From 952525b1f67be3cec5309120f525e8aac9c7540e Mon Sep 17 00:00:00 2001
From: "Peter J. Jones" <peter@jonesbunch.com>
Date: Tue, 18 Aug 2026 14:08:17 +0200
Subject: [PATCH 1/9] std::string_view -> std::string

Arrow/parquet uses std::string_view to represent byte arrays.
However, since they are non-owning objects they need the original
Arrow array to remain resident in memory.  That's unhelpful if we want
to delete arrow arrays after decoding them.

This change copies std::string_view values into std::string objects so
there is no dependency on the original Arrow array memory.
---
 include/mzpeak/data/null_marking.h        |  6 ++--
 include/mzpeak/data/transformer/primary.h |  2 +-
 include/mzpeak/util/algorithm.h           |  3 +-
 include/mzpeak/util/decoders.h            | 26 +++++++++++++++-
 include/mzpeak/util/types.h               | 36 +++++++++++++++++++----
 src/util/executor.cpp                     |  3 +-
 src/util/planner.cpp                      | 10 ++++---
 7 files changed, 71 insertions(+), 15 deletions(-)

diff --git a/include/mzpeak/data/null_marking.h b/include/mzpeak/data/null_marking.h
index c21acca..4cbee07 100644
--- a/include/mzpeak/data/null_marking.h
+++ b/include/mzpeak/data/null_marking.h
@@ -171,7 +171,8 @@ std::optional<T> Decoder<T, U>::operator()(int64_t index)
     prior_.index = index;
 
     if (range.size() == 1) {
-      prior_.value = array_->Value(range.begin);
+      prior_.value =
+          Decoders::unsafe_array_value<enum_type_v<T>>(array_, range.begin);
       prior_.delta = estimator_.predict(prior_.value);
     } else {
       auto slice = array_->Slice(range.begin, range.size());
@@ -182,7 +183,8 @@ std::optional<T> Decoder<T, U>::operator()(int64_t index)
       Decoders::Scalar<T> decoder;
       decoder.decode(slice, values);
 
-      prior_.value = array_->Value(range.anchor(index));
+      prior_.value =
+          Decoders::unsafe_array_value<enum_type_v<T>>(array_, range.anchor(index));
       prior_.delta = Algorithm::median_delta(values, zero_);
     }
 
diff --git a/include/mzpeak/data/transformer/primary.h b/include/mzpeak/data/transformer/primary.h
index 3061d40..dc95f0a 100644
--- a/include/mzpeak/data/transformer/primary.h
+++ b/include/mzpeak/data/transformer/primary.h
@@ -75,7 +75,7 @@ Decoder<T>::Decoder(std::shared_ptr<Signals> signals,
   };
 
   // Decode the `chunk_encoding` column.
-  std::vector<std::string_view> encodings;
+  std::vector<std::string> encodings;
   decode(Schema::BufferFormat::ChunkEncoding, encodings);
   chunk_encoding_.reserve(encodings.size());
 
diff --git a/include/mzpeak/util/algorithm.h b/include/mzpeak/util/algorithm.h
index e0d1f35..4f6950b 100644
--- a/include/mzpeak/util/algorithm.h
+++ b/include/mzpeak/util/algorithm.h
@@ -16,6 +16,7 @@ top-level directory of this repository.
 #include <ranges>
 #include <vector>
 
+#include "mzpeak/util/decoders.h"
 #include "mzpeak/util/types.h"
 
 namespace MzPeak::Util::Algorithm {
@@ -152,7 +153,7 @@ null_delta_decode(typename type_traits<T>::value_type start,
 
   for (int64_t index : std::views::iota(0, length)) {
     if (casted->IsValid(index)) {
-      ValueType delta = casted->Value(index);
+      ValueType delta = Decoders::unsafe_array_value<T>(casted, index);
       last = last.value_or(zero) + delta;
       append(last);
     } else {
diff --git a/include/mzpeak/util/decoders.h b/include/mzpeak/util/decoders.h
index 34f4153..90761b3 100644
--- a/include/mzpeak/util/decoders.h
+++ b/include/mzpeak/util/decoders.h
@@ -43,6 +43,29 @@ concept from_arrow_array =
       { t.decode(a, r) } -> std::same_as<void>;
     };
 
+/******************************************************************************/
+/**
+ * Read a value from an array without checking bounds or if it is NULL.
+ *
+ * This is needed because Parquet/Arrow uses std::string_view for byte
+ * arrays, but that means that the original arrow array needs to
+ * remain resident in memory.  Therefore we need to copy the memory
+ * referenced by a std::string_view into a std::string.
+ */
+template <Type T>
+type_traits<T>::value_type
+unsafe_array_value(const std::shared_ptr<typename type_traits<T>::array_type>& ary,
+                   int64_t index)
+{
+  using A = type_traits<T>::array_type;
+
+  if constexpr (std::is_same_v<A, arrow::StringArray>) {
+    return ary->GetString(index);
+  } else {
+    return ary->Value(index);
+  }
+}
+
 /******************************************************************************/
 /**
  * If the given array is a "list of lists" then visit each element of
@@ -177,7 +200,8 @@ class Scalar final : Helper<Scalar<V, C, N>> {
         std::optional<V> value = std::invoke(null_decoder_, i);
         if (value.has_value()) this->push(dst, std::move(*value));
       } else {
-        this->push(dst, std::move(casted->Value(i)));
+        V value = unsafe_array_value<enum_type_v<V>>(casted, i);
+        this->push(dst, std::move(value));
       }
     }
   }
diff --git a/include/mzpeak/util/types.h b/include/mzpeak/util/types.h
index a3703be..08eb52b 100644
--- a/include/mzpeak/util/types.h
+++ b/include/mzpeak/util/types.h
@@ -14,9 +14,11 @@ directory of this repository.
 #include <cstdint>
 #include <optional>
 #include <parquet/types.h>
+#include <type_traits>
 #include <variant>
 
