ASoA.cxx

Go to the documentation of this file.

// Copyright 2019-2020 CERN and copyright holders of ALICE O2.
// See https://alice-o2.web.cern.ch/copyright for details of the copyright holders.
// All rights not expressly granted are reserved.
//
// This software is distributed under the terms of the GNU General Public
// License v3 (GPL Version 3), copied verbatim in the file "COPYING".
//
// In applying this license CERN does not waive the privileges and immunities
// granted to it by virtue of its status as an Intergovernmental Organization
// or submit itself to any jurisdiction.
 
#include "Framework/ASoA.h"
#include "ArrowDebugHelpers.h"
#include "Framework/RuntimeError.h"
#include <arrow/util/key_value_metadata.h>
#include <arrow/util/config.h>
#include <TMemFile.h>
#include <TClass.h>
#include <TTree.h>
#include <TH1.h>
#include <TError.h>
 
namespace o2::soa
{
void accessingInvalidIndexFor(const char* getter)
{
  throw o2::framework::runtime_error_f("Accessing invalid index for %s", getter);
}
void dereferenceWithWrongType(const char* getter, const char* target)
{
  throw o2::framework::runtime_error_f("Trying to dereference index with a wrong type in %s_as<T> for base target \"%s\". Note that if you have several compatible index targets in your process() signature, the last one will be the one actually bound.", getter, target);
}
void missingFilterDeclaration(int hash, int ai)
{
  throw o2::framework::runtime_error_f("Null selection for %d (arg %d), missing Filter declaration?", hash, ai);
}
 
void getterNotFound(const char* targetColumnLabel)
{
  throw o2::framework::runtime_error_f("Getter for \"%s\" not found", targetColumnLabel);
}
 
void emptyColumnLabel()
{
  throw framework::runtime_error("columnLabel: must not be empty");
}
 
SelectionVector selectionToVector(gandiva::Selection const& sel)
{
  SelectionVector rows;
  rows.resize(sel->GetNumSlots());
  for (auto i = 0; i < sel->GetNumSlots(); ++i) {
    rows[i] = sel->GetIndex(i);
  }
  return rows;
}
 
SelectionVector sliceSelection(std::span<int64_t const> const& mSelectedRows, int64_t nrows, uint64_t offset)
{
  auto start = offset;
  auto end = start + nrows;
  auto start_iterator = std::lower_bound(mSelectedRows.begin(), mSelectedRows.end(), start);
  auto stop_iterator = std::lower_bound(start_iterator, mSelectedRows.end(), end);
  SelectionVector slicedSelection{start_iterator, stop_iterator};
  std::ranges::transform(slicedSelection.begin(), slicedSelection.end(), slicedSelection.begin(),
                         [&start](int64_t idx) {
                           return idx - static_cast<int64_t>(start);
                         });
  return slicedSelection;
}
 
namespace
{
template <typename T>
  requires(std::same_as<T, std::string>)
auto makeString(T const& str)
{
  return str.c_str();
}
template <typename T>
  requires(std::same_as<T, const char*>)
auto makeString(T const& str)
{
  return str;
}
 
template <typename T>
void canNotJoin(std::vector<std::shared_ptr<arrow::Table>> const& tables, std::span<T> labels)
{
  for (auto i = 0U; i < tables.size() - 1; ++i) {
    if (tables[i]->num_rows() != tables[i + 1]->num_rows()) {
      throw o2::framework::runtime_error_f("Tables %s and %s have different sizes (%lld vs %lld) and cannot be joined!",
                                           makeString(labels[i]), makeString(labels[i + 1]), (long long)tables[i]->num_rows(), (long long)tables[i + 1]->num_rows());
    }
  }
}
 
template <typename T>
void IncompatibleRanges(std::vector<ArrowTableRef> const& tables, std::span<T> labels)
{
  auto loc = std::ranges::adjacent_find(tables, [](auto const& l, auto const& r) { return l.range != r.range; });
  if (loc != std::ranges::cend(tables)) {
    auto pos = std::distance(tables.begin(), loc);
    auto next = loc + 1;
    if (loc->range.size != next->range.size) {
      if (labels.empty()) {
        throw o2::framework::runtime_error_f("Tables at positions %d and %d have different sizes (%lld vs %lld) and cannot be joined!",
                                             (int)pos, (int)pos + 1, (long long)loc->range.size, (long long)next->range.size);
      }
      throw o2::framework::runtime_error_f("Tables %s and %s have different sizes (%lld vs %lld) and cannot be joined!",
                                           makeString(labels[pos]), makeString(labels[pos + 1]), (long long)loc->range.size, (long long)next->range.size);
    }
    if (labels.empty()) {
      throw o2::framework::runtime_error_f("Incompatible ranges at %d: (%llu, %lld) vs. (%llu, %lld)",
                                           (int)pos, (unsigned long long)loc->range.offset, (long long)loc->range.size, (unsigned long long)next->range.offset, (long long)next->range.size);
    }
    throw o2::framework::runtime_error_f("Incompatible ranges at %d between %s and %s: (%llu, %lld) vs. (%llu, %lld)",
                                         (int)pos, makeString(labels[pos]), makeString(labels[pos + 1]), (unsigned long long)loc->range.offset, (long long)loc->range.size, (unsigned long long)next->range.offset, (long long)next->range.size);
  }
}
 
