No column detected from CSV with no data (header row only)

[AlexandreGarino]

Hi,

I read a CSV file with no data (header row only) and I noticed that no column was detected.

The resulting table is empty.

final ByteArrayInputStream inputStream = new ByteArrayInputStream("Column1;Column2;Column3\n".getBytes(StandardCharsets.UTF_8));

final CsvReadOptions readOptions = ((CsvReadOptions.Builder) CsvReadOptions.builder(inputStream)
        .columnTypesToDetect(Arrays.asList(ColumnType.STRING)))
        .header(true)
        .separator(';')
        .build();

final Table table = Table.read().csv(readOptions);

assertThat(table.columns(), hasSize(0)); // OK

Did I make something wrong?

Any help would be greatly appreciated.

[lwhite1]

I guess this would be a bug, but a fairly minor one from my perspective.

I don't see the use-case for reading a file with nothing but a header line. Tablesaw is geared more towards analyzing existing datasets than on building datasets in memory. Of course you can create a table entirely in code, but I've only ever used that for testing.

Have you found any work arounds? I think i would look at either (a) trying to load the file with the column types pre-specified (I notice all your columns are strings), or (b), just reading the headers using a standard java file reading approach and creating the table in code, by looping over the names

[AlexandreGarino]

Hi,

The code snippet is here just to reproduce the bug.

The real code scan pragmatically a folder for new CSV files and apply some business logic (we compute new columns based on existing columns) on some criteria.

For now, when we read the table, if the table is empty we copy the file as-is.

[imagejan]

Is there an option now for keeping all columns when reading a header-only file (even when the column types can't be determined of course)?

In our use case, we write many tables automated in batch, and some of them could potentially end up being empty. Nevertheless, we have the same column headers across files. Currently, we get an exception (in MoBIE which depends on tablesaw) because tablesaw returns a Table without columns at all. I would have expected at least all the columnNames() be the same as in the (header-only) input file.

/cc @tischi

[lwhite1]

Columns have to have a type. Maybe if you’re using the same columns repeatedly, you could include one row of dummy data and then delete it? If that’s impossible, you could read the column names and build the table in code, perhaps making them all string columns. They would then need to be converted to the proper types manually.

…

On Wed, Mar 29, 2023 at 4:05 AM Jan Eglinger ***@***.***> wrote: Is there an option now for keeping all columns when reading a header-only file (even when the column *types* can't be determined of course)? In our use case, we write many tables automated in batch, and some of them could potentially end up being empty. Nevertheless, we have the same column headers across files. Currently, we get an exception (in MoBIE which depends on tablesaw) because tablesaw returns a Table without columns at all. I would have expected at least all the columnNames() be the same as in the (header-only) input file. — Reply to this email directly, view it on GitHub <#815 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AA2FPAUWTJOFIZ7WZ7P5QOTW6PUNXANCNFSM4ORMO2RA> . You are receiving this because you modified the open/close state.Message ID: ***@***.***>