Better error message when Table.read cannot open the file (#159)

Our current error message on `Table.read` not being able top open/read file is not helpful. Reason is two-fold:
1) the error pretty printing call was missing an argument, so we got error-formetting-error instead of actual error;
2) the actual error wouldn't be helpful anyway -- we ignored the actual error with reading file (that can be caused by wrong permission settings or wrong path), because we expect failure possibility when trying to match proper parser for given file.

Now not being able to read file is a special class of errors that stop the parser-matching routine and allow to give proper diagnostics to the user.

native_libs/src/Core/Common.h

@@ -167,12 +167,15 @@ std::vector<T> iotaVector(size_t N, T from = T{})
     return ret;
 }
 
+#define THROW_OBJ(EXC_TO_THROW) do {                                                     \
+    std::cerr << __FILE__ << " " << __LINE__ << " " << EXC_TO_THROW.what() << std::endl; \
+    throw EXC_TO_THROW;                                                                  \
+} while(0)
 
-#define THROW(message, ...)  do {                                         \
-	auto msg_ =  fmt::format(message, ##__VA_ARGS__);                     \
-	std::cerr << __FILE__ << " " << __LINE__ << " " << msg_ << std::endl; \
-	std::runtime_error e_to_throw{fmt::format(message, ##__VA_ARGS__)};   \
-	throw e_to_throw;                                                     \
+#define THROW(message, ...)  do {                                       \
+    auto msg_ =  fmt::format(message, ##__VA_ARGS__);                   \
+    std::runtime_error e_to_throw{fmt::format(message, ##__VA_ARGS__)}; \
+    THROW_OBJ(e_to_throw);                                              \
 } while(0)
 
 namespace detail

native_libs/src/IO/IO.cpp

@@ -86,7 +86,7 @@ std::shared_ptr<arrow::Table> readTableFromFile(std::string_view filepath)
         if(auto table = handler->tryReading(filepath))
             return table;
 
-    THROW("Failed to load file {}: it doesn't parse with default settings as any of the supported formats");
+    THROW("Failed to load file {}: it doesn't parse with default settings as any of the supported formats", filepath);
 }
 
 void writeTableToFile(std::string_view filepath, const arrow::Table &table)
@@ -132,7 +132,13 @@ std::ifstream openFileToRead(std::string_view filepath)
 #endif
 
     if(!in)
-        THROW("Cannot open the file to read: {}", filepath);
+    {
+        // Do not bother catching std::ifstream::failure and checking its supposedly inherited `std::system_error::code`.
+        // libstdc++ does not implement it and even on platforms where it works, the error message is not helpful at all.
+        // If we really want to give proper diagnostics for filesystem errors, we would need to get platform-specific.
+        CannotOpenToRead cannotRead{ filepath };
+        THROW_OBJ(cannotRead);
+    }
 
     return in;
 }
@@ -183,6 +189,11 @@ std::shared_ptr<arrow::Table> TableFileHandler::tryReading(std::string_view file
 
         return read(filePath);
     }
+    catch(CannotOpenToRead &)
+    {
+        // this is a serious one -- there is no sense in trying reading something that cannot be read anyway
+        throw;
+    }
     catch(...)
     {
         return nullptr;

native_libs/src/IO/IO.h

@@ -19,6 +19,13 @@ namespace arrow
     class DataType;
 }
 
+struct CannotOpenToRead : std::runtime_error
+{
+    CannotOpenToRead(std::string_view path)
+        : std::runtime_error(fmt::format("Cannot open file to read: {}", path))
+    {}
+};
+
 struct TakeFirstRowAsHeaders {};
 struct GenerateColumnNames {};
 

native_libs/test/Tests.cpp

@@ -1162,6 +1162,12 @@ BOOST_AUTO_TEST_CASE(SliceBoundsChecking)
     BOOST_CHECK_NO_THROW(slice(column, 0, 5));
 }
 
+BOOST_AUTO_TEST_CASE(ReadMissingCsvFile)
+{
+    const auto pathUnlikelyToExist = "data/gbhfudbfiugbiu.csv";
+    BOOST_CHECK_THROW(readTableFromFile(pathUnlikelyToExist), CannotOpenToRead);
+}
+
 BOOST_AUTO_TEST_CASE(ReadCsvFileWithBOM)
 {
     const auto t = FormatCSV{}.read("data/fileWithBOM.csv");