Support ritsuko's custom variable length string arrays in HDF5.

docs/specifications/build.R

@@ -1,7 +1,7 @@
 library(knitr)
 dir.create("compiled", showWarnings=FALSE)
 
-for (v in c("1.0", "1.1", "1.2", "1.3")) {
+for (v in c("1.0", "1.1", "1.2", "1.3", "1.4")) {
     .version <- package_version(v)
     knitr::knit("hdf5.Rmd", output=file.path("compiled", paste0("hdf5-", v, ".md")))
 }

docs/specifications/hdf5.Rmd

@@ -213,6 +213,39 @@ This should be interpreted as a boolean where a non-zero value specifies that we
 }
 ```
 
+```{r, results="asis", echo=FALSE}
+if (.version >= package_version("1.4")) {
+cat('### Variable length string arrays
+
+Arrays of strings can be stored in [**ritsuko**\'s custom variable length string (VLS) array](https://github.com/ArtifactDB/ritsuko).
+This is represented as a HDF5 group (`**/`) with the following attributes:
+
+- `uzuki_object`, a scalar string dataset containing the value `"vector"`.
+  This should use a datatype that can be represented by a UTF-8 encoded string.
+- `uzuki_type`, a scalar string dataset containing `"vls"`.
+
+This group should contain the `pointers` and `heap` datasets.
+
+- The `**/data` dataset should be a 1-dimensional or scalar dataset of a compound datatype of 2 members, `"offset"` and `"length"`.
+  Each member should be of a datatype that can be represented by an unsigned 64-bit integer.
+  If the dataset is scalar, the length of the VLS array is defined as 1.
+- The `**/heap` dataset should be a 1-dimensional dataset of unsigned 8-bit integers.
+
+Each entry of `**/data` refers to a slice `[offset, offset + length)` of the `**/heap` dataset.
+This slice defines a variable-length UTF-8 encoded string of length `length` - unless the slice contains a null terminator, in which case the string is defined as the interval to the first null. 
+Pointers may be in any order, overlapping or non-contiguous, as long as `[offset, offset + length)` lies within the boundaries of the heap.
+
+A `missing-value-placeholder` attribute on the `**/data` dataset may be present, defining a placeholder for missing values.
+The attribute should be a scalar and should be of any HDF5 string datatype that can be represented by a UTF-8 encoded string.
+An entry of `**/data` should be considered as missing if its corresponding string is equal to the placeholder.
+
+The group may also contain `**/names`, a 1-dimensional string dataset of length equal to `**/data`.
+This should use a datatype that can be represented by a UTF-8 encoded string.
+If `**/data` is a scalar, `**/names` should have length 1.
+')
+}
+```
+
 ### Nothing
 
 A "nothing" (a.k.a., "null", "none") value is represented as a HDF5 group with the following attributes:

include/uzuki2/parse_hdf5.hpp

@@ -20,6 +20,7 @@
 
 #include "ritsuko/ritsuko.hpp"
 #include "ritsuko/hdf5/hdf5.hpp"
+#include "ritsuko/hdf5/vls/vls.hpp"
 
 /**
  * @file parse_hdf5.hpp
@@ -286,6 +287,48 @@ std::shared_ptr<Base> parse_inner(const H5::Group& handle, Externals& ext, const
                 present.insert(std::move(x));
             }
 
+        } else if (vector_type == "vls" && !version.lt(1, 4)) {
+            ritsuko::hdf5::vls::validate_pointer_datatype(dhandle.getCompType(), 64, 64);
+            auto hhandle = ritsuko::hdf5::vls::open_heap(handle, "heap");
+            auto missingness = ritsuko::hdf5::open_and_load_optional_string_missing_placeholder(dhandle, "missing-value-placeholder");
+
+            auto ptr = Provisioner::new_String(len, named, is_scalar, StringVector::NONE);
+            output.reset(ptr);
+
+            if (is_scalar) {
+                ritsuko::hdf5::vls::Pointer<uint64_t, uint64_t> vlsptr;