std::shared_ptr<arrow::Table> joinTablesImpl(std::ranges::input_range auto tables)
{
  std::vector<std::shared_ptr<arrow::Field>> fields;
  std::vector<std::shared_ptr<arrow::ChunkedArray>> columns;
  bool notEmpty = (tables.front()->num_rows() != 0);
  std::ranges::for_each(tables, [&fields, &columns, notEmpty](auto const& t) {
    std::ranges::copy(t->fields(), std::back_inserter(fields));
    if (notEmpty) {
      std::ranges::copy(t->columns(), std::back_inserter(columns));
    }
  });
  auto schema = std::make_shared<arrow::Schema>(fields);
  return arrow::Table::Make(schema, columns);
}
 
template <typename T>
ArrowTableRef joinTablesImpl(std::ranges::input_range auto tables, std::span<T> labels)
{
  if (tables.size() == 1) {
    return tables.front();
  }
  IncompatibleRanges(tables, labels);
  ArrowRange commonRange{tables.front().range};
  return {joinTablesImpl(tables), commonRange};
}
} // namespace
 
o2::soa::ArrowTableRef ArrowHelpers::joinTables(std::vector<std::shared_ptr<arrow::Table>>&& tables)
{
  std::vector<ArrowTableRef> refs;
  std::ranges::transform(tables, std::back_inserter(refs), [](auto const& table) { return ArrowTableRef{table}; });
  return joinTablesImpl(refs, std::span<const char* const>());
}
 
o2::soa::ArrowTableRef ArrowHelpers::joinTables(std::vector<o2::soa::ArrowTableRef>&& tables)
{
  return joinTablesImpl(tables, std::span<const char* const>());
}
 
o2::soa::ArrowTableRef ArrowHelpers::joinTables(std::vector<o2::soa::ArrowTableRef>&& tables, std::span<const char* const> labels)
{
  return joinTablesImpl(tables, labels);
}
 
o2::soa::ArrowTableRef ArrowHelpers::joinTables(std::vector<o2::soa::ArrowTableRef>&& tables, std::span<const std::string> labels)
{
  return joinTablesImpl(tables, labels);
}
 
o2::soa::ArrowTableRef ArrowHelpers::joinTables(std::vector<std::shared_ptr<arrow::Table>>&& tables, std::span<const char* const> labels)
{
  canNotJoin(tables, labels);
  return o2::soa::ArrowTableRef{joinTablesImpl(tables)};
}
 
o2::soa::ArrowTableRef ArrowHelpers::joinTables(std::vector<std::shared_ptr<arrow::Table>>&& tables, std::span<const std::string> labels)
{
  canNotJoin(tables, labels);
  return o2::soa::ArrowTableRef{joinTablesImpl(tables)};
}
 
o2::soa::ArrowTableRef ArrowHelpers::concatTables(std::vector<o2::soa::ArrowTableRef>&& tables)
{
  if (tables.size() == 1) {
    return tables.front();
  }
  std::vector<std::shared_ptr<arrow::ChunkedArray>> columns;
  std::vector<std::shared_ptr<arrow::Field>> resultFields = tables.front()->schema()->fields();
  auto compareFields = [](std::shared_ptr<arrow::Field> const& f1, std::shared_ptr<arrow::Field> const& f2) {
    // Let's do this with stable sorting.
    return (!f1->Equals(f2)) && (f1->name() < f2->name());
  };
 
  for (auto i = 1; i < tables.size(); ++i) {
    auto const& fields = tables[i]->fields();
    std::vector<std::shared_ptr<arrow::Field>> intersection;
    std::ranges::set_intersection(resultFields, fields, std::back_inserter(intersection), compareFields);
    resultFields.swap(intersection);
  }
 
  for (auto const& field : resultFields) {
    arrow::ArrayVector chunks;
    for (auto const& table : tables) {
      auto ci = table->schema()->GetFieldIndex(field->name());
      if (ci == -1) {
        throw framework::runtime_error_f("Unable to find field {}", field->name().c_str());
      }
      auto column = table->column(ci);
      auto otherChunks = column->chunks();
      chunks.insert(chunks.end(), otherChunks.begin(), otherChunks.end());
    }
    columns.push_back(std::make_shared<arrow::ChunkedArray>(chunks));
  }
 
  return {arrow::Table::Make(std::make_shared<arrow::Schema>(resultFields), columns)};
}
 