 #include "mzpeak/exception.h"
+#include "mzpeak/util/compat.h" // IWYU pragma: keep
 
 namespace parquet::schema {
 class PrimitiveNode;
@@ -36,7 +38,7 @@ concept supported_type = std::same_as<std::remove_cvref_t<T>, int8_t> ||
                          std::same_as<std::remove_cvref_t<T>, uint64_t> ||
                          std::same_as<std::remove_cvref_t<T>, float> ||
                          std::same_as<std::remove_cvref_t<T>, double> ||
-                         std::same_as<std::remove_cvref_t<T>, std::string_view>;
+                         std::same_as<std::remove_cvref_t<T>, std::string>;
 
 /// A variant that can hold any supported type.
 using any_value_type = std::variant<int8_t,
@@ -47,7 +49,7 @@ using any_value_type = std::variant<int8_t,
                                     uint64_t,
                                     float,
                                     double,
-                                    std::string_view>;
+                                    std::string>;
 
 /// A variant that can hold pairs of any supported type.
 using any_pair_type = std::variant<std::pair<int8_t, int8_t>,
@@ -58,7 +60,7 @@ using any_pair_type = std::variant<std::pair<int8_t, int8_t>,
                                    std::pair<uint64_t, uint64_t>,
                                    std::pair<float, float>,
                                    std::pair<double, double>,
-                                   std::pair<std::string_view, std::string_view>>;
+                                   std::pair<std::string, std::string>>;
 
 /// Enum of supported data types for tracking at run time.
 enum class Type {
@@ -165,7 +167,7 @@ template <> struct type_traits<Type::Float64> {
 
 template <> struct type_traits<Type::ByteArray> {
   static constexpr const char* name = "bytes";
-  using value_type = std::string_view;
+  using value_type = std::string;
   using parquet_type = parquet::ByteArrayType;
   using array_type = arrow::StringArray;
   using builder_type = arrow::StringBuilder;
@@ -209,7 +211,7 @@ template <> struct type_from_value_type<double> {
   static constexpr Type enum_type = Type::Float64;
 };
 
-template <> struct type_from_value_type<std::string_view> {
+template <> struct type_from_value_type<std::string> {
   static constexpr Type enum_type = Type::ByteArray;
 };
 
@@ -217,6 +219,30 @@ template <> struct type_from_value_type<std::string_view> {
 template <supported_type T>
 inline constexpr Type enum_type_v = type_from_value_type<T>::enum_type;
 
+/**
+ * Cast or convert a value from one type (usually Type::parquet_type)
+ * to another (usually Type::value_type).
+ *
+ * This is needed because Parquet/Arrow uses std::string_view for byte
+ * arrays, but that means that the original arrow array needs to
+ * remain resident in memory.  Therefore we need to copy the memory
+ * referenced by a std::string_view into a std::string.
+ */
+template <typename To, typename From> To safe_cast_or_copy(From value)
+{
+  using from_t = std::remove_cvref_t<From>;
+  using to_t = std::remove_cvref_t<To>;
+
+  if constexpr (std::is_convertible_v<From, To>) {
+    return value;
+  } else if constexpr (std::is_same_v<to_t, std::string> &&
+                       std::is_same_v<from_t, parquet::ByteArray>) {
+    return parquet::ByteArrayToString(value);
+  } else {
+    static_assert(false_type<From, To>, "no conversion available");
+  }
+}
+
 /**
  * Return a type for the given parquet node.
  */
diff --git a/src/util/executor.cpp b/src/util/executor.cpp
index 7c3298b..9c1eba7 100644
--- a/src/util/executor.cpp
+++ b/src/util/executor.cpp
@@ -77,7 +77,8 @@ Query::Result<Query::value_t> Executor::Impl::ArrayValueHelper::operator()()
   }
 
   auto casted = std::static_pointer_cast<typename type_traits<T>::array_type>(a);
-  return Query::Result<Query::value_t>(casted->Value(row_index_));
+  auto value = Decoders::unsafe_array_value<T>(casted, row_index_);
+  return Query::Result<Query::value_t>(value);
 }
 
 /******************************************************************************/
diff --git a/src/util/planner.cpp b/src/util/planner.cpp
index dc07f47..fc8cfbe 100644
--- a/src/util/planner.cpp
+++ b/src/util/planner.cpp
@@ -275,14 +275,16 @@ struct ColMinMax final {
     if (index_ != nullptr) {
       using Index = parquet::TypedColumnIndex<P>;
       std::shared_ptr<Index> index = std::static_pointer_cast<Index>(index_);
-      return R(std::make_pair(static_cast<V>(index->min_values()[page_index_]),
-                              static_cast<V>(index->max_values()[page_index_])));
+      V min = safe_cast_or_copy<V>(index->min_values()[page_index_]);
+      V max = safe_cast_or_copy<V>(index->max_values()[page_index_]);
+      return R(std::make_pair(min, max));
     } else if (stats_ != nullptr) {
       using Stats = parquet::TypedStatistics<P>;
       if (!stats_->HasMinMax()) return R::skip();
       std::shared_ptr<Stats> stats = std::static_pointer_cast<Stats>(stats_);
-      return R(std::make_pair(static_cast<V>(stats->min()),
-                              static_cast<V>(stats->max())));
+      V min = safe_cast_or_copy<V>(stats->min());
+      V max = safe_cast_or_copy<V>(stats->max());
+      return R(std::make_pair(min, max));
     } else {
       return R::skip();
     }

From 680277f416c463b60df3c75f187fc0312e9aeb93 Mon Sep 17 00:00:00 2001
From: "Peter J. Jones" <peter@jonesbunch.com>
Date: Tue, 18 Aug 2026 14:34:12 +0200
Subject: [PATCH 2/9] Delete arrow arrays after decoding them

