1 // Copyright 2012 Google Inc. All Rights Reserved.
3 // Licensed under the Apache License, Version 2.0 (the "License");
4 // you may not use this file except in compliance with the License.
5 // You may obtain a copy of the License at
7 // http://www.apache.org/licenses/LICENSE-2.0
9 // Unless required by applicable law or agreed to in writing, software
10 // distributed under the License is distributed on an "AS IS" BASIS,
11 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 // See the License for the specific language governing permissions and
13 // limitations under the License.
30 // The version is stored as 4 bytes after the signature and also serves as a
31 // byte order mark. Signature and version combined are 16 bytes long.
32 const char kFileSignature[] = "# ninjadeps\n";
33 const int kCurrentVersion = 1;
35 // Since the size field is 2 bytes and the top bit marks deps entries, a single
36 // record can be at most 32 kB. Set the buffer size to this and flush the file
37 // buffer after every record to make sure records aren't written partially.
38 const int kMaxBufferSize = 1 << 15;
44 bool DepsLog::OpenForWrite(const string& path, string* err) {
45 if (needs_recompaction_) {
47 if (!Recompact(path, err))
51 file_ = fopen(path.c_str(), "ab");
53 *err = strerror(errno);
56 setvbuf(file_, NULL, _IOFBF, kMaxBufferSize);
57 SetCloseOnExec(fileno(file_));
59 // Opening a file in append mode doesn't set the file pointer to the file's
60 // end on Windows. Do that explicitly.
61 fseek(file_, 0, SEEK_END);
63 if (ftell(file_) == 0) {
64 if (fwrite(kFileSignature, sizeof(kFileSignature) - 1, 1, file_) < 1) {
65 *err = strerror(errno);
68 if (fwrite(&kCurrentVersion, 4, 1, file_) < 1) {
69 *err = strerror(errno);
73 if (fflush(file_) != 0) {
74 *err = strerror(errno);
80 bool DepsLog::RecordDeps(Node* node, TimeStamp mtime,
81 const vector<Node*>& nodes) {
82 return RecordDeps(node, mtime, nodes.size(),
83 nodes.empty() ? NULL : (Node**)&nodes.front());
86 bool DepsLog::RecordDeps(Node* node, TimeStamp mtime,
87 int node_count, Node** nodes) {
88 // Track whether there's any new data to be recorded.
89 bool made_change = false;
91 // Assign ids to all nodes that are missing one.
97 for (int i = 0; i < node_count; ++i) {
98 if (nodes[i]->id() < 0) {
99 if (!RecordId(nodes[i]))
105 // See if the new data is different than the existing data, if any.
107 Deps* deps = GetDeps(node);
109 deps->mtime != mtime ||
110 deps->node_count != node_count) {
113 for (int i = 0; i < node_count; ++i) {
114 if (deps->nodes[i] != nodes[i]) {
122 // Don't write anything if there's no new info.
126 // Update on-disk representation.
127 uint16_t size = 4 * (1 + 1 + (uint16_t)node_count);
128 size |= 0x8000; // Deps record: set high bit.
129 if (fwrite(&size, 2, 1, file_) < 1)
132 if (fwrite(&id, 4, 1, file_) < 1)
134 int timestamp = mtime;
135 if (fwrite(×tamp, 4, 1, file_) < 1)
137 for (int i = 0; i < node_count; ++i) {
139 if (fwrite(&id, 4, 1, file_) < 1)
142 if (fflush(file_) != 0)
145 // Update in-memory representation.
146 Deps* deps = new Deps(mtime, node_count);
147 for (int i = 0; i < node_count; ++i)
148 deps->nodes[i] = nodes[i];
149 UpdateDeps(node->id(), deps);
154 void DepsLog::Close() {
160 bool DepsLog::Load(const string& path, State* state, string* err) {
161 METRIC_RECORD(".ninja_deps load");
163 FILE* f = fopen(path.c_str(), "rb");
167 *err = strerror(errno);
171 bool valid_header = true;
173 if (!fgets(buf, sizeof(buf), f) || fread(&version, 4, 1, f) < 1)
174 valid_header = false;
175 if (!valid_header || strcmp(buf, kFileSignature) != 0 ||
176 version != kCurrentVersion) {
177 *err = "bad deps log signature or version; starting over";
179 unlink(path.c_str());
180 // Don't report this as a failure. An empty deps log will cause
181 // us to rebuild the outputs anyway.