arrow::ChunkedArray* getIndexFromLabel(arrow::Table* table, std::string_view label)
{
  // Take the exact-match common case first (string_view comparison checks length
  // then memcmp), and fall back to a case-insensitive scan only when the labels
  // differ in case.
  auto field = std::ranges::find_if(table->schema()->fields(), [label](std::shared_ptr<arrow::Field> const& f) {
    std::string_view name = f->name();
    return label == name ||
           std::ranges::equal(label, name, [](char c1, char c2) {
             return asciiToLower(c1) == asciiToLower(c2);
           });
  });
  if (field == table->schema()->fields().end()) {
    o2::framework::throw_error(o2::framework::runtime_error_f("Unable to find column with label %s.", label));
  }
  return table->column(std::distance(table->schema()->fields().begin(), field)).get();
}
 
void notBoundTable(const char* tableName)
{
  throw o2::framework::runtime_error_f("Index pointing to %s is not bound! Did you subscribe to the table?", tableName);
}
 
void notFoundColumn(const char* label, const char* key)
{
  throw o2::framework::runtime_error_f(R"(Preslice not valid: table "%s" (or join based on it) does not have column "%s")", label, key);
}
 
void missingOptionalPreslice(const char* label, const char* key)
{
  throw o2::framework::runtime_error_f(R"(Optional Preslice with missing binding used: table "%s" (or join based on it) does not have column "%s")", label, key);
}
 
void* extractCCDBPayload(char* payload, size_t size, TClass const* cl, const char* what)
{
  Int_t previousErrorLevel = gErrorIgnoreLevel;
  gErrorIgnoreLevel = kFatal;
  // does it have a flattened headers map attached in the end?
  TMemFile file("name", (char*)payload, size, "READ");
  gErrorIgnoreLevel = previousErrorLevel;
  if (file.IsZombie()) {
    return nullptr;
  }
 
  if (!cl) {
    return nullptr;
  }
  auto object = file.GetObjectChecked(what, cl);
  if (!object) {
    // it could be that object was stored with previous convention
    // where the classname was taken as key
    std::string objectName(cl->GetName());
    objectName.erase(std::find_if(objectName.rbegin(), objectName.rend(), [](unsigned char ch) {
                       return !std::isspace(ch);
                     }).base(),
                     objectName.end());
    objectName.erase(objectName.begin(), std::find_if(objectName.begin(), objectName.end(), [](unsigned char ch) {
                       return !std::isspace(ch);
                     }));
 
    object = file.GetObjectChecked(objectName.c_str(), cl);
    LOG(warn) << "Did not find object under expected name " << what;
    if (!object) {
      return nullptr;
    }
    LOG(warn) << "Found object under deprecated name " << cl->GetName();
  }
  auto result = object;
  // We need to handle some specific cases as ROOT ties them deeply
  // to the file they are contained in
  if (cl->IsTObject()) {
    // make a clone
    // detach from the file
    auto tree = dynamic_cast<TTree*>((TObject*)object);
    if (tree) {
      tree->LoadBaskets(0x1L << 32); // make tree memory based
      tree->SetDirectory(nullptr);
      result = tree;
    } else {
      auto h = dynamic_cast<TH1*>((TObject*)object);
      if (h) {
        h->SetDirectory(nullptr);
        result = h;
      }
    }
  }
  return result;
}
 
std::function<framework::ConcreteDataMatcher(framework::ConcreteDataMatcher&&)> originReplacement(header::DataOrigin newOrigin)
{
  return [newOrigin](framework::ConcreteDataMatcher&& m) {
    if ((m.origin == header::DataOrigin{"AOD"}) && (newOrigin != header::DataOrigin{"AOD"})) {
      m.origin = newOrigin;
    }
    return m;
  };
}
 
} // namespace o2::soa
 
namespace o2::framework
{
std::string cutString(std::string&& str)
{
  auto pos = str.find('_');
  if (pos != std::string::npos) {
    str.erase(pos);
  }
  return str;
}
 
std::string strToUpper(std::string&& str)
{
  std::transform(str.begin(), str.end(), str.begin(), [](unsigned char c) { return std::toupper(c); });
  return str;
}
 
bool PreslicePolicyBase::isMissing() const
{
  return binding == "[MISSING]";
}
 
Entry const& PreslicePolicyBase::getBindingKey() const
{
  return bindingKey;
}
 
void PreslicePolicySorted::updateSliceInfo(SliceInfoPtr&& si)
{
  sliceInfo = si;
}
 
void PreslicePolicyGeneral::updateSliceInfo(SliceInfoUnsortedPtr&& si)
{
  sliceInfo = si;
}
 
o2::soa::ArrowTableRef PreslicePolicySorted::getSliceFor(int value, o2::soa::ArrowTableRef const& input) const
{
  auto [offset_, count] = this->sliceInfo.getSliceFor(value);
  return input.slice({static_cast<uint64_t>(offset_), count});
}
 
std::span<const int64_t> PreslicePolicyGeneral::getSliceFor(int value) const
{
  return this->sliceInfo.getSliceFor(value);
}
} // namespace o2::framework