+                dhandle.read(&vlsptr, ritsuko::hdf5::vls::define_pointer_datatype<uint64_t, uint64_t>());
+
+                hsize_t len = vlsptr.length;
+                H5::DataSpace mspace(1, &len);
+                hsize_t offset = vlsptr.offset;
+                hsize_t hlen = ritsuko::hdf5::get_1d_length(hhandle, false);
+                H5::DataSpace dspace(1, &hlen);
+                dspace.selectHyperslab(H5S_SELECT_SET, &len, &offset);
+
+                std::vector<uint8_t> buffer(vlsptr.length);
+                hhandle.read(buffer.data(), H5::PredType::NATIVE_UINT8, mspace, dspace);
+                auto cptr = reinterpret_cast<const char*>(buffer.data());
+                std::string str(cptr, cptr + ritsuko::hdf5::find_string_length(cptr, vlsptr.length));
+
+                if (missingness.has_value() && str == *missingness) {
+                    ptr->set_missing(0);
+                } else {
+                    ptr->set(0, std::move(str));
+                }
+
+            } else {
+                ritsuko::hdf5::vls::Stream1dArray<uint64_t, uint64_t> stream(&dhandle, &hhandle, len, buffer_size);
+                for (hsize_t i = 0; i < len; ++i, stream.next()) {
+                    auto x = stream.steal();
+                    if (missingness.has_value() && x == *missingness) {
+                        ptr->set_missing(i);
+                    } else {
+                        ptr->set(i, std::move(x));
+                    }
+                }
+            }
+
         } else if (vector_type == "string" || (version.equals(1, 0) && (vector_type == "date" || vector_type == "date-time"))) {
             StringVector::Format format = StringVector::NONE;
             if (version.equals(1, 0)) {

tests/CMakeLists.txt

@@ -23,6 +23,7 @@ add_executable(
     src/factor.cpp
     src/number.cpp
     src/string.cpp
+    src/vls.cpp
     src/date.cpp
     src/datetime.cpp
     src/external.cpp

tests/src/string.cpp

@@ -235,3 +235,5 @@ TEST(JsonStringTest, CheckError) {
      *** See integer.cpp for vector error tests. ***
      ***********************************************/
 }
+
+

tests/src/utils.h

@@ -99,15 +99,7 @@ H5::DataSet create_dataset(const H5::Group& parent, const std::string& name, con
     }
 
     auto dhandle = parent.createDataSet(name, dtype, dspace, cplist);
-
-    if constexpr(std::is_same<T, int>::value) {
-        dhandle.write(values.data(), H5::PredType::NATIVE_INT);
-    } else if constexpr(std::is_same<T, double>::value) {
-        dhandle.write(values.data(), H5::PredType::NATIVE_DOUBLE);
-    } else {
-        throw std::runtime_error("unknown type!");
-    }
-
+    dhandle.write(values.data(), ritsuko::hdf5::as_numeric_datatype<T>());
     return dhandle;
 }
 

tests/src/vls.cpp

@@ -0,0 +1,191 @@
+#include <gtest/gtest.h>
+#include <gmock/gmock.h>
+
+#include "uzuki2/parse_hdf5.hpp"
+
+#include "utils.h"
+
+TEST(Hdf5VlsTest, Basic) {
+    auto path = "TEST-vls.h5";
+    std::string heap = "abcdefghijklmno";
+    size_t nlen = 10;
+
+    {
+        H5::H5File handle(path, H5F_ACC_TRUNC);
+        auto vhandle = vector_opener(handle, "blub", "vls");
+        add_version(vhandle, "1.4");
+
+        auto hhandle = create_dataset(vhandle, "heap", heap.size(), H5::PredType::NATIVE_UINT8);
+        const unsigned char* hptr = reinterpret_cast<const unsigned char*>(heap.c_str());
+        hhandle.write(hptr, H5::PredType::NATIVE_UCHAR);
+
+        std::vector<ritsuko::hdf5::vls::Pointer<uint64_t, uint64_t> > pointers(nlen);
+        size_t n = 0;
+        for (size_t i = 0; i < nlen; ++i) {
+            pointers[i].offset = n;
+            size_t count = (i % 2) + 1; // for some interesting differences.