186 bool read_failed = false;
187 int unique_dep_record_count = 0;
188 int total_dep_record_count = 0;
193 if (fread(&size, 2, 1, f) < 1) {
198 bool is_deps = (size >> 15) != 0;
199 size = size & 0x7FFF;
201 if (fread(buf, size, 1, f) < 1) {
207 assert(size % 4 == 0);
208 int* deps_data = reinterpret_cast<int*>(buf);
209 int out_id = deps_data[0];
210 int mtime = deps_data[1];
212 int deps_count = (size / 4) - 2;
214 Deps* deps = new Deps(mtime, deps_count);
215 for (int i = 0; i < deps_count; ++i) {
216 assert(deps_data[i] < (int)nodes_.size());
217 assert(nodes_[deps_data[i]]);
218 deps->nodes[i] = nodes_[deps_data[i]];
221 total_dep_record_count++;
222 if (!UpdateDeps(out_id, deps))
223 ++unique_dep_record_count;
225 StringPiece path(buf, size);
226 Node* node = state->GetNode(path);
227 assert(node->id() < 0);
228 node->set_id(nodes_.size());
229 nodes_.push_back(node);
234 // An error occurred while loading; try to recover by truncating the
235 // file to the last fully-read record.
237 *err = strerror(ferror(f));
239 *err = "premature end of file";
243 if (!Truncate(path.c_str(), offset, err))
246 // The truncate succeeded; we'll just report the load error as a
247 // warning because the build can proceed.
248 *err += "; recovering";
254 // Rebuild the log if there are too many dead records.
255 int kMinCompactionEntryCount = 1000;
256 int kCompactionRatio = 3;
257 if (total_dep_record_count > kMinCompactionEntryCount &&
258 total_dep_record_count > unique_dep_record_count * kCompactionRatio) {
259 needs_recompaction_ = true;
265 DepsLog::Deps* DepsLog::GetDeps(Node* node) {
266 // Abort if the node has no id (never referenced in the deps) or if
267 // there's no deps recorded for the node.
268 if (node->id() < 0 || node->id() >= (int)deps_.size())
270 return deps_[node->id()];
273 bool DepsLog::Recompact(const string& path, string* err) {
274 METRIC_RECORD(".ninja_deps recompact");
275 printf("Recompacting deps...\n");
277 string temp_path = path + ".recompact";
279 // OpenForWrite() opens for append. Make sure it's not appending to a
280 // left-over file from a previous recompaction attempt that crashed somehow.
281 unlink(temp_path.c_str());
284 if (!new_log.OpenForWrite(temp_path, err))
287 // Clear all known ids so that new ones can be reassigned. The new indices
288 // will refer to the ordering in new_log, not in the current log.
289 for (vector<Node*>::iterator i = nodes_.begin(); i != nodes_.end(); ++i)
292 // Write out all deps again.
293 for (int old_id = 0; old_id < (int)deps_.size(); ++old_id) {
294 Deps* deps = deps_[old_id];
295 if (!deps) continue; // If nodes_[old_id] is a leaf, it has no deps.
297 if (!new_log.RecordDeps(nodes_[old_id], deps->mtime,
298 deps->node_count, deps->nodes)) {
306 // All nodes now have ids that refer to new_log, so steal its data.
307 deps_.swap(new_log.deps_);
308 nodes_.swap(new_log.nodes_);
310 if (unlink(path.c_str()) < 0) {
311 *err = strerror(errno);
315 if (rename(temp_path.c_str(), path.c_str()) < 0) {
316 *err = strerror(errno);
323 bool DepsLog::UpdateDeps(int out_id, Deps* deps) {
324 if (out_id >= (int)deps_.size())
325 deps_.resize(out_id + 1);
327 bool delete_old = deps_[out_id] != NULL;
329 delete deps_[out_id];
330 deps_[out_id] = deps;
334 bool DepsLog::RecordId(Node* node) {
335 uint16_t size = (uint16_t)node->path().size();
336 if (fwrite(&size, 2, 1, file_) < 1)
338 if (fwrite(node->path().data(), node->path().size(), 1, file_) < 1)
339 return false; // assuming node->path().size() > 0
340 if (fflush(file_) != 0)
343 node->set_id(nodes_.size());
344 nodes_.push_back(node);