain full delta for this node.
187
282k
  FullDelta += Delta;
188
189
  // Find the insertion point, the first delta whose index is >= FileIndex.
190
282k
  unsigned i = 0, e = getNumValuesUsed();
191
1.65M
  while (i != e && 
FileIndex > getValue(i).FileLoc1.57M
)
192
1.37M
    ++i;
193
194
  // If we found an a record for exactly this file index, just merge this
195
  // value into the pre-existing record and finish early.
196
282k
  if (i != e && 
getValue(i).FileLoc == FileIndex198k
) {
197
    // NOTE: Delta could drop to zero here.  This means that the delta entry is
198
    // useless and could be removed.  Supporting erases is more complex than
199
    // leaving an entry with Delta=0, so we just leave an entry with Delta=0 in
200
    // the tree.
201
15.5k
    Values[i].Delta += Delta;
202
15.5k
    return false;
203
15.5k
  }
204
205
  // Otherwise, we found an insertion point, and we know that the value at the
206
  // specified index is > FileIndex.  Handle the leaf case first.
207
266k
  if (isLeaf()) {
208
120k
    if (!isFull()) {
209
      // For an insertion into a non-full leaf node, just insert the value in
210
      // its sorted position.  This requires moving later values over.
211
113k
      if (i != e)
212
84.1k
        memmove(&Values[i+1], &Values[i], sizeof(Values[0])*(e-i));
213
113k
      Values[i] = SourceDelta::get(FileIndex, Delta);
214
113k
      ++NumValuesUsed;
215
113k
      return false;
216
113k
    }
217
218
    // Otherwise, if this is leaf is full, split the node at its median, insert
219
    // the value into one of the children, and return the result.
220
6.49k
    assert(InsertRes && "No result location specified");
221
0
    DoSplit(*InsertRes);
222
223
6.49k
    if (InsertRes->Split.FileLoc > FileIndex)
224
777
      InsertRes->LHS->DoInsertion(FileIndex, Delta, nullptr /*can't fail*/);
225
5.71k
    else
226
5.71k
      InsertRes->RHS->DoInsertion(FileIndex, Delta, nullptr /*can't fail*/);
227
6.49k
    return true;
228
120k
  }
229
230
  // Otherwise, this is an interior node.  Send the request down the tree.
231
146k
  auto *IN = cast<DeltaTreeInteriorNode>(this);
232
146k
  if (!IN->Children[i]->DoInsertion(FileIndex, Delta, InsertRes))
233
139k
    return false; // If there was space in the child, just return.
234
235
  // Okay, this split the subtree, producing a new value and two children to
236
  // insert here.  If this node is non-full, we can just insert it directly.
237
6.59k
  if (!isFull()) {
238
    // Now that we have two nodes and a new element, insert the perclated value
239
    // into ourself by moving all the later values/children down, then inserting
240
    // the new one.
241
6.03k
    if (i != e)
242
4.38k
      memmove(&IN->Children[i+2], &IN->Children[i+1],
243
4.38k
              (e-i)*sizeof(IN->Children[0]));
244
6.03k
    IN->Children[i] = InsertRes->LHS;
245
6.03k
    IN->Children[i+1] = InsertRes->RHS;
246
247
6.03k
    if (e != i)
248
4.38k
      memmove(&Values[i+1], &Values[i], (e-i)*sizeof(Values[0]));
249
6.03k
    Values[i] = InsertRes->Split;
250
6.03k
    ++NumValuesUsed;
251
6.03k
    return false;
252
6.03k
  }
253
254
  // Finally, if this interior node was full and a node is percolated up, split
255
  // ourself and return that up the chain.  Start by saving all our info to
256
  // avoid having the split clobber it.
257
558
  IN->Children[i] = InsertRes->LHS;
258
558
  DeltaTreeNode *SubRHS = InsertRes->RHS;
259
558
  SourceDelta SubSplit = InsertRes->Split;
260
261
  // Do the split.
262
558
  DoSplit(*InsertRes);
263
264
  // Figure out where to insert SubRHS/NewSplit.
265
558
  DeltaTreeInteriorNode *InsertSide;
266
558
  if (SubSplit.FileLoc < InsertRes->Split.FileLoc)
267
40
    InsertSide = cast<DeltaTreeInteriorNode>(InsertRes->LHS);
268
518
  else
269
518
    InsertSide = cast<DeltaTreeInteriorNode>(InsertRes->RHS);
270
271
  // We now have a non-empty interior node 'InsertSide' to insert
272
  // SubRHS/SubSplit into.  Find out where to insert SubSplit.
273
274
  // Find the insertion point, the first delta whose index is >SubSplit.FileLoc.
