[Rust] proposal: add a new encoder setter method for fixed length array copying from slice ref

[wbprime]

It's very common to serialize an obj as SBE, when obj having a field of String type (or similar) while SBE defining it as fixed length primitive array.

When encoding a field of fixed length primitive array type, one must:

1. first construct an array
2. copy bytes into this array
3. pass this array to Encoder and copy bytes into its backing ByteBuf

It may be more convenient to generate a new method copy from directly slice ref.

For example, there exists a field named "account" of char[16].

Currently it generates a method called "account" with parameter of type &[u8; 16].

/// primitive array field 'account'
/// - min value: 32
/// - max value: 126
/// - null value: 0
/// - characterEncoding: US-ASCII
/// - semanticType: String
/// - encodedOffset: 0
/// - encodedLength: 16
/// - version: 0
#[inline]
pub fn account(&mut self, value: &[u8; 16]) {
    let offset = self.offset;
    let buf = self.get_buf_mut();
    buf.put_bytes_at(offset, value);
}

Following method is proposed to be generated as well:

pub fn account_from_slice_ref(&mut self, value: &[u8]) {
    // Typical implementation

    let offset = self.offset;
    let buf = self.get_buf_mut();

    let len = value.len();
    // How to deal with cases input slice is too large: quietly copy leading bytes of fixed size or return Result or panic ?
    if len > 16 {
        buf.put_slice_at(offset, &value[..16]);
    } else {
        buf.put_slice_at(offset, value);

        // Padding with ZERO
        buf.put_u8_at(len + 0, 0_u8);
    }
}

[wbprime]

Hi Iv created a new pr #1026 to address this issue.

Please tell me what you think.

[vyazelenko]

@mward Can you have a look? Thanks.

[mward]

I'll take a look. Thanks

[mward]

@wbprime - I like the idea of supporting slices, in addition to arrays, but PR #1026 generates an enormous amount of code. I'm not sure it is necessary to support all permutations of undersized slices. Do any other implementations provide functionality similar to this?

[wbprime]

@mward Ok, will try to compact the implementation.

[wbprime]

@mward just updated the implementation:

1. use loops to copy data for a sub set of permutations
2. use zero when padding, drop support for customed pading value

Could you help review it ?

[mward]

I'll check it out

[mward]

@wbprime I created a new PR #1040 to resolve this. I incorporated the tests you created. Please take a look

[wbprime]

Hi @mward I think we should not introduce breaking changes in generated code say for primitive array setters. Maybe you can choose a new name for slice setter, like the iter variant do.

[wbprime]

@mward I just leave some comments on your pr, feel free to accept or reject them.

As en is not my native language, just ignore the words or sentences may look impolite/rude to you.

[mward]

Hi @mward I think we should not introduce breaking changes in generated code say for primitive array setters. Maybe you can choose a new name for slice setter, like the iter variant do.

changing function param from array to slice doesn't break anything, consumer of library won't need to make any changes

[wbprime]

An array ref is not the safe thing as a slice ref. An array ref has length info built into its type while a slice ref doest.

[mward]

True, I meant signature wise