No need to keep this memory around longer than necessary.
---
 include/mzpeak/data/encoding.h |  2 ++
 include/mzpeak/util/slice.h    | 24 +++++++++++++++---------
 src/util/slice.cpp             |  6 ++++--
 test/executor_test.cpp         |  9 +++------
 4 files changed, 24 insertions(+), 17 deletions(-)

diff --git a/include/mzpeak/data/encoding.h b/include/mzpeak/data/encoding.h
index 88024b1..b8ee50e 100644
--- a/include/mzpeak/data/encoding.h
+++ b/include/mzpeak/data/encoding.h
@@ -161,6 +161,8 @@ void Decoder<T>::decode_with_nulls(const ArrayIndex::Dimension& dim,
 
   if (!col.has_value()) {
     throw ParquetError("unable to decode dimension, not in schema: " + dim.name);
+  } else if (!slice_->has_column(col.value())) {
+    return; // No data to decode so we can exit early.
   }
 
   auto go = [&](auto&& decoder) -> void { slice_->array(col.value(), v, decoder); };
diff --git a/include/mzpeak/util/slice.h b/include/mzpeak/util/slice.h
index d29c1e9..cb0372d 100644
--- a/include/mzpeak/util/slice.h
+++ b/include/mzpeak/util/slice.h
@@ -48,13 +48,18 @@ class Slice final {
   /**
    * Return the raw array for the given field.
    *
+   * The returned raw array is removed from the internal storage
+   * therefore calling this method again with the same column will
+   * fail.
+   *
    * NOTE: If you request a field that does not exist in the slice
    * this function will return a nullptr.
    */
-  std::shared_ptr<Raw> raw(const Column&) const;
+  std::shared_ptr<Raw> raw(const Column&);
 
   /**
-   * Decode the first non-null value.
+   * Decode the first non-null value from the given column.  The
+   * column is then removed from internal storage.
    *
    * Template Parameters:
    *
@@ -63,10 +68,11 @@ class Slice final {
    *   - R: The destination object to update with the decoded value
    */
   template <typename T, typename R = std::optional<typename T::value_type>>
-  void singleton(const Column&, R&, T&& = {}) const;
+  void singleton(const Column&, R&, T&& = {});
 
   /**
-   * Exact and decode an array.
+   * Extract and decode an array.  The array is then removed from the
+   * internal storage.
    *
    * Use one of the decoders defined in `decoders.h`, or write your own.
    *
@@ -78,12 +84,12 @@ class Slice final {
    */
   template <typename T, typename V = std::vector<typename T::value_type>>
     requires Decoders::from_arrow_array<T, V>
-  void array(const Column&, V&, T&& = {}) const;
+  void array(const Column&, V&, T&& = {});
 
   /****************************************************************************/
   template <typename T, typename V = std::vector<typename T::value_type>>
     requires Decoders::from_arrow_array<T, V>
-  void array(const Column&, V&, T&) const;
+  void array(const Column&, V&, T&);
 
 private:
   friend class MzPeak::Util::Executor;
@@ -100,7 +106,7 @@ class Slice final {
 
 /******************************************************************************/
 template <typename T, typename R>
-void Slice::singleton(const Column& field, R& dst, T&& t) const
+void Slice::singleton(const Column& field, R& dst, T&& t)
 {
   std::shared_ptr<Raw> chunks = raw(field);
 
@@ -130,7 +136,7 @@ void Slice::singleton(const Column& field, R& dst, T&& t) const
 /******************************************************************************/
 template <typename T, typename V>
   requires Decoders::from_arrow_array<T, V>
-void Slice::array(const Column& field, V& v, T&& t) const
+void Slice::array(const Column& field, V& v, T&& t)
 {
   array(field, v, t);
 }
@@ -138,7 +144,7 @@ void Slice::array(const Column& field, V& v, T&& t) const
 /******************************************************************************/
 template <typename T, typename V>
   requires Decoders::from_arrow_array<T, V>
-void Slice::array(const Column& field, V& v, T& t) const
+void Slice::array(const Column& field, V& v, T& t)
 {
   std::shared_ptr<Raw> chunks = raw(field);
   if (chunks == nullptr) return;
diff --git a/src/util/slice.cpp b/src/util/slice.cpp
index a79ba45..21ad787 100644
--- a/src/util/slice.cpp
+++ b/src/util/slice.cpp
@@ -52,14 +52,16 @@ bool Slice::has_column(const Column& column) const
 }
 
 /******************************************************************************/
-std::shared_ptr<Slice::Raw> Slice::raw(const Column& field) const
+std::shared_ptr<Slice::Raw> Slice::raw(const Column& field)
 {
   auto it = impl_->arrays_.find(impl_->key(field));
 
   if (it == impl_->arrays_.end()) {
     return nullptr;
   } else {
-    return it->second;
+    auto sp = it->second;
+    impl_->arrays_.erase(it);
+    return sp;
   }
 }
 
diff --git a/test/executor_test.cpp b/test/executor_test.cpp
index b32d80b..2921edb 100644
--- a/test/executor_test.cpp
+++ b/test/executor_test.cpp
@@ -47,9 +47,7 @@ BOOST_AUTO_TEST_CASE(can_find_spectrum)
   Util::Executor executor = parquet->executor(projection);
   const auto& slice = executor.execute(plan);
   BOOST_TEST((slice->fields() == projection.get()));
-
-  auto raw = slice->raw(*mz_field);
-  BOOST_TEST((raw != nullptr));
+  BOOST_TEST(slice->has_column(mz_field.value()));
 
   // Decode, dropping null values.
   std::vector<double> mz;
@@ -88,10 +86,9 @@ BOOST_AUTO_TEST_CASE(can_read_uint8_t)
 
   Util::Executor executor = parquet->executor(projection);
   const auto& slice = executor.execute(plan);
-  BOOST_TEST((slice->fields() == projection.get()));
 