275
558
  i = 0; e = InsertSide->getNumValuesUsed();
276
3.25k
  while (i != e && 
SubSplit.FileLoc > InsertSide->getValue(i).FileLoc3.12k
)
277
2.69k
    ++i;
278
279
  // Now we know that i is the place to insert the split value into.  Insert it
280
  // and the child right after it.
281
558
  if (i != e)
282
436
    memmove(&InsertSide->Children[i+2], &InsertSide->Children[i+1],
283
436
            (e-i)*sizeof(IN->Children[0]));
284
558
  InsertSide->Children[i+1] = SubRHS;
285
286
558
  if (e != i)
287
436
    memmove(&InsertSide->Values[i+1], &InsertSide->Values[i],
288
436
            (e-i)*sizeof(Values[0]));
289
558
  InsertSide->Values[i] = SubSplit;
290
558
  ++InsertSide->NumValuesUsed;
291
558
  InsertSide->FullDelta += SubSplit.Delta + SubRHS->getFullDelta();
292
558
  return true;
293
6.59k
}
294
295
/// DoSplit - Split the currently full node (which has 2*WidthFactor-1 values)
296
/// into two subtrees each with "WidthFactor-1" values and a pivot value.
297
/// Return the pieces in InsertRes.
298
7.05k
void DeltaTreeNode::DoSplit(InsertResult &InsertRes) {
299
7.05k
  assert(isFull() && "Why split a non-full node?");
300
301
  // Since this node is full, it contains 2*WidthFactor-1 values.  We move
302
  // the first 'WidthFactor-1' values to the LHS child (which we leave in this
303
  // node), propagate one value up, and move the last 'WidthFactor-1' values
304
  // into the RHS child.
305
306
  // Create the new child node.
307
0
  DeltaTreeNode *NewNode;
308
7.05k
  if (auto *IN = dyn_cast<DeltaTreeInteriorNode>(this)) {
309
    // If this is an interior node, also move over 'WidthFactor' children
310
    // into the new node.
311
558
    DeltaTreeInteriorNode *New = new DeltaTreeInteriorNode();
312
558
    memcpy(&New->Children[0], &IN->Children[WidthFactor],
313
558
           WidthFactor*sizeof(IN->Children[0]));
314
558
    NewNode = New;
315
6.49k
  } else {
316
    // Just create the new leaf node.
317
6.49k
    NewNode = new DeltaTreeNode();
318
6.49k
  }
319
320
  // Move over the last 'WidthFactor-1' values from here to NewNode.
321
7.05k
  memcpy(&NewNode->Values[0], &Values[WidthFactor],
322
7.05k
         (WidthFactor-1)*sizeof(Values[0]));
323
324
  // Decrease the number of values in the two nodes.
325
7.05k
  NewNode->NumValuesUsed = NumValuesUsed = WidthFactor-1;
326
327
  // Recompute the two nodes' full delta.
328
7.05k
  NewNode->RecomputeFullDeltaLocally();
329
7.05k
  RecomputeFullDeltaLocally();
330
331
7.05k
  InsertRes.LHS = this;
332
7.05k
  InsertRes.RHS = NewNode;
333
7.05k
  InsertRes.Split = Values[WidthFactor-1];
334
7.05k
}
335
336
//===----------------------------------------------------------------------===//
337
//                        DeltaTree Implementation
338
//===----------------------------------------------------------------------===//
339
340
//#define VERIFY_TREE
341
342
#ifdef VERIFY_TREE
343
/// VerifyTree - Walk the btree performing assertions on various properties to
344
/// verify consistency.  This is useful for debugging new changes to the tree.
345
static void VerifyTree(const DeltaTreeNode *N) {
346
  const auto *IN = dyn_cast<DeltaTreeInteriorNode>(N);
347
  if (IN == 0) {
348
    // Verify leaves, just ensure that FullDelta matches up and the elements
349
    // are in proper order.
350
    int FullDelta = 0;
351
    for (unsigned i = 0, e = N->getNumValuesUsed(); i != e; ++i) {
352
      if (i)
353
        assert(N->getValue(i-1).FileLoc < N->getValue(i).FileLoc);
354
      FullDelta += N->getValue(i).Delta;
355
    }
356
    assert(FullDelta == N->getFullDelta());
357
    return;
358
  }
359
360
  // Verify interior nodes: Ensure that FullDelta matches up and the
361
  // elements are in proper order and the children are in proper order.
362
  int FullDelta = 0;
363
  for (unsigned i = 0, e = IN->getNumValuesUsed(); i != e; ++i) {
364
    const SourceDelta &IVal = N->getValue(i);
365
    const DeltaTreeNode *IChild = IN->getChild(i);
366
    if (i)
367
      assert(IN->getValue(i-1).FileLoc < IVal.FileLoc);
368
    FullDelta += IVal.Delta;
369
    FullDelta += IChild->getFullDelta();
370
371
    // The largest value in child #i should be smaller than FileLoc.
