Final polish

 * Run emscripten tests
 * Very minor BasicCSVParser constructor fix
 * Documentation updates

Author: vincentlaucsb

.github/workflows/cmake-multi-platform.yml

@@ -146,11 +146,16 @@ jobs:
         -DCSV_ENABLE_THREADS=OFF
         -DCSV_CXX_STANDARD=20
         -DCSV_BUILD_SINGLE_INCLUDE_TEST=ON
+        -DCMAKE_CROSSCOMPILING_EMULATOR=node
         -DCMAKE_BUILD_TYPE=Release
         -S ${{ github.workspace }}
 
     - name: Build (single-threaded emscripten)
       run: cmake --build ${{ github.workspace }}/build-single-thread-emscripten --config Release
 
+    - name: Test (single-threaded emscripten)
+      working-directory: ${{ github.workspace }}/build-single-thread-emscripten
+      run: ctest --build-config Release --output-on-failure
+
 
 

include/internal/basic_csv_parser.cpp

@@ -6,11 +6,19 @@ namespace csv {
     namespace internals {
         CSV_INLINE size_t get_file_size(csv::string_view filename) {
             std::ifstream infile(std::string(filename), std::ios::binary);
+            if (!infile.is_open()) {
+                throw std::runtime_error("Cannot open file " + std::string(filename));
+            }
+
             const auto start = infile.tellg();
             infile.seekg(0, std::ios::end);
             const auto end = infile.tellg();
 
-            return end - start;
+            if (start < 0 || end < 0) {
+                throw std::runtime_error("Cannot determine file size for " + std::string(filename));
+            }
+
+            return static_cast<size_t>(end - start);
         }
 
         CSV_INLINE std::string get_csv_head(csv::string_view filename) {

include/internal/csv_reader.hpp

@@ -159,25 +159,26 @@ namespace csv {
          *  Constructors for iterating over large files and parsing in-memory sources.
          */
          ///@{
-        /** @brief Construct CSVReader from filename using memory-mapped I/O
+        /** @brief Construct CSVReader from filename.
          * 
-         * CODE PATH 1 of 2: Uses MmapParser with mio library for maximum performance.
-         * This is fundamentally different from the stream-based constructor below.
+         * Native builds use CODE PATH 1 of 2: MmapParser with mio for maximum performance.
+         * Emscripten builds fall back to the stream-based implementation because mmap is unavailable.
          * 
-         * @note Bugs can exist in this path independently of the stream path (and vice versa)
-         * @note When writing tests that validate I/O behavior, BOTH paths must be tested
-         * @see StreamParser for the alternative implementation
+         * @note On native builds, bugs can exist in this path independently of the stream path
+         * @note When writing tests that validate I/O behavior, test both filename and stream constructors
+         * @see StreamParser for the stream-based alternative
          */
         CSVReader(csv::string_view filename, CSVFormat format = CSVFormat::guess_csv());
 
         /** @brief Construct CSVReader from std::istream
          * 
-         * CODE PATH 2 of 2: Uses StreamParser with different internal implementation than
-         * the memory-mapped constructor above. Issue #281 was specific to THIS path only.
+         * Uses StreamParser. On native builds this is CODE PATH 2 of 2 and remains independent
+         * from the filename-based mmap path. On Emscripten, the filename constructor also funnels
+         * through this implementation.
          *
          *  @tparam TStream An input stream deriving from `std::istream`
          *  @note CSV format guessing works differently here - must manually specify dialect
-         *  @note When writing tests that validate I/O behavior, BOTH paths must be tested
+         *  @note On native builds, tests that validate I/O behavior should cover both constructors
          *  @see MmapParser for the memory-mapped alternative
          */
         template<typename TStream,

include/internal/data_type.hpp

@@ -34,10 +34,17 @@ namespace csv {
     static_assert(DataType::CSV_INT64 < DataType::CSV_DOUBLE, "Integer types should come before floating point value types.");
 
     namespace internals {
-        /** Compute 10 to the power of n */
+        /** Compute 10 to the power of n.
+         *  Only integral exponents are supported; fractional exponents
+         *  are never needed since CSV scientific notation exponents are
+         *  always integers (enforced by the CSV_INT8..CSV_INT64 guard
+         *  in _process_potential_exponential before calling this).
+         */
         template<typename T>
         CSV_CONST CONSTEXPR_14
         long double pow10(const T& n) noexcept {
+            static_assert(std::is_integral<T>::value, "pow10 only supports integral exponents");
+            
             long double multiplicand = n > 0 ? 10 : 0.1,
                 ret = 1;
 
@@ -186,7 +193,7 @@ namespace csv {
 
             // Exponents in scientific notation should not be decimal numbers
             if (result >= DataType::CSV_INT8 && result < DataType::CSV_DOUBLE) {
-                if (out) *out = coeff * pow10(exponent);
+                if (out) *out = coeff * pow10(static_cast<long long>(exponent));
                 return DataType::CSV_DOUBLE;
             }