+            pointers[i].length = count;
+            n += count;
+        }
+        auto ptype = ritsuko::hdf5::vls::define_pointer_datatype<uint64_t, uint64_t>();
+        auto phandle = create_dataset(vhandle, "data", pointers.size(), ptype);
+        phandle.write(pointers.data(), ptype);
+    }
+
+    // Check that it works correctly.
+    {
+        auto parsed = load_hdf5(path, "blub");
+        EXPECT_EQ(parsed->type(), uzuki2::STRING);
+        auto sptr = static_cast<const DefaultStringVector*>(parsed.get());
+        EXPECT_EQ(sptr->size(), nlen);
+        std::vector<std::string> expected { "a", "bc", "d", "ef", "g", "hi", "j", "kl", "m", "no" };
+        EXPECT_EQ(sptr->base.values, expected);
+    }
+
+    // Adding a missing value placeholder.
+    {
+        {
+            H5::H5File handle(path, H5F_ACC_RDWR);
+            auto vhandle = handle.openDataSet("blub/data");
+            H5::StrType stype(0, H5T_VARIABLE);
+            auto ahandle = vhandle.createAttribute("missing-value-placeholder", stype, H5S_SCALAR);
+            ahandle.write(stype, std::string("hi"));
+        }
+
+        auto parsed = load_hdf5(path, "blub");
+        EXPECT_EQ(parsed->type(), uzuki2::STRING);
+        auto sptr = static_cast<const DefaultStringVector*>(parsed.get());
+        EXPECT_EQ(sptr->base.values[5], "ich bin missing"); // the test's missing placeholder.
+
+        // Adding the wrong missing value placeholder.
+        {
+            H5::H5File handle(path, H5F_ACC_RDWR);
+            auto vhandle = handle.openDataSet("blub/data");
+            vhandle.removeAttr("missing-value-placeholder");
+            vhandle.createAttribute("missing-value-placeholder", H5::PredType::NATIVE_INT, H5S_SCALAR);
+        }
+        expect_hdf5_error(path, "blub", "string datatype");
+
+        // Removing for the next checks.
+        {
+            H5::H5File handle(path, H5F_ACC_RDWR);
+            auto vhandle = handle.openDataSet("blub/data");
+            vhandle.removeAttr("missing-value-placeholder");
+        }
+    }
+}
+
+TEST(Hdf5VlsTest, Failures) {
+    auto path = "TEST-vls.h5";
+    std::string heap = "abcdefghijklmno";
+    size_t nlen = 10;
+
+    // Shortening the heap to check that we perform bounds checks on the pointers.
+    {
+        H5::H5File handle(path, H5F_ACC_TRUNC);
+        auto ghandle = vector_opener(handle, "blub", "vls");
+        add_version(ghandle, "1.4");
+
+        hsize_t zero = 0;
+        H5::DataSpace hspace(1, &zero);
+        ghandle.createDataSet("heap", H5::PredType::NATIVE_UINT8, hspace);
+
+        std::vector<ritsuko::hdf5::vls::Pointer<uint64_t, uint64_t> > pointers(nlen);
+        for (size_t i = 0; i < nlen; ++i) {
+            pointers[i].offset = i;
+            pointers[i].length = 1;
+        }
+        auto ptype = ritsuko::hdf5::vls::define_pointer_datatype<uint64_t, uint64_t>();
+        auto phandle = create_dataset(ghandle, "data", pointers.size(), ptype);
+        phandle.write(pointers.data(), ptype);
+    }
+    expect_hdf5_error(path, "blub", "out of range");
+
+    // Checking that we check for 64-bit unsigned integer types. 