372
    assert(IChild->getValue(IChild->getNumValuesUsed()-1).FileLoc <
373
           IVal.FileLoc);
374
375
    // The smallest value in child #i+1 should be larger than FileLoc.
376
    assert(IN->getChild(i+1)->getValue(0).FileLoc > IVal.FileLoc);
377
    VerifyTree(IChild);
378
  }
379
380
  FullDelta += IN->getChild(IN->getNumValuesUsed())->getFullDelta();
381
382
  assert(FullDelta == N->getFullDelta());
383
}
384
#endif  // VERIFY_TREE
385
386
376k
static DeltaTreeNode *getRoot(void *Root) {
387
376k
  return (DeltaTreeNode*)Root;
388
376k
}
389
390
19.2k
DeltaTree::DeltaTree() {
391
19.2k
  Root = new DeltaTreeNode();
392
19.2k
}
393
394
38.5k
DeltaTree::DeltaTree(const DeltaTree &RHS) {
395
  // Currently we only support copying when the RHS is empty.
396
38.5k
  assert(getRoot(RHS.Root)->getNumValuesUsed() == 0 &&
397
38.5k
         "Can only copy empty tree");
398
0
  Root = new DeltaTreeNode();
399
38.5k
}
400
401
57.7k
DeltaTree::~DeltaTree() {
402
57.7k
  getRoot(Root)->Destroy();
403
57.7k
}
404
405
/// getDeltaAt - Return the accumulated delta at the specified file offset.
406
/// This includes all insertions or delections that occurred *before* the
407
/// specified file index.
408
151k
int DeltaTree::getDeltaAt(unsigned FileIndex) const {
409
151k
  const DeltaTreeNode *Node = getRoot(Root);
410
411
151k
  int Result = 0;
412
413
  // Walk down the tree.
414
301k
  while (true) {
415
    // For all nodes, include any local deltas before the specified file
416
    // index by summing them up directly.  Keep track of how many were
417
    // included.
418
301k
    unsigned NumValsGreater = 0;
419
1.72M
    for (unsigned e = Node->getNumValuesUsed(); NumValsGreater != e;
420
1.62M
         
++NumValsGreater1.42M
) {
421
1.62M
      const SourceDelta &Val = Node->getValue(NumValsGreater);
422
423
1.62M
      if (Val.FileLoc >= FileIndex)
424
198k
        break;
425
1.42M
      Result += Val.Delta;
426
1.42M
    }
427
428
    // If we have an interior node, include information about children and
429
    // recurse.  Otherwise, if we have a leaf, we're done.
430
301k
    const auto *IN = dyn_cast<DeltaTreeInteriorNode>(Node);
431
301k
    if (!IN) 
return Result150k
;
432
433
    // Include any children to the left of the values we skipped, all of
434
    // their deltas should be included as well.
435
966k
    
for (unsigned i = 0; 151k
i != NumValsGreater;
++i814k
)
436
814k
      Result += IN->getChild(i)->getFullDelta();
437
438
    // If we found exactly the value we were looking for, break off the
439
    // search early.  There is no need to search the RHS of the value for
440
    // partial results.
441
151k
    if (NumValsGreater != Node->getNumValuesUsed() &&
442
151k
        
Node->getValue(NumValsGreater).FileLoc == FileIndex94.9k
)
443
810
      return Result+IN->getChild(NumValsGreater)->getFullDelta();
444
445
    // Otherwise, traverse down the tree.  The selected subtree may be
446
    // partially included in the range.
447
150k
    Node = IN->getChild(NumValsGreater);
448
150k
  }
449
  // NOT REACHED.
450
151k
}
451
452
/// AddDelta - When a change is made that shifts around the text buffer,
453
/// this method is used to record that info.  It inserts a delta of 'Delta'
454
/// into the current DeltaTree at offset FileIndex.
455
129k
void DeltaTree::AddDelta(unsigned FileIndex, int Delta) {
456
129k
  assert(Delta && "Adding a noop?");
457
0
  DeltaTreeNode *MyRoot = getRoot(Root);
458
459
129k
  DeltaTreeNode::InsertResult InsertRes;
460
129k
  if (MyRoot->DoInsertion(FileIndex, Delta, &InsertRes)) {
461
455
    Root = new DeltaTreeInteriorNode(InsertRes);
462
#ifdef VERIFY_TREE
463
    MyRoot = Root;
464
#endif
465
455
  }
466
467
#ifdef VERIFY_TREE
468
  VerifyTree(MyRoot);
469
#endif
470
129k
}