1//===- DbiModuleList.cpp - PDB module information list --------------------===//
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#include "llvm/DebugInfo/PDB/Native/DbiModuleList.h"
10#include "llvm/ADT/StringRef.h"
11#include "llvm/ADT/iterator_range.h"
12#include "llvm/DebugInfo/PDB/Native/RawError.h"
13#include "llvm/DebugInfo/PDB/Native/RawTypes.h"
14#include "llvm/Support/BinaryStreamReader.h"
15#include "llvm/Support/Error.h"
16#include <algorithm>
17#include <cassert>
18#include <cstddef>
19#include <cstdint>
20
21using namespace llvm;
22using namespace llvm::pdb;
23
24DbiModuleSourceFilesIterator::DbiModuleSourceFilesIterator(
25 const DbiModuleList &Modules, uint32_t Modi, uint16_t Filei)
26 : Modules(&Modules), Modi(Modi), Filei(Filei) {
27 setValue();
28}
29
30bool DbiModuleSourceFilesIterator::
31operator==(const DbiModuleSourceFilesIterator &R) const {
32 // incompatible iterators are never equal
33 if (!isCompatible(R))
34 return false;
35
36 // If they're compatible, and they're both ends, then they're equal.
37 if (isEnd() && R.isEnd())
38 return true;
39
40 // If one is an end and the other is not, they're not equal.
41 if (isEnd() != R.isEnd())
42 return false;
43
44 // Now we know:
45 // - They're compatible
46 // - They're not *both* end iterators
47 // - Their endness is the same.
48 // Thus, they're compatible iterators pointing to a valid file on the same
49 // module. All we need to check are the file indices.
50 assert(Modules == R.Modules);
51 assert(Modi == R.Modi);
52 assert(!isEnd());
53 assert(!R.isEnd());
54
55 return (Filei == R.Filei);
56}
57
58bool DbiModuleSourceFilesIterator::
59operator<(const DbiModuleSourceFilesIterator &R) const {
60 assert(isCompatible(R));
61
62 // It's not sufficient to compare the file indices, because default
63 // constructed iterators could be equal to iterators with valid indices. To
64 // account for this, early-out if they're equal.
65 if (*this == R)
66 return false;
67
68 return Filei < R.Filei;
69}
70
71std::ptrdiff_t DbiModuleSourceFilesIterator::
72operator-(const DbiModuleSourceFilesIterator &R) const {
73 assert(isCompatible(R));
74 assert(!(*this < R));
75
76 // If they're both end iterators, the distance is 0.
77 if (isEnd() && R.isEnd())
78 return 0;
79
80 assert(!R.isEnd());
81
82 // At this point, R cannot be end, but *this can, which means that *this
83 // might be a universal end iterator with none of its fields set. So in that
84 // case have to rely on R as the authority to figure out how many files there
85 // are to compute the distance.
86 uint32_t Thisi = Filei;
87 if (isEnd()) {
88 uint32_t RealModi = R.Modi;
89 Thisi = R.Modules->getSourceFileCount(Modi: RealModi);
90 }
91
92 assert(Thisi >= R.Filei);
93 return Thisi - R.Filei;
94}
95
96DbiModuleSourceFilesIterator &DbiModuleSourceFilesIterator::
97operator+=(std::ptrdiff_t N) {
98 assert(!isEnd());
99
100 Filei += N;
101 assert(Filei <= Modules->getSourceFileCount(Modi));
102 setValue();
103 return *this;
104}
105
106DbiModuleSourceFilesIterator &DbiModuleSourceFilesIterator::
107operator-=(std::ptrdiff_t N) {
108 // Note that we can subtract from an end iterator, but not a universal end
109 // iterator.
110 assert(!isUniversalEnd());
111
112 assert(N <= Filei);
113
114 Filei -= N;
115 return *this;
116}
117
118void DbiModuleSourceFilesIterator::setValue() {
119 if (isEnd()) {
120 ThisValue = "";
121 return;
122 }
123
124 uint32_t Off = Modules->ModuleInitialFileIndex[Modi] + Filei;
125 auto ExpectedValue = Modules->getFileName(Index: Off);
126 if (!ExpectedValue) {
127 consumeError(Err: ExpectedValue.takeError());
128 Filei = Modules->getSourceFileCount(Modi);
129 } else
130 ThisValue = *ExpectedValue;
131}
132
133bool DbiModuleSourceFilesIterator::isEnd() const {
134 if (isUniversalEnd())
135 return true;
136
137 assert(Modules);
138 assert(Modi <= Modules->getModuleCount());
139 assert(Filei <= Modules->getSourceFileCount(Modi));
140
141 if (Modi == Modules->getModuleCount())
142 return true;
143 if (Filei == Modules->getSourceFileCount(Modi))
144 return true;
145 return false;
146}
147
148bool DbiModuleSourceFilesIterator::isUniversalEnd() const { return !Modules; }
149
150bool DbiModuleSourceFilesIterator::isCompatible(
151 const DbiModuleSourceFilesIterator &R) const {
152 // Universal iterators are compatible with any other iterator.
153 if (isUniversalEnd() || R.isUniversalEnd())
154 return true;
155
156 // At this point, neither iterator is a universal end iterator, although one
157 // or both might be non-universal end iterators. Regardless, the module index
158 // is valid, so they are compatible if and only if they refer to the same
159 // module.