+    {
+        H5::H5File handle(path, H5F_ACC_RDWR);
+        auto ghandle = handle.openGroup("blub");
+        ghandle.unlink("data");
+
+        std::vector<ritsuko::hdf5::vls::Pointer<int, int> > pointers(3);
+        for (auto& p : pointers) {
+            p.offset = 0;
+            p.length = 0;
+        }
+        hsize_t plen = pointers.size();
+        H5::DataSpace pspace(1, &plen);
+        auto ptype = ritsuko::hdf5::vls::define_pointer_datatype<int, int>();
+        auto phandle = ghandle.createDataSet("data", ptype, pspace);
+        phandle.write(pointers.data(), ptype);
+    }
+    expect_hdf5_error(path, "blub", "64-bit unsigned integer");
+
+    // Checking that this only works in the latest version.
+    {
+        H5::H5File handle(path, H5F_ACC_RDWR);
+        auto vhandle = handle.openGroup("blub");
+        vhandle.removeAttr("uzuki_version");
+    }
+    expect_hdf5_error(path, "blub", "unknown vector type");
+}
+
+TEST(Hdf5VlsTest, Scalar) {
+    auto path = "TEST-vls.h5";
+    std::string heap = "abcdefghijklmno";
+
+    {
+        H5::H5File handle(path, H5F_ACC_TRUNC);
+        auto ghandle = vector_opener(handle, "blub", "vls");
+        add_version(ghandle, "1.4");
+
+        auto hhandle = create_dataset(ghandle, "heap", heap.size(), H5::PredType::NATIVE_UINT8);
+        const unsigned char* hptr = reinterpret_cast<const unsigned char*>(heap.c_str());
+        hhandle.write(hptr, H5::PredType::NATIVE_UCHAR);
+
+        ritsuko::hdf5::vls::Pointer<uint8_t, uint8_t> ptr;
+        ptr.offset = 0; ptr.length = 10;
+        auto ptype = ritsuko::hdf5::vls::define_pointer_datatype<uint8_t, uint8_t>();
+        auto phandle = ghandle.createDataSet("data", ptype, H5S_SCALAR);
+        phandle.write(&ptr, ptype);
+    }
+    {
+        auto parsed = load_hdf5(path, "blub");
+        EXPECT_EQ(parsed->type(), uzuki2::STRING);
+        auto sptr = static_cast<const DefaultStringVector*>(parsed.get());
+        EXPECT_EQ(sptr->size(), 1);
+        EXPECT_EQ(sptr->base.values.front(), "abcdefghij");
+    }
+
+    // Checking that it works correctly with early termination.
+    {
+        H5::H5File handle(path, H5F_ACC_RDWR);
+        auto ghandle = handle.openGroup("blub");
+        auto hhandle = ghandle.openDataSet("heap");
+        std::vector<uint8_t> replacement(heap.size());
+        hhandle.write(replacement.data(), H5::PredType::NATIVE_UINT8);
+    }
+    {
+        auto parsed = load_hdf5(path, "blub");
+        EXPECT_EQ(parsed->type(), uzuki2::STRING);
+        auto sptr = static_cast<const DefaultStringVector*>(parsed.get());
+        EXPECT_EQ(sptr->size(), 1);
+        EXPECT_EQ(sptr->base.values.front(), "");
+    }
+
+    // Checking that scalar works correctly with missing values.
+    {
+        H5::H5File handle(path, H5F_ACC_RDWR);
+        auto ghandle = handle.openGroup("blub");
+        auto dhandle = ghandle.openDataSet("data");
+        H5::StrType stype(0, 10);
+        auto ahandle = dhandle.createAttribute("missing-value-placeholder", stype, H5S_SCALAR);
+        ahandle.write(stype, std::string{});
+    }
+    {
+        auto parsed = load_hdf5(path, "blub");
+        EXPECT_EQ(parsed->type(), uzuki2::STRING);
+        auto sptr = static_cast<const DefaultStringVector*>(parsed.get());
+        EXPECT_EQ(sptr->size(), 1);
+        EXPECT_EQ(sptr->base.values.front(), "ich bin missing");
+    }
+}