-  auto raw = slice->raw(*ms_level);
-  BOOST_TEST((raw != nullptr));
+  BOOST_TEST((slice->fields() == projection.get()));
+  BOOST_TEST(slice->has_column(ms_level.value()));
 
   std::vector<uint8_t> levels;
   slice->array<Util::Decoders::Scalar<uint8_t>>(*ms_level, levels);

From 1529adf6c6982bfbdb6bbcd2108237fa25fb7368 Mon Sep 17 00:00:00 2001
From: "Peter J. Jones" <peter@jonesbunch.com>
Date: Tue, 18 Aug 2026 14:35:07 +0200
Subject: [PATCH 3/9] Remove more uses of std::optional::operator*

---
 include/mzpeak/data/transformer/primary.h |  4 ++--
 test/executor_test.cpp                    | 12 ++++++------
 2 files changed, 8 insertions(+), 8 deletions(-)

diff --git a/include/mzpeak/data/transformer/primary.h b/include/mzpeak/data/transformer/primary.h
index dc95f0a..a718280 100644
--- a/include/mzpeak/data/transformer/primary.h
+++ b/include/mzpeak/data/transformer/primary.h
@@ -71,7 +71,7 @@ Decoder<T>::Decoder(std::shared_ptr<Signals> signals,
       throw InvalidFormatError(msg);
     }
 
-    slice->array(*column, dest, Util::Decoders::Scalar<U>());
+    slice->array(column.value(), dest, Util::Decoders::Scalar<U>());
   };
 
   // Decode the `chunk_encoding` column.
@@ -86,7 +86,7 @@ Decoder<T>::Decoder(std::shared_ptr<Signals> signals,
       throw InvalidFormatError("invalid chunk encoding CV: " + std::string(s));
     }
 
-    chunk_encoding_.emplace_back(*cv);
+    chunk_encoding_.emplace_back(cv.value());
   }
 
   // Decode the `chunk_start` column.
diff --git a/test/executor_test.cpp b/test/executor_test.cpp
index 2921edb..cbf199f 100644
--- a/test/executor_test.cpp
+++ b/test/executor_test.cpp
@@ -35,14 +35,14 @@ BOOST_AUTO_TEST_CASE(can_find_spectrum)
   auto mz_field = parquet->field("point", "mz");
   BOOST_TEST(mz_field.has_value());
 
-  auto query = Util::Query::Builder(*index_field).eq<uint64_t>(1ul);
+  auto query = Util::Query::Builder(index_field.value()).eq<uint64_t>(1ul);
 
   Util::Planner planner = parquet->planner(query);
   auto plan = planner.plan();
   BOOST_TEST(plan.ranges.size() == 1ul);
 
   Util::Projection projection;
-  projection.project(*mz_field);
+  projection.project(mz_field.value());
 
   Util::Executor executor = parquet->executor(projection);
   const auto& slice = executor.execute(plan);
@@ -51,7 +51,7 @@ BOOST_AUTO_TEST_CASE(can_find_spectrum)
 
   // Decode, dropping null values.
   std::vector<double> mz;
-  slice->array<Util::Decoders::Scalar<double>>(*mz_field, mz);
+  slice->array<Util::Decoders::Scalar<double>>(mz_field.value(), mz);
   BOOST_TEST(mz.size() == 15063);
   BOOST_TEST(mz[0] == 200.09, boost::test_tools::tolerance(0.001));
   BOOST_TEST(mz[mz.size() - 1] == 1999.81, boost::test_tools::tolerance(0.001));
@@ -76,13 +76,13 @@ BOOST_AUTO_TEST_CASE(can_read_uint8_t)
   auto ms_level = parquet->field("root", "ms_level");
   BOOST_TEST(ms_level.has_value());
 
-  auto query = Util::Query::Builder(*index_field).eq<uint64_t>(0ul);
+  auto query = Util::Query::Builder(index_field.value()).eq<uint64_t>(0ul);
   Util::Planner planner = parquet->planner(query);
   auto plan = planner.plan();
   BOOST_TEST(plan.ranges.size() == 1ul);
 
   Util::Projection projection;
-  projection.project(*ms_level);
+  projection.project(ms_level.value());
 
   Util::Executor executor = parquet->executor(projection);
   const auto& slice = executor.execute(plan);
@@ -91,7 +91,7 @@ BOOST_AUTO_TEST_CASE(can_read_uint8_t)
   BOOST_TEST(slice->has_column(ms_level.value()));
 
   std::vector<uint8_t> levels;
-  slice->array<Util::Decoders::Scalar<uint8_t>>(*ms_level, levels);
+  slice->array<Util::Decoders::Scalar<uint8_t>>(ms_level.value(), levels);
   BOOST_TEST(levels.size() == 1);
   BOOST_TEST(levels[0] == 1);
 }

From 6d1f9727e0da3fdcb66a9d84e6dae4d3cebd2a9c Mon Sep 17 00:00:00 2001
From: "Peter J. Jones" <peter@jonesbunch.com>
Date: Tue, 18 Aug 2026 14:48:04 +0200
Subject: [PATCH 4/9] Save a few stack bytes by passing std::string_view by
 value

---
 include/mzpeak/data/signals.h          |  2 +-
 include/mzpeak/index.h                 |  2 +-
 include/mzpeak/metadata/table.h        |  2 +-
 include/mzpeak/schema/buffer_format.h  |  2 +-
 include/mzpeak/schema/cv.h             |  8 ++++----
 include/mzpeak/schema/data_kind.h      |  2 +-
 include/mzpeak/schema/entity_type.h    |  2 +-
 include/mzpeak/schema/group.h          |  4 ++--
 include/mzpeak/schema/psi/array_type.h |  2 +-
 include/mzpeak/util/manager.h          |  3 +--
 include/mzpeak/util/parquet.h          |  7 +++----
 include/mzpeak/util/projection.h       |  2 +-
 src/data/signals.cpp                   |  2 +-
 src/index.cpp                          |  3 +--
 src/metadata/table.cpp                 |  2 +-
 src/schema/buffer_format.cpp           |  2 +-
 src/schema/cv.cpp                      |  2 +-
 src/schema/data_kind.cpp               |  2 +-
 src/schema/entity_type.cpp             |  2 +-
 src/schema/group.cpp                   |  4 ++--
 src/schema/psi/array_type.cpp          |  2 +-
 src/util/executor.cpp                  |  2 +-
 src/util/manager.cpp                   |  2 +-
 src/util/parquet.cpp                   | 13 ++++++-------
 src/util/projection.cpp                |  2 +-
 25 files changed, 37 insertions(+), 41 deletions(-)

