RUST-1998 Remove lossy utf8 as a decoder option #550
Conversation
abr-egn
force-pushed
the
RUST-1998/lossy-utf8
branch
from
3df66fd to
328ac6d
Compare
| @@ -386,13 +386,6 @@ impl crate::DateTime { | |||
| } | |||
| } | |||
|
|
|||
| #[deprecated(since = "2.3.0", note = "Use try_to_rfc3339_string instead.")] | |||
There was a problem hiding this comment.
This is completely unrelated but I happened to notice it and figured we should remove deprecated methods while we're doing a major version release. I did a grep and didn't find any others.
src/raw/document.rs
Outdated
| @@ -531,6 +531,21 @@ impl RawDocument { | |||
| let bytes = self.cstring_bytes_at(start_at)?; | |||
| try_to_str(bytes) | |||
| } | |||
|
|
|||
| /// Copy this into a [`Document`], returning an error if invalid BSON is encountered. | |||
| pub fn to_document(&self) -> Result<Document> { | |||
There was a problem hiding this comment.
I moved this from RawDocumentBuf since that was calling the TryInto impl on RawDocument anyway and this way both types can use it (RawDocumentBuf derefs to this one).
src/raw/error.rs
Outdated
| Self::MalformedValue { message: r_message }, | ||
| ) => l_message == r_message, | ||
| (Self::Utf8EncodingError, Self::Utf8EncodingError) => true, | ||
| (Self::DeError(_), Self::DeError(_)) => true, |
There was a problem hiding this comment.
crate::de:Error doesn't impl PartialEq because it wraps std::io::Error, which also doesn't, hence the manual impl. I figure this whole situation can be revisted for RUST-1406.
This is all needed so I can use crate::de::reader_to_vec in RawDocumentBuf::from_reader; the former uses the crate::de error hierarchy, the latter the crate::raw one.
| @@ -266,7 +268,47 @@ impl<'a> RawElement<'a> { | |||
| }) | |||
| } | |||
|
|
|||
| pub(crate) fn value_utf8_lossy(&self) -> Result<Option<Utf8LossyBson<'a>>> { | |||
| pub fn value_utf8_lossy(&self) -> Result<RawBson> { | |||
There was a problem hiding this comment.
I opted to expose this method rather than the lower-level value_utf8_lossy_inner because in my estimation the confluence between needing to deal with invalid utf8 and needing to save the memory cost of copying a single element is very small, and outside of that this is a strictly more useful interface.
Yup, it'll be equivalent performance-wise, since using the wrapper type just enables behavior the same way |
abr-egn
force-pushed
the
RUST-1998/lossy-utf8
branch
from
96b7844 to
6930185
Compare
abr-egn
force-pushed
the
RUST-1998/lossy-utf8
branch
from
dc3585d to
8679191
Compare
RUST-1998
This removes lossy utf8 decoding as a public option for the bson serde decoder, and as a consequence many of the API entry points. Dealing with lossy utf8 is now done in one of three ways depending on the specific need:
Utf8LossyDeserializationhelper type, if the user is going viaserdeand knows that particular fields might have bad data;RawElement::value_utf8_lossymethod, which allows fine-grained iteration over potentially-invalid raw data, orRawDocument::to_document_utf8_lossymethod, which is a convenience wrapper over the above to bulk-convert an entireRawDocumentat once.