[Breaking] Require format to be specified in input URI. (#9077)

Previously, we use `libsvm` as default when format is not specified. However, the dmlc
data parser is not particularly robust against errors, and the most common type of error
is undefined format.

Along with which, we will recommend users to use other data loader instead. We will
continue the maintenance of the parsers as it's currently used for many internal tests
including federated learning.

tests/cpp/common/test_hist_util.h

@@ -88,7 +88,8 @@ inline std::shared_ptr<DMatrix> GetExternalMemoryDMatrixFromData(
     fo << row_data.str() << "\n";
   }
   fo.close();
-  return std::shared_ptr<DMatrix>(DMatrix::Load(tmp_file + "#" + tmp_file + ".cache"));
+  return std::shared_ptr<DMatrix>(
+      DMatrix::Load(tmp_file + "?format=libsvm" + "#" + tmp_file + ".cache"));
 }
 
 // Test that elements are approximately equally distributed among bins

tests/cpp/data/test_file_iterator.cc

@@ -29,16 +29,16 @@ TEST(FileIterator, Basic) {
   {
     auto zpath = tmpdir.path + "/0-based.svm";
     CreateBigTestData(zpath, 3 * 64, true);
-    zpath += "?indexing_mode=0";
-    FileIterator iter{zpath, 0, 1, "libsvm"};
+    zpath += "?indexing_mode=0&format=libsvm";
+    FileIterator iter{zpath, 0, 1};
     check_n_features(&iter);
   }
 
   {
     auto opath = tmpdir.path + "/1-based.svm";
     CreateBigTestData(opath, 3 * 64, false);
-    opath += "?indexing_mode=1";
-    FileIterator iter{opath, 0, 1, "libsvm"};
+    opath += "?indexing_mode=1&format=libsvm";
+    FileIterator iter{opath, 0, 1};
     check_n_features(&iter);
   }
 }

tests/cpp/data/test_metainfo.cc

@@ -157,8 +157,7 @@ TEST(MetaInfo, LoadQid) {
   dmlc::TemporaryDirectory tempdir;
   std::string tmp_file = tempdir.path + "/qid_test.libsvm";
   {
-    std::unique_ptr<dmlc::Stream> fs(
-      dmlc::Stream::Create(tmp_file.c_str(), "w"));
+    std::unique_ptr<dmlc::Stream> fs(dmlc::Stream::Create(tmp_file.c_str(), "w"));
     dmlc::ostream os(fs.get());
     os << R"qid(3 qid:1 1:1 2:1 3:0 4:0.2 5:0
                 2 qid:1 1:0 2:0 3:1 4:0.1 5:1
@@ -175,7 +174,7 @@ TEST(MetaInfo, LoadQid) {
     os.set_stream(nullptr);
   }
   std::unique_ptr<xgboost::DMatrix> dmat(
-    xgboost::DMatrix::Load(tmp_file, true, xgboost::DataSplitMode::kRow, "libsvm"));
+      xgboost::DMatrix::Load(tmp_file + "?format=libsvm", true, xgboost::DataSplitMode::kRow));
 
   const xgboost::MetaInfo& info = dmat->Info();
   const std::vector<xgboost::bst_uint> expected_group_ptr{0, 4, 8, 12};

tests/cpp/data/test_simple_dmatrix.cc

@@ -17,11 +17,15 @@
 
 using namespace xgboost;  // NOLINT
 
+namespace {
+std::string UriSVM(std::string name) { return name + "?format=libsvm"; }
+}  // namespace
+
 TEST(SimpleDMatrix, MetaInfo) {
   dmlc::TemporaryDirectory tempdir;
   const std::string tmp_file = tempdir.path + "/simple.libsvm";
   CreateSimpleTestData(tmp_file);
-  xgboost::DMatrix *dmat = xgboost::DMatrix::Load(tmp_file);
+  xgboost::DMatrix *dmat = xgboost::DMatrix::Load(UriSVM(tmp_file));
 
   // Test the metadata that was parsed
   EXPECT_EQ(dmat->Info().num_row_, 2);
@@ -37,7 +41,7 @@ TEST(SimpleDMatrix, RowAccess) {
   dmlc::TemporaryDirectory tempdir;
   const std::string tmp_file = tempdir.path + "/simple.libsvm";
   CreateSimpleTestData(tmp_file);
-  xgboost::DMatrix *dmat = xgboost::DMatrix::Load(tmp_file, false);
+  xgboost::DMatrix *dmat = xgboost::DMatrix::Load(UriSVM(tmp_file), false);
 
   // Loop over the batches and count the records
   int64_t row_count = 0;
@@ -60,7 +64,7 @@ TEST(SimpleDMatrix, ColAccessWithoutBatches) {
   dmlc::TemporaryDirectory tempdir;
   const std::string tmp_file = tempdir.path + "/simple.libsvm";
   CreateSimpleTestData(tmp_file);
-  xgboost::DMatrix *dmat = xgboost::DMatrix::Load(tmp_file);
+  xgboost::DMatrix *dmat = xgboost::DMatrix::Load(UriSVM(tmp_file));
 
   ASSERT_TRUE(dmat->SingleColBlock());
 
