On Sat, Nov 9, 2024, at 08:07, Joel Jacobson wrote:
> Here's a draft of an idea I'm considering (not yet implemented):
I realize the last part about optional quoting is unnecessary,
since if quoting is desired, users could just use the 'csv' format.
Revised draft of the idea (not yet implemented):
- Fast path for newline-free types:
For the list of built-in types where we know the ::text representation cannot
contain newlines, we take the fast path in NextCopyFromRawFields(),
pointing cstate->raw_fields[0] directly to cstate->line_buf.data.
- Handling newlines for other types:
For any other types, we would need to scan the string for newline characters.
If a newline is encountered, it results in an error.
This brings up the question on what to offer users wanting to export text
values containing newlines.
To address this need, I think that's out of scope for the 'list' format,
and is better handled by a separate 'value' format:
- Such a format would be specialized for exporting a value "as is" to a file,
or importing en entire file as a single value.
- Such a value could be a physical single-column single-row,
but could also be constructed using e.g. string_agg().
- The 'value' format could also easily support import/export
binary data (bytea), to e.g. allow importing/exporting images, etc.
Dimensionality perspective on formats:
2D formats: 'text', 'csv', 'binary' (tabular formats)
1D format: 'list' (single-column)
0D format: 'value' (single-column, single-row)
/Joel