diff --git a/include/mzpeak/data/signals.h b/include/mzpeak/data/signals.h
index 7cef70c..a2e69eb 100644
--- a/include/mzpeak/data/signals.h
+++ b/include/mzpeak/data/signals.h
@@ -56,7 +56,7 @@ class Signals {
    *
    * Useful if you need to manually construct queries.
    */
-  std::optional<Schema::Column> column(const std::string_view&) const;
+  std::optional<Schema::Column> column(std::string_view) const;
 
   /**
    * Low-level interface for accessing a column given an array index entry.
diff --git a/include/mzpeak/index.h b/include/mzpeak/index.h
index 93dfdb8..1fe7acf 100644
--- a/include/mzpeak/index.h
+++ b/include/mzpeak/index.h
@@ -43,7 +43,7 @@ class Index {
   /**
    * Find a file in the mzPeak archive with the given name.
    */
-  std::vector<Schema::File>::const_iterator find(const std::string_view&) const;
+  std::vector<Schema::File>::const_iterator find(std::string_view) const;
 
   /**
    * Access the spectra in the file.
diff --git a/include/mzpeak/metadata/table.h b/include/mzpeak/metadata/table.h
index 8b1cb22..a3ecc21 100644
--- a/include/mzpeak/metadata/table.h
+++ b/include/mzpeak/metadata/table.h
@@ -31,7 +31,7 @@ class Table final {
    * Return a group with the given name.  If the group does not
    * exist in the schema return `nullptr`.
    */
-  std::shared_ptr<Schema::Group> group(const std::string_view&) const;
+  std::shared_ptr<Schema::Group> group(std::string_view) const;
 
   /**
    * Read all rows from the given group where the index column
diff --git a/include/mzpeak/schema/buffer_format.h b/include/mzpeak/schema/buffer_format.h
index d0bcfc4..de4c8a9 100644
--- a/include/mzpeak/schema/buffer_format.h
+++ b/include/mzpeak/schema/buffer_format.h
@@ -61,6 +61,6 @@ std::string buffer_format_to_string(BufferFormat);
 /**
  * Parse an BufferFormat from a string.
  */
-BufferFormat buffer_format_from_string(const std::string_view&);
+BufferFormat buffer_format_from_string(std::string_view);
 
 } // namespace MzPeak::Schema
diff --git a/include/mzpeak/schema/cv.h b/include/mzpeak/schema/cv.h
index 6b4f9ec..0b75ce2 100644
--- a/include/mzpeak/schema/cv.h
+++ b/include/mzpeak/schema/cv.h
@@ -19,9 +19,9 @@ namespace MzPeak::Schema {
 class CV {
 public:
   /// Constructor.
-  CV(const std::string_view& code, const std::string_view& accession)
-      : code_(std::move(code))
-      , accession_(std::move(accession))
+  CV(std::string_view code, std::string_view accession)
+      : code_(code)
+      , accession_(accession)
   {
   }
 
@@ -29,7 +29,7 @@ class CV {
   ~CV() = default;
 
   /// Parse a string like "MS:1000511"
-  static std::optional<CV> from_string(const std::string_view&);
+  static std::optional<CV> from_string(std::string_view);
 
   /// Convert this CV term to a string like "MS:1000511"
   std::string to_string() const;
diff --git a/include/mzpeak/schema/data_kind.h b/include/mzpeak/schema/data_kind.h
index 7bfb16b..5adb5d2 100644
--- a/include/mzpeak/schema/data_kind.h
+++ b/include/mzpeak/schema/data_kind.h
@@ -46,6 +46,6 @@ std::string data_kind_to_string(DataKind);
 /**
  * Parse a DataKind from a string view.
  */
-DataKind data_kind_from_string(const std::string_view&);
+DataKind data_kind_from_string(std::string_view);
 
 } // namespace MzPeak::Schema
diff --git a/include/mzpeak/schema/entity_type.h b/include/mzpeak/schema/entity_type.h
index d041830..e0a72bd 100644
--- a/include/mzpeak/schema/entity_type.h
+++ b/include/mzpeak/schema/entity_type.h
@@ -38,6 +38,6 @@ std::string entity_type_to_string(EntityType);
 /**
  * Parse an EntityType from a string view.
  */
-EntityType entity_type_from_string(const std::string_view&);
+EntityType entity_type_from_string(std::string_view);
 
 } // namespace MzPeak::Schema
diff --git a/include/mzpeak/schema/group.h b/include/mzpeak/schema/group.h
index 4e041d7..db44a07 100644
--- a/include/mzpeak/schema/group.h
+++ b/include/mzpeak/schema/group.h
@@ -73,7 +73,7 @@ class Group final {
     };
 
     /// Constructor from an encoded column name.
-    explicit Field(const std::string_view& column_name,
+    explicit Field(std::string_view column_name,
                    index_type rel_index,
                    index_type abs_index);
 
@@ -171,7 +171,7 @@ class Group final {
    * NOTE: For metadata groups this is the cleaned name, not the raw
    * schema node name.
    */
-  std::optional<std::shared_ptr<const Field>> field(const std::string_view&&) const;
+  std::optional<std::shared_ptr<const Field>> field(std::string_view) const;
 