160 return Modi == R.Modi;
161}
162
163Error DbiModuleList::initialize(BinaryStreamRef ModInfo,
164 BinaryStreamRef FileInfo) {
165 if (auto EC = initializeModInfo(ModInfo))
166 return EC;
167 if (auto EC = initializeFileInfo(FileInfo))
168 return EC;
169
170 return Error::success();
171}
172
173Error DbiModuleList::initializeModInfo(BinaryStreamRef ModInfo) {
174 ModInfoSubstream = ModInfo;
175
176 if (ModInfo.getLength() == 0)
177 return Error::success();
178
179 BinaryStreamReader Reader(ModInfo);
180
181 if (auto EC = Reader.readArray(Array&: Descriptors, Size: ModInfo.getLength()))
182 return EC;
183
184 return Error::success();
185}
186
187Error DbiModuleList::initializeFileInfo(BinaryStreamRef FileInfo) {
188 FileInfoSubstream = FileInfo;
189
190 if (FileInfo.getLength() == 0)
191 return Error::success();
192
193 BinaryStreamReader FISR(FileInfo);
194 if (auto EC = FISR.readObject(Dest&: FileInfoHeader))
195 return EC;
196
197 // First is an array of `NumModules` module indices. This does not seem to be
198 // used for anything meaningful, so we ignore it.
199 FixedStreamArray<support::ulittle16_t> ModuleIndices;
200 if (auto EC = FISR.readArray(Array&: ModuleIndices, NumItems: FileInfoHeader->NumModules))
201 return EC;
202 if (auto EC = FISR.readArray(Array&: ModFileCountArray, NumItems: FileInfoHeader->NumModules))
203 return EC;
204
205 // Compute the real number of source files. We can't trust the value in
206 // `FileInfoHeader->NumSourceFiles` because it is a unit16, and the sum of all
207 // source file counts might be larger than a unit16. So we compute the real
208 // count by summing up the individual counts.
209 uint32_t NumSourceFiles = 0;
210 for (auto Count : ModFileCountArray)
211 NumSourceFiles += Count;
212
213 // In the reference implementation, this array is where the pointer documented
214 // at the definition of ModuleInfoHeader::FileNameOffs points to. Note that
215 // although the field in ModuleInfoHeader is ignored this array is not, as it
216 // is the authority on where each filename begins in the names buffer.
217 if (auto EC = FISR.readArray(Array&: FileNameOffsets, NumItems: NumSourceFiles))
218 return EC;
219
220 if (auto EC = FISR.readStreamRef(Ref&: NamesBuffer))
221 return EC;
222
223 auto DescriptorIter = Descriptors.begin();
224 uint32_t NextFileIndex = 0;
225 ModuleInitialFileIndex.resize(new_size: FileInfoHeader->NumModules);
226 ModuleDescriptorOffsets.resize(new_size: FileInfoHeader->NumModules);
227 for (size_t I = 0; I < FileInfoHeader->NumModules; ++I) {
228 assert(DescriptorIter != Descriptors.end());
229 ModuleInitialFileIndex[I] = NextFileIndex;
230 ModuleDescriptorOffsets[I] = DescriptorIter.offset();
231
232 NextFileIndex += ModFileCountArray[I];
233 ++DescriptorIter;
234 }
235
236 assert(DescriptorIter == Descriptors.end());
237 assert(NextFileIndex == NumSourceFiles);
238
239 return Error::success();
240}
241
242uint32_t DbiModuleList::getModuleCount() const {
243 // Workaround to avoid the crash until upstream issue is fixed:
244 // https://github.com/llvm/llvm-project/issues/55214
245 return FileInfoHeader ? FileInfoHeader->NumModules : 0;
246}
247
248uint32_t DbiModuleList::getSourceFileCount() const {
249 return FileNameOffsets.size();
250}
251
252uint16_t DbiModuleList::getSourceFileCount(uint32_t Modi) const {
253 return ModFileCountArray[Modi];
254}
255
256DbiModuleDescriptor DbiModuleList::getModuleDescriptor(uint32_t Modi) const {
257 assert(Modi < getModuleCount());
258 uint32_t Offset = ModuleDescriptorOffsets[Modi];
259 auto Iter = Descriptors.at(Offset);
260 assert(Iter != Descriptors.end());
261 return *Iter;
262}
263
264iterator_range<DbiModuleSourceFilesIterator>
265DbiModuleList::source_files(uint32_t Modi) const {
266 return make_range<DbiModuleSourceFilesIterator>(
267 x: DbiModuleSourceFilesIterator(*this, Modi, 0),
268 y: DbiModuleSourceFilesIterator());
269}
270
271Expected<StringRef> DbiModuleList::getFileName(uint32_t Index) const {
272 BinaryStreamReader Names(NamesBuffer);
273 if (Index >= getSourceFileCount())
274 return make_error<RawError>(Args: raw_error_code::index_out_of_bounds);
275
276 uint32_t FileOffset = FileNameOffsets[Index];
277 Names.setOffset(FileOffset);
278 StringRef Name;
279 if (auto EC = Names.readCString(Dest&: Name))
280 return std::move(EC);
281 return Name;
282}
283