Zengtudor/ccls
1
0
Fork 0
mirror of https://github.com/MaskRay/ccls.git synced 2025-01-31 18:00:26 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Spend time on indexer thread doing a higher quality IndexUpdate merge to reduce querydb import time

Browse Source
This commit is contained in:
Jacob Dufault 2017-04-16 01:31:28 -07:00
parent 04a855ebc7
commit 55f527b975
1 changed files with 44 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

59
src/query.cc
View File

@ -58,6 +58,38 @@ QueryVar::DefUpdate ToQuery(const IdMap& id_map, const IndexedVarDef::Def& var)
return result; return result;
} }
// Adds the mergeable updates in |source| to |dest|. If a mergeable update for
// the destination type already exists, it will be combined. This makes merging
// updates take longer but reduces import time on the querydb thread.
template <typename TId, typename TValue>
void AddMergeableRange(
std::vector<MergeableUpdate<TId, TValue>>* dest,
const std::vector<MergeableUpdate<TId, TValue>>& source) {
// TODO: Consider caching the lookup table. It can probably save even more
// time at the cost of some additional memory.
// Build lookup table.
google::dense_hash_map<TId, size_t> id_to_index;
id_to_index.set_empty_key(TId(-1));
id_to_index.resize(dest->size());
for (size_t i = 0; i < dest->size(); ++i)
id_to_index[(*dest)[i].id] = i;
// Add entries. Try to add them to an existing entry.
for (const auto& entry : source) {
auto it = id_to_index.find(entry.id);
if (it != id_to_index.end()) {
AddRange(&(*dest)[it->second].to_add, entry.to_add);
AddRange(&(*dest)[it->second].to_remove, entry.to_remove);
}
else {
dest->push_back(entry);
}
}
}
} // namespace } // namespace
@ -539,34 +571,31 @@ IndexUpdate::IndexUpdate(const IdMap& previous_id_map, const IdMap& current_id_m
} }
void IndexUpdate::Merge(const IndexUpdate& update) { void IndexUpdate::Merge(const IndexUpdate& update) {
// TODO: When merging we should look at existing entries #define INDEX_UPDATE_APPEND(name) \
// and join them so we only run the map access once in the
// querydb. This will probably be expensive though, we may
// want to cache additional state so we could do a set
// intersection.
#define INDEX_UPDATE_MERGE(name) \
AddRange(&name, update.name); AddRange(&name, update.name);
#define INDEX_UPDATE_MERGE(name) \
AddMergeableRange(&name, update.name);
INDEX_UPDATE_MERGE(files_removed); INDEX_UPDATE_APPEND(files_removed);
INDEX_UPDATE_MERGE(files_def_update); INDEX_UPDATE_APPEND(files_def_update);
INDEX_UPDATE_MERGE(types_removed); INDEX_UPDATE_APPEND(types_removed);
INDEX_UPDATE_MERGE(types_def_update); INDEX_UPDATE_APPEND(types_def_update);
INDEX_UPDATE_MERGE(types_derived); INDEX_UPDATE_MERGE(types_derived);
INDEX_UPDATE_MERGE(types_instantiations); INDEX_UPDATE_MERGE(types_instantiations);
INDEX_UPDATE_MERGE(types_uses); INDEX_UPDATE_MERGE(types_uses);
INDEX_UPDATE_MERGE(funcs_removed); INDEX_UPDATE_APPEND(funcs_removed);
INDEX_UPDATE_MERGE(funcs_def_update); INDEX_UPDATE_APPEND(funcs_def_update);
INDEX_UPDATE_MERGE(funcs_declarations); INDEX_UPDATE_MERGE(funcs_declarations);
INDEX_UPDATE_MERGE(funcs_derived); INDEX_UPDATE_MERGE(funcs_derived);
INDEX_UPDATE_MERGE(funcs_callers); INDEX_UPDATE_MERGE(funcs_callers);
INDEX_UPDATE_MERGE(vars_removed); INDEX_UPDATE_APPEND(vars_removed);
INDEX_UPDATE_MERGE(vars_def_update); INDEX_UPDATE_APPEND(vars_def_update);
INDEX_UPDATE_MERGE(vars_uses); INDEX_UPDATE_MERGE(vars_uses);
#undef INDEX_UPDATE_APPEND
#undef INDEX_UPDATE_MERGE #undef INDEX_UPDATE_MERGE
} }