@@ -387,7 +391,7 @@ TEST(SimpleDMatrix, SaveLoadBinary) {
   dmlc::TemporaryDirectory tempdir;
   const std::string tmp_file = tempdir.path + "/simple.libsvm";
   CreateSimpleTestData(tmp_file);
-  xgboost::DMatrix * dmat = xgboost::DMatrix::Load(tmp_file);
+  xgboost::DMatrix * dmat = xgboost::DMatrix::Load(UriSVM(tmp_file));
   data::SimpleDMatrix *simple_dmat = dynamic_cast<data::SimpleDMatrix*>(dmat);
 
   const std::string tmp_binfile = tempdir.path + "/csr_source.binary";

tests/cpp/data/test_sparse_page_dmatrix.cc

@@ -16,14 +16,19 @@
 #include "../helpers.h"
 
 using namespace xgboost;  // NOLINT
+namespace {
+std::string UriSVM(std::string name, std::string cache) {
+  return name + "?format=libsvm" + "#" + cache + ".cache";
+}
+}  // namespace
 
 template <typename Page>
 void TestSparseDMatrixLoadFile() {
   dmlc::TemporaryDirectory tmpdir;
   auto opath = tmpdir.path + "/1-based.svm";
   CreateBigTestData(opath, 3 * 64, false);
-  opath += "?indexing_mode=1";
-  data::FileIterator iter{opath, 0, 1, "libsvm"};
+  opath += "?indexing_mode=1&format=libsvm";
+  data::FileIterator iter{opath, 0, 1};
   auto n_threads = 0;
   data::SparsePageDMatrix m{&iter,
                             iter.Proxy(),
@@ -112,15 +117,13 @@ TEST(SparsePageDMatrix, MetaInfo) {
   size_t constexpr kEntries = 24;
   CreateBigTestData(tmp_file, kEntries);
 
-  xgboost::DMatrix *dmat = xgboost::DMatrix::Load(tmp_file + "#" + tmp_file + ".cache", false);
+  std::unique_ptr<DMatrix> dmat{xgboost::DMatrix::Load(UriSVM(tmp_file, tmp_file), false)};
 
   // Test the metadata that was parsed
   EXPECT_EQ(dmat->Info().num_row_, 8ul);
   EXPECT_EQ(dmat->Info().num_col_, 5ul);
   EXPECT_EQ(dmat->Info().num_nonzero_, kEntries);
   EXPECT_EQ(dmat->Info().labels.Size(), dmat->Info().num_row_);
-
-  delete dmat;
 }
 
 TEST(SparsePageDMatrix, RowAccess) {
@@ -139,7 +142,7 @@ TEST(SparsePageDMatrix, ColAccess) {
   dmlc::TemporaryDirectory tempdir;
   const std::string tmp_file = tempdir.path + "/simple.libsvm";
   CreateSimpleTestData(tmp_file);
-  xgboost::DMatrix *dmat = xgboost::DMatrix::Load(tmp_file + "#" + tmp_file + ".cache");
+  xgboost::DMatrix *dmat = xgboost::DMatrix::Load(UriSVM(tmp_file, tmp_file));
 
   // Loop over the batches and assert the data is as expected
   size_t iter = 0;
@@ -231,7 +234,7 @@ auto TestSparsePageDMatrixDeterminism(int32_t threads) {
   std::string filename = tempdir.path + "/simple.libsvm";
   CreateBigTestData(filename, 1 << 16);
 
-  data::FileIterator iter(filename, 0, 1, "auto");
+  data::FileIterator iter(filename + "?format=libsvm", 0, 1);
   std::unique_ptr<DMatrix> sparse{
       new data::SparsePageDMatrix{&iter, iter.Proxy(), data::fileiter::Reset, data::fileiter::Next,
                                   std::numeric_limits<float>::quiet_NaN(), threads, filename}};

tests/cpp/data/test_sparse_page_dmatrix.cu

@@ -13,7 +13,7 @@ TEST(SparsePageDMatrix, EllpackPage) {
   dmlc::TemporaryDirectory tempdir;
   const std::string tmp_file = tempdir.path + "/simple.libsvm";
   CreateSimpleTestData(tmp_file);
-  DMatrix* dmat = DMatrix::Load(tmp_file + "#" + tmp_file + ".cache");
+  DMatrix* dmat = DMatrix::Load(tmp_file + "?format=libsvm" + "#" + tmp_file + ".cache");
 
   // Loop over the batches and assert the data is as expected
   size_t n = 0;

tests/cpp/helpers.cc

@@ -548,7 +548,7 @@ std::unique_ptr<DMatrix> CreateSparsePageDMatrixWithRC(
   }
   fo.close();
 
-  std::string uri = tmp_file;
+  std::string uri = tmp_file + "?format=libsvm";
   if (page_size > 0) {
     uri += "#" + tmp_file + ".cache";
   }

tests/cpp/test_learner.cc

@@ -126,7 +126,8 @@ TEST(Learner, SLOW_CheckMultiBatch) {  // NOLINT
   dmlc::TemporaryDirectory tempdir;
   const std::string tmp_file = tempdir.path + "/big.libsvm";
   CreateBigTestData(tmp_file, 50000);
-  std::shared_ptr<DMatrix> dmat(xgboost::DMatrix::Load(tmp_file + "#" + tmp_file + ".cache"));
+  std::shared_ptr<DMatrix> dmat(
+      xgboost::DMatrix::Load(tmp_file + "?format=libsvm" + "#" + tmp_file + ".cache"));
   EXPECT_FALSE(dmat->SingleColBlock());
   size_t num_row = dmat->Info().num_row_;
   std::vector<bst_float> labels(num_row);