1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
#[cfg(feature = "write")]
use crate::internal::encodings::varint::encode_prefix_varint;
use crate::prelude::*;
use std::vec::IntoIter;

#[cfg(feature = "write")]
impl<'a, T: Writable<'a>> Writable<'a> for Vec<T> {
    type WriterArray = VecArrayWriter<'a, T::WriterArray>;
    fn write_root<'b: 'a>(value: &'b Self, bytes: &mut Vec<u8>, lens: &mut Vec<usize>) -> RootTypeId {
        match value.len() {
            0 => RootTypeId::Array0,
            1 => {
                let type_index = bytes.len();
                bytes.push(0);
                let type_id = T::write_root(&value[0], bytes, lens);
                bytes[type_index] = type_id.into();
                RootTypeId::Array1
            }
            _ => {
                // TODO: Seems kind of redundant to have both the array len,
                // and the bytes len. Though, it's not for obvious reasons.
                // Maybe sometimes we can infer from context. Eg: bool
                encode_prefix_varint(value.len() as u64, bytes);

                let type_index = bytes.len();
                bytes.push(0);

                // TODO: When there are types that are already
                // primitive (eg: Vec<f64>) it doesn't make sense
                // to buffer at this level. Specialization may
                // be useful here.
                //
                // TODO: See below, and just call buffer on the vec
                // and flush it!
                let mut writer = T::WriterArray::default();
                for item in value {
                    writer.buffer(item);
                }
                let type_id = writer.flush(bytes, lens);
                bytes[type_index] = type_id.into();
                RootTypeId::ArrayN
            }
        }
    }
}

#[cfg(feature = "read")]
impl<T: Readable> Readable for Vec<T> {
    type ReaderArray = Option<VecArrayReader<T::ReaderArray>>;
    fn read(sticks: DynRootBranch<'_>) -> ReadResult<Self> {
        match sticks {
            DynRootBranch::Array0 => Ok(Vec::new()),
            DynRootBranch::Array1(inner) => {
                let inner = T::read(*inner)?;
                Ok(vec![inner])
            }
            DynRootBranch::Array { len, values } => {
                let mut v = Vec::with_capacity(len);
                // TODO: Some of what the code is actually doing here is silly.
                // Actual ReaderArray's may be IntoIter, which moved out of a Vec
                // that we wanted in the first place. An overload here would be nice.
                let mut reader = T::ReaderArray::new(values)?;
                for _ in 0..len {
                    v.push(reader.read_next()?);
                }
                Ok(v)
                // We could try to verify the file here by trying
                // to read one more value, but that doesn't work well
                // for block based readers (eg: bool) because they round up
                // the number of values read to pad the block
            }
            _ => Err(ReadError::SchemaMismatch),
        }
    }
}

#[cfg(feature = "write")]
#[derive(Debug, Default)]
pub struct VecArrayWriter<'a, T> {
    // TODO: usize
    len: <u64 as Writable<'a>>::WriterArray,
    // Using Option here enables recursion when necessary.
    values: Option<T>,
}

#[cfg(feature = "read")]
pub struct VecArrayReader<T> {
    // TODO: usize
    len: IntoIter<u64>,
    values: T,
}

#[cfg(feature = "write")]
impl<'a, T: WriterArray<'a>> WriterArray<'a> for VecArrayWriter<'a, T> {
    type Write = Vec<T::Write>;
    fn buffer<'b: 'a>(&mut self, value: &'b Self::Write) {
        // TODO: Consider whether buffer should actually just
        // do something non-flat, (like literally push the Vec<T> into another Vec<T>)
        // and the flattening could happen later at flush time. This may reduce memory cost.
        // Careful though.
        // I feel though that somehow this outer buffer type
        // could fix the specialization problem above for single-vec
        // values.
        self.len.buffer(&(value.len() as u64));
        let values = self.values.get_or_insert_with(Default::default);
        for item in value {
            values.buffer(item);
        }
    }
    fn flush(self, bytes: &mut Vec<u8>, lens: &mut Vec<usize>) -> ArrayTypeId {
        let Self { len, values } = self;
        if let Some(values) = values {
            let type_index = bytes.len();
            bytes.push(0);
            let type_id = values.flush(bytes, lens);
            debug_assert_ne!(type_id, ArrayTypeId::Void); // If this is Void, it's ambigous
            bytes[type_index] = type_id.into();

            // TODO: Maybe combine the permutations of valid int compressors here with ArrayVar to save a byte
            // here every time. Eg: ArrayVarSimple16 ArrayVarIntPrefixVar
            let type_index = bytes.len();
            bytes.push(0);
            let len_type_id = len.flush(bytes, lens);
            bytes[type_index] = len_type_id.into();
        } else {
            bytes.push(ArrayTypeId::Void.into())
        }

        ArrayTypeId::ArrayVar
    }
}

#[cfg(feature = "read")]
impl<T: ReaderArray> ReaderArray for Option<VecArrayReader<T>> {
    type Read = Vec<T::Read>;
    fn new(sticks: DynArrayBranch<'_>) -> ReadResult<Self> {
        match sticks {
            DynArrayBranch::Array0 => Ok(None),
            DynArrayBranch::Array { len, values } => {
                let values = T::new(*values)?;
                let len = <<u64 as Readable>::ReaderArray as ReaderArray>::new(*len)?;
                Ok(Some(VecArrayReader { len, values }))
            }
            _ => Err(ReadError::SchemaMismatch),
        }
    }
    fn read_next(&mut self) -> ReadResult<Self::Read> {
        if let Some(inner) = self {
            let len = inner.len.read_next()?;
            let mut result = Vec::with_capacity(len as usize);
            for _ in 0..len {
                result.push(inner.values.read_next()?);
            }
            Ok(result)
        } else {
            Ok(Vec::new())
        }
    }
}