Adds type conversions between PG types and arrow types #19
Conversation
| let numeric: AnyNumeric = unsafe { | ||
| direct_function_call( | ||
| pg_sys::numeric_in, | ||
| &[numeric_str.into_datum(), 0.into_datum(), 0.into_datum()], |
There was a problem hiding this comment.
the 0 typemod seems a bit strange to me, since -1 is normally used as invalid typemod, but 0 kind of works by accident because it does not pass is_valid_numeric_typmod
I think it's probably ok to pass -1 here with a comment that we don't need numeric_in to validate the input.
| let digit = (decimal % 10) as i8; | ||
| decimal_digits.push(digit); | ||
| decimal /= 10; | ||
| } |
There was a problem hiding this comment.
Is there any way to construct the numeric directly without creating a string?
There was a problem hiding this comment.
probably via writing as bytea and then calling numeric_recv, but string seems like a better idea to me
| .expect("cannot adjust PG timestamptz to Unix timestamptz") | ||
| }; | ||
|
|
||
| let adjusted_timestamptz_as_bytes: Vec<u8> = unsafe { |
There was a problem hiding this comment.
maybe:
// convert timestamptz to big endian bytes using timestamptz_send
There was a problem hiding this comment.
I think expect part sounds like a comment.
let adjusted_timestamptz_as_bytes: Vec<u8> = unsafe {
direct_function_call(
pg_sys::timestamptz_send,
&[adjusted_timestamptz.into_datum()],
)
.expect("cannot convert timestamptz to bytes")
};|
|
||
| let sign = if numeric_str.starts_with('-') { -1 } else { 1 }; | ||
|
|
||
| let numeric_str = numeric_str.replace(['-', '+', '.'], ""); |
There was a problem hiding this comment.
could use a comment on why we don't need to worry about where the fraction point is.
| pub(crate) fn numeric_to_i128(numeric: AnyNumeric) -> i128 { | ||
| // obtain numeric's string representation | ||
| let numeric_str: &CStr = unsafe { | ||
| direct_function_call(pg_sys::numeric_out, &[numeric.into_datum()]) |
There was a problem hiding this comment.
I kind liked that we were able to use send functions above. Is that not a possibility here? It seems at least the digits part looks vaguely similar.
There was a problem hiding this comment.
it generates different byte representation than what parquet expects.
There was a problem hiding this comment.
could use a comment then
aykut-bozkurt
force-pushed
the
crunchy_review
branch
from
157b1e6 to
0d37a90
Compare
aykut-bozkurt
force-pushed
the
crunchy_main
branch
from
9b40920 to
e7e694f
Compare
|
Addressed at |
These conversions will be needed when we convert PG types to arrow types and vice-versa.