RUST-877 Delay replacement document serialization until Operation::build
#942
Conversation
Operation::buildOperation::build
a7e4c93 to
8fe50c3
Compare
| @@ -119,6 +125,17 @@ pub(crate) fn read_document_bytes<R: Read>(mut reader: R) -> Result<Vec<u8>> { | |||
| Ok(bytes) | |||
| } | |||
|
|
|||
| pub(crate) fn extend_raw_document_buf( | |||
There was a problem hiding this comment.
A slightly optimized version of this could just append all of the non-size bytes from other to this and then re-calculate the size, might be something worth adding to the bson API.
There was a problem hiding this comment.
Agreed, filed RUST-1754 for this.
| @@ -177,6 +178,19 @@ impl<T> Command<T> { | |||
| } | |||
| } | |||
|
|
|||
| impl Command<RawDocumentBuf> { | |||
| pub(crate) fn into_bson_bytes(mut self) -> Result<Vec<u8>> { | |||
There was a problem hiding this comment.
This method (when it's called in Operation::serialize_command) bypasses running the body of the command back through the serializer (i.e. the default implementation for serialize_command) and instead appends the rest of the command to the body. The tradeoff here is that the raw BSON created on 187 is immediately copied on line 189; we could manually serialize and append each field directly to command instead, but I think the size of rest_of_command is small enough that it shouldn't make much of a difference in the vast majority of cases.
Most Commands for other operations currently use the default Document type for T, but in the future (maybe in tandem with RUST-1747) I think we should change them to use RawDocumentBuf and then standardize this approach for serializing commands.
There was a problem hiding this comment.
We've talked in the past about deleting these low-value operation tests as they become outdated. These changes broke most of them so I just removed the whole file.
src/operation/update/test.rs
Outdated
| @@ -177,6 +178,19 @@ impl<T> Command<T> { | |||
| } | |||
| } | |||
|
|
|||
| impl Command<RawDocumentBuf> { | |||
| pub(crate) fn into_bson_bytes(mut self) -> Result<Vec<u8>> { | |||
| @@ -119,6 +125,17 @@ pub(crate) fn read_document_bytes<R: Read>(mut reader: R) -> Result<Vec<u8>> { | |||
| Ok(bytes) | |||
| } | |||
|
|
|||
| pub(crate) fn extend_raw_document_buf( | |||
There was a problem hiding this comment.
Agreed, filed RUST-1754 for this.
| pub(crate) fn to_bson(&self) -> Bson { | ||
| match self { | ||
| Hint::Keys(ref d) => Bson::Document(d.clone()), | ||
| Hint::Name(ref s) => Bson::String(s.clone()), | ||
| } | ||
| } | ||
|
|
||
| pub(crate) fn to_raw_bson(&self) -> Result<RawBson> { |
There was a problem hiding this comment.
It seems like this could just be to_bson().try_into()? ?
There was a problem hiding this comment.
I added this as a minor optimization to avoid an intermediary allocation of a regular Bson via to_bson. It did seem unnecessary to have both to_bson and to_raw_bson though so I just refactored the few uses of to_bson and removed it.
Ideally we'd have a to_raw_bson(value: impl Serialize) method in bson, but I'm not sure how that would actually work; the raw serializer spits out raw document bytes as opposed to the non-raw serializer, which generates BSON values.
| ns: Namespace, | ||
| query: Document, | ||
| modification: Modification<'a, R>, | ||
| human_readable_serialization: Option<bool>, |
There was a problem hiding this comment.
Could this be an un-serialized field in FindAndModifyOptions rather than adding it as an explicit parameter?
There was a problem hiding this comment.
this was slightly more ergonomic because it allows implementing From for the various options rather than writing custom methods. this is maybe just personal preference too but it made sense to me to have only fields that will be serialized and sent to the server within the options. this did make me realize that there's no need to unwrap_or_default the options anymore so I removed that logic
No description provided.