   /**
    * Find a field given its CV type.
diff --git a/include/mzpeak/schema/psi/array_type.h b/include/mzpeak/schema/psi/array_type.h
index 88a2a27..898932c 100644
--- a/include/mzpeak/schema/psi/array_type.h
+++ b/include/mzpeak/schema/psi/array_type.h
@@ -150,6 +150,6 @@ std::string array_type_to_string(ArrayType);
 /**
  * Parse an ArrayType from a string.
  */
-ArrayType array_type_from_string(const std::string_view&);
+ArrayType array_type_from_string(std::string_view);
 
 } // namespace MzPeak::Schema::PSI
diff --git a/include/mzpeak/util/manager.h b/include/mzpeak/util/manager.h
index ca88079..5d4f467 100644
--- a/include/mzpeak/util/manager.h
+++ b/include/mzpeak/util/manager.h
@@ -32,8 +32,7 @@ class Manager final {
   /**
    * Find a file given its name.
    */
-  std::vector<Schema::File>::const_iterator
-  find_file(const std::string_view& name) const;
+  std::vector<Schema::File>::const_iterator find_file(std::string_view name) const;
 
   /**
    * Open a Parquet file from the mzPeak archive.
diff --git a/include/mzpeak/util/parquet.h b/include/mzpeak/util/parquet.h
index 3c7cd5f..9ae1b49 100644
--- a/include/mzpeak/util/parquet.h
+++ b/include/mzpeak/util/parquet.h
@@ -45,8 +45,7 @@ class Parquet final {
   /**
    * Return a Group and Field matching the given names.
    */
-  std::optional<Schema::Column> field(const std::string_view&,
-                                      const std::string_view&) const;
+  std::optional<Schema::Column> field(std::string_view, std::string_view) const;
 
   /**
    * Access the file metadata.
@@ -57,13 +56,13 @@ class Parquet final {
    * Fetch a string value from the metadata key-value store.
    */
   std::optional<std::string> kv_string(const file_metadata_t&,
-                                       const std::string_view&) const;
+                                       std::string_view) const;
 
   /**
    * Fetch a std::size_t value from the metadata key-value store.
    */
   std::optional<std::size_t> kv_size_t(const file_metadata_t&,
-                                       const std::string_view&) const;
+                                       std::string_view) const;
 
   /**
    * Directly access the FileReader.  This reference is only valid
diff --git a/include/mzpeak/util/projection.h b/include/mzpeak/util/projection.h
index 5b4a53f..6bf909c 100644
--- a/include/mzpeak/util/projection.h
+++ b/include/mzpeak/util/projection.h
@@ -43,7 +43,7 @@ class Projection final {
   /**
    * Project a column using a field name.
    */
-  Result project(const std::shared_ptr<Schema::Group>&, const std::string_view&&);
+  Result project(const std::shared_ptr<Schema::Group>&, std::string_view);
 
   /**
    * Look up a CV type and project that.
diff --git a/src/data/signals.cpp b/src/data/signals.cpp
index 9e5b660..1b743bb 100644
--- a/src/data/signals.cpp
+++ b/src/data/signals.cpp
@@ -102,7 +102,7 @@ std::size_t Signals::record_count() const
 }
 
 /******************************************************************************/
-std::optional<Schema::Column> Signals::column(const std::string_view& name) const
+std::optional<Schema::Column> Signals::column(std::string_view name) const
 {
   return impl_->parquet_->field(impl_->array_index_->prefix(), name);
 }
diff --git a/src/index.cpp b/src/index.cpp
index 9c41888..41330b0 100644
--- a/src/index.cpp
+++ b/src/index.cpp
@@ -27,8 +27,7 @@ Index::Index(std::unique_ptr<MzPeak::IO::Archive> archive)
 const std::vector<Schema::File>& Index::files() const { return manager_->files(); }
 
 /******************************************************************************/
-std::vector<Schema::File>::const_iterator
-Index::find(const std::string_view& name) const
+std::vector<Schema::File>::const_iterator Index::find(std::string_view name) const
 {
   return manager_->find_file(name);
 }
diff --git a/src/metadata/table.cpp b/src/metadata/table.cpp
index 46197c6..0fbbb00 100644
--- a/src/metadata/table.cpp
+++ b/src/metadata/table.cpp
@@ -50,7 +50,7 @@ Table::Table(std::unique_ptr<Util::Parquet> parquet)
 Table::~Table() = default;
 
 /******************************************************************************/
-std::shared_ptr<Schema::Group> Table::group(const std::string_view& name) const
+std::shared_ptr<Schema::Group> Table::group(std::string_view name) const
 {
   const std::shared_ptr<Schema::GroupMap>& map = impl_->parquet_->groups();
   auto it = map->find(std::string(name));
diff --git a/src/schema/buffer_format.cpp b/src/schema/buffer_format.cpp
index ddf56db..20f34b6 100644
--- a/src/schema/buffer_format.cpp
+++ b/src/schema/buffer_format.cpp
@@ -38,7 +38,7 @@ std::string buffer_format_to_string(BufferFormat v)
 }
 
 /******************************************************************************/
-BufferFormat buffer_format_from_string(const std::string_view& s)
+BufferFormat buffer_format_from_string(std::string_view s)
 {
   using enum BufferFormat;
 
diff --git a/src/schema/cv.cpp b/src/schema/cv.cpp
index 6ff4ddf..69c1bdf 100644
--- a/src/schema/cv.cpp
+++ b/src/schema/cv.cpp
@@ -11,7 +11,7 @@ top-level directory of this repository.
 namespace MzPeak::Schema {
 
 /******************************************************************************/
-std::optional<CV> CV::from_string(const std::string_view& s)
+std::optional<CV> CV::from_string(std::string_view s)
 {
   std::string_view::size_type sep_pos = s.find(':');
 
diff --git a/src/schema/data_kind.cpp b/src/schema/data_kind.cpp
index 58905e3..200ce4a 100644
--- a/src/schema/data_kind.cpp
+++ b/src/schema/data_kind.cpp
@@ -32,7 +32,7 @@ std::string data_kind_to_string(DataKind dk)
 }
 
 /******************************************************************************/
-DataKind data_kind_from_string(const std::string_view& s)
+DataKind data_kind_from_string(std::string_view s)
 {
   using enum DataKind;
 
diff --git a/src/schema/entity_type.cpp b/src/schema/entity_type.cpp
index 537b1fe..ec7b047 100644
--- a/src/schema/entity_type.cpp
+++ b/src/schema/entity_type.cpp
@@ -29,7 +29,7 @@ std::string entity_type_to_string(EntityType et)
   return "other";
 }
 
-EntityType entity_type_from_string(const std::string_view& s)
+EntityType entity_type_from_string(std::string_view s)
 {
   using enum EntityType;
 
diff --git a/src/schema/group.cpp b/src/schema/group.cpp
index 63e1642..fdedadb 100644
--- a/src/schema/group.cpp
+++ b/src/schema/group.cpp
@@ -80,7 +80,7 @@ field_type_from_parquet(const std::shared_ptr<parquet::schema::GroupNode>& node)
 }
 
 /******************************************************************************/
-Group::Field::Field(const std::string_view& column_name,
+Group::Field::Field(std::string_view column_name,
                     index_type rel_index,
                     index_type abs_index)
     : rel_index_(rel_index)
@@ -208,7 +208,7 @@ Group::index_type Group::index() const { return index_; }
 
 /******************************************************************************/
 std::optional<std::shared_ptr<const Group::Field>>
-Group::field(const std::string_view&& name) const
+Group::field(std::string_view name) const
 {
   auto it = fields_.find(std::string{name});
 
diff --git a/src/schema/psi/array_type.cpp b/src/schema/psi/array_type.cpp
index 629f387..412cf12 100644
--- a/src/schema/psi/array_type.cpp
+++ b/src/schema/psi/array_type.cpp
@@ -64,7 +64,7 @@ std::string array_type_to_string(ArrayType v)
 }
 
 /******************************************************************************/
-ArrayType array_type_from_string(const std::string_view& s)
+ArrayType array_type_from_string(std::string_view s)
 {
   using enum ArrayType;
 
diff --git a/src/util/executor.cpp b/src/util/executor.cpp
index 9c1eba7..4832ca4 100644
--- a/src/util/executor.cpp
+++ b/src/util/executor.cpp
@@ -30,7 +30,7 @@ struct Executor::Impl {
   }
 
   /// Helper to check a result and throw an error if necessary.
-  template <typename T> T check(const std::string_view& msg, arrow::Result<T> r)
+  template <typename T> T check(std::string_view msg, arrow::Result<T> r)
   {
     if (!r.ok()) {
       std::string error("while executing a query: " + std::string(msg) + ": ");
diff --git a/src/util/manager.cpp b/src/util/manager.cpp
index 77c00b7..39af105 100644
--- a/src/util/manager.cpp
+++ b/src/util/manager.cpp
@@ -66,7 +66,7 @@ const std::vector<Schema::File>& Manager::files() const { return files_; }
 
 /******************************************************************************/
 std::vector<Schema::File>::const_iterator
-Manager::find_file(const std::string_view& name) const
+Manager::find_file(std::string_view name) const
 {
   return std::ranges::find(files_, name, &Schema::File::file_name);
 }
diff --git a/src/util/parquet.cpp b/src/util/parquet.cpp
index fab820d..a7b0a7a 100644
--- a/src/util/parquet.cpp
+++ b/src/util/parquet.cpp
@@ -24,7 +24,7 @@ namespace MzPeak::Util {
 
 /******************************************************************************/
 std::optional<std::string> get_kv_string(const Parquet::file_metadata_t& fmd,
-                                         const std::string_view& key)
+                                         std::string_view key)
 {
   auto result(fmd->key_value_metadata()->Get(key));
 
@@ -37,7 +37,7 @@ std::optional<std::string> get_kv_string(const Parquet::file_metadata_t& fmd,
 
 /******************************************************************************/
 std::optional<std::size_t> get_kv_uint(const Parquet::file_metadata_t& fmd,
-                                       const std::string_view& key)
+                                       std::string_view key)
 {
   return get_kv_string(fmd, key).and_then(
       [](const std::string& s) -> std::optional<std::size_t> {
@@ -147,9 +147,8 @@ const std::shared_ptr<Schema::GroupMap>& Parquet::groups() const
 }
 
 /******************************************************************************/
-std::optional<Schema::Column>
-Parquet::field(const std::string_view& group_name,
-               const std::string_view& field_name) const
+std::optional<Schema::Column> Parquet::field(std::string_view group_name,
+                                             std::string_view field_name) const
 {
   auto group_ptr = impl_->groups_->find(std::string{group_name});
   if (group_ptr == impl_->groups_->end()) return {};
@@ -168,14 +167,14 @@ Parquet::file_metadata_t Parquet::file_metadata() const
 
 /******************************************************************************/
 std::optional<std::string> Parquet::kv_string(const file_metadata_t& fmd,
-                                              const std::string_view& key) const
+                                              std::string_view key) const
 {
   return get_kv_string(fmd, key);
 }
 
 /******************************************************************************/
 std::optional<std::size_t> Parquet::kv_size_t(const file_metadata_t& fmd,
-                                              const std::string_view& key) const
+                                              std::string_view key) const
 {
   return get_kv_uint(fmd, key);
 }
diff --git a/src/util/projection.cpp b/src/util/projection.cpp
index dee9964..c2588c0 100644
--- a/src/util/projection.cpp
+++ b/src/util/projection.cpp
@@ -41,7 +41,7 @@ Projection::Result Projection::project(
 
 /******************************************************************************/
 Projection::Result Projection::project(const std::shared_ptr<Schema::Group>& group,
-                                       const std::string_view&& name)
+                                       std::string_view name)
 {
   return project(group, group->field(std::move(name)));
 }

From 529dd8f45d3f7db28877cd8ebe21316e78951d74 Mon Sep 17 00:00:00 2001
From: "Peter J. Jones" <peter@jonesbunch.com>
Date: Tue, 18 Aug 2026 14:48:25 +0200
Subject: [PATCH 5/9] Correctly prevent copying

---
 src/io/zip.cpp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/io/zip.cpp b/src/io/zip.cpp
index 9ecf5ad..41ff62c 100644
--- a/src/io/zip.cpp
+++ b/src/io/zip.cpp
@@ -84,8 +84,8 @@ class ZipFile_ final : public MzPeak::IO::File {
     fs::path path_;
 
   private:
-    Impl& operator=(const Impl&) = default;
-    Impl(const Impl&) = default;
+    Impl& operator=(const Impl&) = delete;
+    Impl(const Impl&) = delete;
   };
 
   std::shared_ptr<Impl> impl_;

From b981b03eb65d802f417b4e01356c20383beadddd Mon Sep 17 00:00:00 2001
From: "Peter J. Jones" <peter@jonesbunch.com>
Date: Tue, 18 Aug 2026 14:48:37 +0200
Subject: [PATCH 6/9] Unneeded call to reserve

---
 include/mzpeak/data/encoding.h | 1 -
 1 file changed, 1 deletion(-)

diff --git a/include/mzpeak/data/encoding.h b/include/mzpeak/data/encoding.h
index b8ee50e..2c86801 100644
--- a/include/mzpeak/data/encoding.h
+++ b/include/mzpeak/data/encoding.h
@@ -122,7 +122,6 @@ void Decoder<T>::remap(const ArrayIndex::Dimension& dim, std::vector<V>& v) cons
   } else {
     std::vector<F> tmp;
     decode<F>(dim, tmp);
-    v.reserve(tmp.size());
     v.insert(v.end(), tmp.begin(), tmp.end());
   }
 }

From f67853e94d74f956486c0ecb892610697a1dadfc Mon Sep 17 00:00:00 2001
From: "Peter J. Jones" <peter@jonesbunch.com>
Date: Tue, 18 Aug 2026 14:54:25 +0200
Subject: [PATCH 7/9] Add missing header files

---
 include/mzpeak/index.h        | 1 +
 include/mzpeak/util/manager.h | 2 ++
 2 files changed, 3 insertions(+)

diff --git a/include/mzpeak/index.h b/include/mzpeak/index.h
index 1fe7acf..8730603 100644
--- a/include/mzpeak/index.h
+++ b/include/mzpeak/index.h
@@ -9,6 +9,7 @@ directory of this repository.
 #pragma once
 
 #include <memory>
+#include <string_view>
 #include <vector>
 
 namespace MzPeak {
diff --git a/include/mzpeak/util/manager.h b/include/mzpeak/util/manager.h
index 5d4f467..533e589 100644
--- a/include/mzpeak/util/manager.h
+++ b/include/mzpeak/util/manager.h
@@ -9,6 +9,8 @@ directory of this repository.
 #pragma once
 
 #include <memory>
+#include <string_view>
+#include <vector>
 
 #include "mzpeak/io/archive.h"
 #include "mzpeak/schema/file.h"

From 8579280130b1381789b81630090bb8ccd24a56fb Mon Sep 17 00:00:00 2001
From: "Peter J. Jones" <peter@jonesbunch.com>
Date: Tue, 18 Aug 2026 14:56:43 +0200
Subject: [PATCH 8/9] Use correct exception

---
 include/mzpeak/metadata/table.h | 3 +--
 src/metadata/table.cpp          | 6 +++---
 2 files changed, 4 insertions(+), 5 deletions(-)

diff --git a/include/mzpeak/metadata/table.h b/include/mzpeak/metadata/table.h
index a3ecc21..1b79b4b 100644
--- a/include/mzpeak/metadata/table.h
+++ b/include/mzpeak/metadata/table.h
@@ -28,8 +28,7 @@ class Table final {
   ~Table();
 
   /**
-   * Return a group with the given name.  If the group does not
-   * exist in the schema return `nullptr`.
+   * Return a group with the given name.
    */
   std::shared_ptr<Schema::Group> group(std::string_view) const;
 
diff --git a/src/metadata/table.cpp b/src/metadata/table.cpp
index 0fbbb00..06ca5bc 100644
--- a/src/metadata/table.cpp
+++ b/src/metadata/table.cpp
@@ -33,7 +33,7 @@ Table::Impl::Impl(std::unique_ptr<Util::Parquet> parquet)
 
   if (file.data_kind() != Schema::DataKind::Metadata) {
     std::string msg("file is not a metadata file: " + file.file_name());
-    throw ParquetError(msg);
+    throw InvalidFormatError(msg);
   }
 }
 
@@ -56,8 +56,8 @@ std::shared_ptr<Schema::Group> Table::group(std::string_view name) const
   auto it = map->find(std::string(name));
 
   if (it == map->end()) {
-    // FIXME: Replace with the InvalidFormat exception.
-    throw ParquetError("schema is missing the " + std::string(name) + " group");
+    throw InvalidFormatError("schema is missing the " + std::string(name) +
+                             " group");
   } else {
     return it->second;
   }

From 8cbd93b712ceb1ebc5fd4e4f4930732b5204d52d Mon Sep 17 00:00:00 2001
From: "Peter J. Jones" <peter@jonesbunch.com>
Date: Tue, 18 Aug 2026 14:56:53 +0200
Subject: [PATCH 9/9] Add missing documentation

---
 include/mzpeak/util/manager.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/include/mzpeak/util/manager.h b/include/mzpeak/util/manager.h
index 533e589..a3dd001 100644
--- a/include/mzpeak/util/manager.h
+++ b/include/mzpeak/util/manager.h
@@ -19,7 +19,7 @@ directory of this repository.
 namespace MzPeak::Util {
 
 /**
- * FIXME: Write documentation!
+ * File manager for parquet files.
  */
 class Manager final {
 public: