chore: add vendor directory

3 files changed, 871 insertions, 0 deletions

diff --git a/vendor/base64/src/engine/general_purpose/decode.rs b/vendor/base64/src/engine/general_purpose/decode.rs
new file mode 100644
index 00000000..b55d3fc5
--- /dev/null
+++ b/vendor/base64/src/engine/general_purpose/decode.rs
@@ -0,0 +1,357 @@
+use crate::{
+    engine::{general_purpose::INVALID_VALUE, DecodeEstimate, DecodeMetadata, DecodePaddingMode},
+    DecodeError, DecodeSliceError, PAD_BYTE,
+};
+
+#[doc(hidden)]
+pub struct GeneralPurposeEstimate {
+    /// input len % 4
+    rem: usize,
+    conservative_decoded_len: usize,
+}
+
+impl GeneralPurposeEstimate {
+    pub(crate) fn new(encoded_len: usize) -> Self {
+        let rem = encoded_len % 4;
+        Self {
+            rem,
+            conservative_decoded_len: (encoded_len / 4 + (rem > 0) as usize) * 3,
+        }
+    }
+}
+
+impl DecodeEstimate for GeneralPurposeEstimate {
+    fn decoded_len_estimate(&self) -> usize {
+        self.conservative_decoded_len
+    }
+}
+
+/// Helper to avoid duplicating num_chunks calculation, which is costly on short inputs.
+/// Returns the decode metadata, or an error.
+// We're on the fragile edge of compiler heuristics here. If this is not inlined, slow. If this is
+// inlined(always), a different slow. plain ol' inline makes the benchmarks happiest at the moment,
+// but this is fragile and the best setting changes with only minor code modifications.
+#[inline]
+pub(crate) fn decode_helper(
+    input: &[u8],
+    estimate: GeneralPurposeEstimate,
+    output: &mut [u8],
+    decode_table: &[u8; 256],
+    decode_allow_trailing_bits: bool,
+    padding_mode: DecodePaddingMode,
+) -> Result<DecodeMetadata, DecodeSliceError> {
+    let input_complete_nonterminal_quads_len =
+        complete_quads_len(input, estimate.rem, output.len(), decode_table)?;
+
+    const UNROLLED_INPUT_CHUNK_SIZE: usize = 32;
+    const UNROLLED_OUTPUT_CHUNK_SIZE: usize = UNROLLED_INPUT_CHUNK_SIZE / 4 * 3;
+
+    let input_complete_quads_after_unrolled_chunks_len =
+        input_complete_nonterminal_quads_len % UNROLLED_INPUT_CHUNK_SIZE;
+
+    let input_unrolled_loop_len =
+        input_complete_nonterminal_quads_len - input_complete_quads_after_unrolled_chunks_len;
+
+    // chunks of 32 bytes
+    for (chunk_index, chunk) in input[..input_unrolled_loop_len]
+        .chunks_exact(UNROLLED_INPUT_CHUNK_SIZE)
+        .enumerate()
+    {
+        let input_index = chunk_index * UNROLLED_INPUT_CHUNK_SIZE;
+        let chunk_output = &mut output[chunk_index * UNROLLED_OUTPUT_CHUNK_SIZE
+            ..(chunk_index + 1) * UNROLLED_OUTPUT_CHUNK_SIZE];
+
+        decode_chunk_8(
+            &chunk[0..8],
+            input_index,
+            decode_table,
+            &mut chunk_output[0..6],
+        )?;
+        decode_chunk_8(
+            &chunk[8..16],
+            input_index + 8,
+            decode_table,
+            &mut chunk_output[6..12],
+        )?;
+        decode_chunk_8(
+            &chunk[16..24],
+            input_index + 16,
+            decode_table,
+            &mut chunk_output[12..18],
+        )?;
+        decode_chunk_8(
+            &chunk[24..32],
+            input_index + 24,
+            decode_table,
+            &mut chunk_output[18..24],
+        )?;
+    }
+
+    // remaining quads, except for the last possibly partial one, as it may have padding
+    let output_unrolled_loop_len = input_unrolled_loop_len / 4 * 3;
+    let output_complete_quad_len = input_complete_nonterminal_quads_len / 4 * 3;
+    {
+        let output_after_unroll = &mut output[output_unrolled_loop_len..output_complete_quad_len];
+
+        for (chunk_index, chunk) in input
+            [input_unrolled_loop_len..input_complete_nonterminal_quads_len]
+            .chunks_exact(4)
+            .enumerate()
+        {
+            let chunk_output = &mut output_after_unroll[chunk_index * 3..chunk_index * 3 + 3];
+
+            decode_chunk_4(
+                chunk,
+                input_unrolled_loop_len + chunk_index * 4,
+                decode_table,
+                chunk_output,
+            )?;
+        }
+    }
+
+    super::decode_suffix::decode_suffix(
+        input,
+        input_complete_nonterminal_quads_len,
+        output,
+        output_complete_quad_len,
+        decode_table,
+        decode_allow_trailing_bits,
+        padding_mode,
+    )
+}
+
+/// Returns the length of complete quads, except for the last one, even if it is complete.
+///
+/// Returns an error if the output len is not big enough for decoding those complete quads, or if
+/// the input % 4 == 1, and that last byte is an invalid value other than a pad byte.
+///
+/// - `input` is the base64 input
+/// - `input_len_rem` is input len % 4
+/// - `output_len` is the length of the output slice
+pub(crate) fn complete_quads_len(
+    input: &[u8],
+    input_len_rem: usize,
+    output_len: usize,
+    decode_table: &[u8; 256],
+) -> Result<usize, DecodeSliceError> {
+    debug_assert!(input.len() % 4 == input_len_rem);
+
+    // detect a trailing invalid byte, like a newline, as a user convenience
+    if input_len_rem == 1 {
+        let last_byte = input[input.len() - 1];
+        // exclude pad bytes; might be part of padding that extends from earlier in the input
+        if last_byte != PAD_BYTE && decode_table[usize::from(last_byte)] == INVALID_VALUE {
+            return Err(DecodeError::InvalidByte(input.len() - 1, last_byte).into());
+        }
+    };
+
+    // skip last quad, even if it's complete, as it may have padding
+    let input_complete_nonterminal_quads_len = input
+        .len()
+        .saturating_sub(input_len_rem)
+        // if rem was 0, subtract 4 to avoid padding
+        .saturating_sub((input_len_rem == 0) as usize * 4);
+    debug_assert!(
+        input.is_empty() || (1..=4).contains(&(input.len() - input_complete_nonterminal_quads_len))
+    );
+
+    // check that everything except the last quad handled by decode_suffix will fit
+    if output_len < input_complete_nonterminal_quads_len / 4 * 3 {
+        return Err(DecodeSliceError::OutputSliceTooSmall);
+    };
+    Ok(input_complete_nonterminal_quads_len)
+}
+
+/// Decode 8 bytes of input into 6 bytes of output.
+///
+/// `input` is the 8 bytes to decode.
+/// `index_at_start_of_input` is the offset in the overall input (used for reporting errors
+/// accurately)
+/// `decode_table` is the lookup table for the particular base64 alphabet.
+/// `output` will have its first 6 bytes overwritten
+// yes, really inline (worth 30-50% speedup)
+#[inline(always)]
+fn decode_chunk_8(
+    input: &[u8],
+    index_at_start_of_input: usize,
+    decode_table: &[u8; 256],
+    output: &mut [u8],
+) -> Result<(), DecodeError> {
+    let morsel = decode_table[usize::from(input[0])];
+    if morsel == INVALID_VALUE {
+        return Err(DecodeError::InvalidByte(index_at_start_of_input, input[0]));
+    }
+    let mut accum = u64::from(morsel) << 58;
+
+    let morsel = decode_table[usize::from(input[1])];
+    if morsel == INVALID_VALUE {
+        return Err(DecodeError::InvalidByte(
+            index_at_start_of_input + 1,
+            input[1],
+        ));
+    }
+    accum |= u64::from(morsel) << 52;
+
+    let morsel = decode_table[usize::from(input[2])];
+    if morsel == INVALID_VALUE {
+        return Err(DecodeError::InvalidByte(
+            index_at_start_of_input + 2,
+            input[2],
+        ));
+    }
+    accum |= u64::from(morsel) << 46;
+
+    let morsel = decode_table[usize::from(input[3])];
+    if morsel == INVALID_VALUE {
+        return Err(DecodeError::InvalidByte(
+            index_at_start_of_input + 3,
+            input[3],
+        ));
+    }
+    accum |= u64::from(morsel) << 40;
+
+    let morsel = decode_table[usize::from(input[4])];
+    if morsel == INVALID_VALUE {
+        return Err(DecodeError::InvalidByte(
+            index_at_start_of_input + 4,
+            input[4],
+        ));
+    }
+    accum |= u64::from(morsel) << 34;
+
+    let morsel = decode_table[usize::from(input[5])];
+    if morsel == INVALID_VALUE {
+        return Err(DecodeError::InvalidByte(
+            index_at_start_of_input + 5,
+            input[5],
+        ));
+    }
+    accum |= u64::from(morsel) << 28;
+
+    let morsel = decode_table[usize::from(input[6])];
+    if morsel == INVALID_VALUE {
+        return Err(DecodeError::InvalidByte(
+            index_at_start_of_input + 6,
+            input[6],
+        ));
+    }
+    accum |= u64::from(morsel) << 22;
+
+    let morsel = decode_table[usize::from(input[7])];
+    if morsel == INVALID_VALUE {
+        return Err(DecodeError::InvalidByte(
+            index_at_start_of_input + 7,
+            input[7],
+        ));
+    }
+    accum |= u64::from(morsel) << 16;
+
+    output[..6].copy_from_slice(&accum.to_be_bytes()[..6]);
+
+    Ok(())
+}
+
+/// Like [decode_chunk_8] but for 4 bytes of input and 3 bytes of output.
+#[inline(always)]
+fn decode_chunk_4(
+    input: &[u8],
+    index_at_start_of_input: usize,
+    decode_table: &[u8; 256],
+    output: &mut [u8],
+) -> Result<(), DecodeError> {
+    let morsel = decode_table[usize::from(input[0])];
+    if morsel == INVALID_VALUE {
+        return Err(DecodeError::InvalidByte(index_at_start_of_input, input[0]));
+    }
+    let mut accum = u32::from(morsel) << 26;
+
+    let morsel = decode_table[usize::from(input[1])];
+    if morsel == INVALID_VALUE {
+        return Err(DecodeError::InvalidByte(
+            index_at_start_of_input + 1,
+            input[1],
+        ));
+    }
+    accum |= u32::from(morsel) << 20;
+
+    let morsel = decode_table[usize::from(input[2])];
+    if morsel == INVALID_VALUE {
+        return Err(DecodeError::InvalidByte(
+            index_at_start_of_input + 2,
+            input[2],
+        ));
+    }
+    accum |= u32::from(morsel) << 14;
+
+    let morsel = decode_table[usize::from(input[3])];
+    if morsel == INVALID_VALUE {
+        return Err(DecodeError::InvalidByte(
+            index_at_start_of_input + 3,
+            input[3],
+        ));
+    }
+    accum |= u32::from(morsel) << 8;
+
+    output[..3].copy_from_slice(&accum.to_be_bytes()[..3]);
+
+    Ok(())
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    use crate::engine::general_purpose::STANDARD;
+
+    #[test]
+    fn decode_chunk_8_writes_only_6_bytes() {
+        let input = b"Zm9vYmFy"; // "foobar"
+        let mut output = [0_u8, 1, 2, 3, 4, 5, 6, 7];
+
+        decode_chunk_8(&input[..], 0, &STANDARD.decode_table, &mut output).unwrap();
+        assert_eq!(&vec![b'f', b'o', b'o', b'b', b'a', b'r', 6, 7], &output);
+    }
+
+    #[test]
+    fn decode_chunk_4_writes_only_3_bytes() {
+        let input = b"Zm9v"; // "foobar"
+        let mut output = [0_u8, 1, 2, 3];
+
+        decode_chunk_4(&input[..], 0, &STANDARD.decode_table, &mut output).unwrap();
+        assert_eq!(&vec![b'f', b'o', b'o', 3], &output);
+    }
+
+    #[test]
+    fn estimate_short_lengths() {
+        for (range, decoded_len_estimate) in [
+            (0..=0, 0),
+            (1..=4, 3),
+            (5..=8, 6),
+            (9..=12, 9),
+            (13..=16, 12),
+            (17..=20, 15),
+        ] {
+            for encoded_len in range {
+                let estimate = GeneralPurposeEstimate::new(encoded_len);
+                assert_eq!(decoded_len_estimate, estimate.decoded_len_estimate());
+            }
+        }
+    }
+
+    #[test]
+    fn estimate_via_u128_inflation() {
+        // cover both ends of usize
+        (0..1000)
+            .chain(usize::MAX - 1000..=usize::MAX)
+            .for_each(|encoded_len| {
+                // inflate to 128 bit type to be able to safely use the easy formulas
+                let len_128 = encoded_len as u128;
+
+                let estimate = GeneralPurposeEstimate::new(encoded_len);
+                assert_eq!(
+                    (len_128 + 3) / 4 * 3,
+                    estimate.conservative_decoded_len as u128
+                );
+            })
+    }
+}
diff --git a/vendor/base64/src/engine/general_purpose/decode_suffix.rs b/vendor/base64/src/engine/general_purpose/decode_suffix.rs
new file mode 100644
index 00000000..02aaf514
--- /dev/null
+++ b/vendor/base64/src/engine/general_purpose/decode_suffix.rs
@@ -0,0 +1,162 @@
+use crate::{
+    engine::{general_purpose::INVALID_VALUE, DecodeMetadata, DecodePaddingMode},
+    DecodeError, DecodeSliceError, PAD_BYTE,
+};
+
+/// Decode the last 0-4 bytes, checking for trailing set bits and padding per the provided
+/// parameters.
+///
+/// Returns the decode metadata representing the total number of bytes decoded, including the ones
+/// indicated as already written by `output_index`.
+pub(crate) fn decode_suffix(
+    input: &[u8],
+    input_index: usize,
+    output: &mut [u8],
+    mut output_index: usize,
+    decode_table: &[u8; 256],
+    decode_allow_trailing_bits: bool,
+    padding_mode: DecodePaddingMode,
+) -> Result<DecodeMetadata, DecodeSliceError> {
+    debug_assert!((input.len() - input_index) <= 4);
+
+    // Decode any leftovers that might not be a complete input chunk of 4 bytes.
+    // Use a u32 as a stack-resident 4 byte buffer.
+    let mut morsels_in_leftover = 0;
+    let mut padding_bytes_count = 0;
+    // offset from input_index
+    let mut first_padding_offset: usize = 0;
+    let mut last_symbol = 0_u8;
+    let mut morsels = [0_u8; 4];
+
+    for (leftover_index, &b) in input[input_index..].iter().enumerate() {
+        // '=' padding
+        if b == PAD_BYTE {
+            // There can be bad padding bytes in a few ways:
+            // 1 - Padding with non-padding characters after it
+            // 2 - Padding after zero or one characters in the current quad (should only
+            //     be after 2 or 3 chars)
+            // 3 - More than two characters of padding. If 3 or 4 padding chars
+            //     are in the same quad, that implies it will be caught by #2.
+            //     If it spreads from one quad to another, it will be an invalid byte
+            //     in the first quad.
+            // 4 - Non-canonical padding -- 1 byte when it should be 2, etc.
+            //     Per config, non-canonical but still functional non- or partially-padded base64
+            //     may be treated as an error condition.
+
+            if leftover_index < 2 {
+                // Check for error #2.
+                // Either the previous byte was padding, in which case we would have already hit
+                // this case, or it wasn't, in which case this is the first such error.
+                debug_assert!(
+                    leftover_index == 0 || (leftover_index == 1 && padding_bytes_count == 0)
+                );
+                let bad_padding_index = input_index + leftover_index;
+                return Err(DecodeError::InvalidByte(bad_padding_index, b).into());
+            }
+
+            if padding_bytes_count == 0 {
+                first_padding_offset = leftover_index;
+            }
+
+            padding_bytes_count += 1;
+            continue;
+        }
+
+        // Check for case #1.
+        // To make '=' handling consistent with the main loop, don't allow
+        // non-suffix '=' in trailing chunk either. Report error as first
+        // erroneous padding.
+        if padding_bytes_count > 0 {
+            return Err(
+                DecodeError::InvalidByte(input_index + first_padding_offset, PAD_BYTE).into(),
+            );
+        }
+
+        last_symbol = b;
+
+        // can use up to 8 * 6 = 48 bits of the u64, if last chunk has no padding.
+        // Pack the leftovers from left to right.
+        let morsel = decode_table[b as usize];
+        if morsel == INVALID_VALUE {
+            return Err(DecodeError::InvalidByte(input_index + leftover_index, b).into());
+        }
+
+        morsels[morsels_in_leftover] = morsel;
+        morsels_in_leftover += 1;
+    }
+
+    // If there was 1 trailing byte, and it was valid, and we got to this point without hitting
+    // an invalid byte, now we can report invalid length
+    if !input.is_empty() && morsels_in_leftover < 2 {
+        return Err(DecodeError::InvalidLength(input_index + morsels_in_leftover).into());
+    }
+
+    match padding_mode {
+        DecodePaddingMode::Indifferent => { /* everything we care about was already checked */ }
+        DecodePaddingMode::RequireCanonical => {
+            // allow empty input
+            if (padding_bytes_count + morsels_in_leftover) % 4 != 0 {
+                return Err(DecodeError::InvalidPadding.into());
+            }
+        }
+        DecodePaddingMode::RequireNone => {
+            if padding_bytes_count > 0 {
+                // check at the end to make sure we let the cases of padding that should be InvalidByte
+                // get hit
+                return Err(DecodeError::InvalidPadding.into());
+            }
+        }
+    }
+
+    // When encoding 1 trailing byte (e.g. 0xFF), 2 base64 bytes ("/w") are needed.
+    // / is the symbol for 63 (0x3F, bottom 6 bits all set) and w is 48 (0x30, top 2 bits
+    // of bottom 6 bits set).
+    // When decoding two symbols back to one trailing byte, any final symbol higher than
+    // w would still decode to the original byte because we only care about the top two
+    // bits in the bottom 6, but would be a non-canonical encoding. So, we calculate a
+    // mask based on how many bits are used for just the canonical encoding, and optionally
+    // error if any other bits are set. In the example of one encoded byte -> 2 symbols,
+    // 2 symbols can technically encode 12 bits, but the last 4 are non-canonical, and
+    // useless since there are no more symbols to provide the necessary 4 additional bits
+    // to finish the second original byte.
+
+    let leftover_bytes_to_append = morsels_in_leftover * 6 / 8;
+    // Put the up to 6 complete bytes as the high bytes.
+    // Gain a couple percent speedup from nudging these ORs to use more ILP with a two-way split.
+    let mut leftover_num = (u32::from(morsels[0]) << 26)
+        | (u32::from(morsels[1]) << 20)
+        | (u32::from(morsels[2]) << 14)
+        | (u32::from(morsels[3]) << 8);
+
+    // if there are bits set outside the bits we care about, last symbol encodes trailing bits that
+    // will not be included in the output
+    let mask = !0_u32 >> (leftover_bytes_to_append * 8);
+    if !decode_allow_trailing_bits && (leftover_num & mask) != 0 {
+        // last morsel is at `morsels_in_leftover` - 1
+        return Err(DecodeError::InvalidLastSymbol(
+            input_index + morsels_in_leftover - 1,
+            last_symbol,
+        )
+        .into());
+    }
+
+    // Strangely, this approach benchmarks better than writing bytes one at a time,
+    // or copy_from_slice into output.
+    for _ in 0..leftover_bytes_to_append {
+        let hi_byte = (leftover_num >> 24) as u8;
+        leftover_num <<= 8;
+        *output
+            .get_mut(output_index)
+            .ok_or(DecodeSliceError::OutputSliceTooSmall)? = hi_byte;
+        output_index += 1;
+    }
+
+    Ok(DecodeMetadata::new(
+        output_index,
+        if padding_bytes_count > 0 {
+            Some(input_index + first_padding_offset)
+        } else {
+            None
+        },
+    ))
+}
diff --git a/vendor/base64/src/engine/general_purpose/mod.rs b/vendor/base64/src/engine/general_purpose/mod.rs
new file mode 100644
index 00000000..6fe95809
--- /dev/null
+++ b/vendor/base64/src/engine/general_purpose/mod.rs
@@ -0,0 +1,352 @@
+//! Provides the [GeneralPurpose] engine and associated config types.
+use crate::{
+    alphabet,
+    alphabet::Alphabet,
+    engine::{Config, DecodeMetadata, DecodePaddingMode},
+    DecodeSliceError,
+};
+use core::convert::TryInto;
+
+pub(crate) mod decode;
+pub(crate) mod decode_suffix;
+
+pub use decode::GeneralPurposeEstimate;
+
+pub(crate) const INVALID_VALUE: u8 = 255;
+
+/// A general-purpose base64 engine.
+///
+/// - It uses no vector CPU instructions, so it will work on any system.
+/// - It is reasonably fast (~2-3GiB/s).
+/// - It is not constant-time, though, so it is vulnerable to timing side-channel attacks. For loading cryptographic keys, etc, it is suggested to use the forthcoming constant-time implementation.
+
+#[derive(Debug, Clone)]
+pub struct GeneralPurpose {
+    encode_table: [u8; 64],
+    decode_table: [u8; 256],
+    config: GeneralPurposeConfig,
+}
+
+impl GeneralPurpose {
+    /// Create a `GeneralPurpose` engine from an [Alphabet].
+    ///
+    /// While not very expensive to initialize, ideally these should be cached
+    /// if the engine will be used repeatedly.
+    pub const fn new(alphabet: &Alphabet, config: GeneralPurposeConfig) -> Self {
+        Self {
+            encode_table: encode_table(alphabet),
+            decode_table: decode_table(alphabet),
+            config,
+        }
+    }
+}
+
+impl super::Engine for GeneralPurpose {
+    type Config = GeneralPurposeConfig;
+    type DecodeEstimate = GeneralPurposeEstimate;
+
+    fn internal_encode(&self, input: &[u8], output: &mut [u8]) -> usize {
+        let mut input_index: usize = 0;
+
+        const BLOCKS_PER_FAST_LOOP: usize = 4;
+        const LOW_SIX_BITS: u64 = 0x3F;
+
+        // we read 8 bytes at a time (u64) but only actually consume 6 of those bytes. Thus, we need
+        // 2 trailing bytes to be available to read..
+        let last_fast_index = input.len().saturating_sub(BLOCKS_PER_FAST_LOOP * 6 + 2);
+        let mut output_index = 0;
+
+        if last_fast_index > 0 {
+            while input_index <= last_fast_index {
+                // Major performance wins from letting the optimizer do the bounds check once, mostly
+                // on the output side
+                let input_chunk =
+                    &input[input_index..(input_index + (BLOCKS_PER_FAST_LOOP * 6 + 2))];
+                let output_chunk =
+                    &mut output[output_index..(output_index + BLOCKS_PER_FAST_LOOP * 8)];
+
+                // Hand-unrolling for 32 vs 16 or 8 bytes produces yields performance about equivalent
+                // to unsafe pointer code on a Xeon E5-1650v3. 64 byte unrolling was slightly better for
+                // large inputs but significantly worse for 50-byte input, unsurprisingly. I suspect
+                // that it's a not uncommon use case to encode smallish chunks of data (e.g. a 64-byte
+                // SHA-512 digest), so it would be nice if that fit in the unrolled loop at least once.
+                // Plus, single-digit percentage performance differences might well be quite different
+                // on different hardware.
+
+                let input_u64 = read_u64(&input_chunk[0..]);
+
+                output_chunk[0] = self.encode_table[((input_u64 >> 58) & LOW_SIX_BITS) as usize];
+                output_chunk[1] = self.encode_table[((input_u64 >> 52) & LOW_SIX_BITS) as usize];
+                output_chunk[2] = self.encode_table[((input_u64 >> 46) & LOW_SIX_BITS) as usize];
+                output_chunk[3] = self.encode_table[((input_u64 >> 40) & LOW_SIX_BITS) as usize];
+                output_chunk[4] = self.encode_table[((input_u64 >> 34) & LOW_SIX_BITS) as usize];
+                output_chunk[5] = self.encode_table[((input_u64 >> 28) & LOW_SIX_BITS) as usize];
+                output_chunk[6] = self.encode_table[((input_u64 >> 22) & LOW_SIX_BITS) as usize];
+                output_chunk[7] = self.encode_table[((input_u64 >> 16) & LOW_SIX_BITS) as usize];
+
+                let input_u64 = read_u64(&input_chunk[6..]);
+
+                output_chunk[8] = self.encode_table[((input_u64 >> 58) & LOW_SIX_BITS) as usize];
+                output_chunk[9] = self.encode_table[((input_u64 >> 52) & LOW_SIX_BITS) as usize];
+                output_chunk[10] = self.encode_table[((input_u64 >> 46) & LOW_SIX_BITS) as usize];
+                output_chunk[11] = self.encode_table[((input_u64 >> 40) & LOW_SIX_BITS) as usize];
+                output_chunk[12] = self.encode_table[((input_u64 >> 34) & LOW_SIX_BITS) as usize];
+                output_chunk[13] = self.encode_table[((input_u64 >> 28) & LOW_SIX_BITS) as usize];
+                output_chunk[14] = self.encode_table[((input_u64 >> 22) & LOW_SIX_BITS) as usize];
+                output_chunk[15] = self.encode_table[((input_u64 >> 16) & LOW_SIX_BITS) as usize];
+
+                let input_u64 = read_u64(&input_chunk[12..]);
+
+                output_chunk[16] = self.encode_table[((input_u64 >> 58) & LOW_SIX_BITS) as usize];
+                output_chunk[17] = self.encode_table[((input_u64 >> 52) & LOW_SIX_BITS) as usize];
+                output_chunk[18] = self.encode_table[((input_u64 >> 46) & LOW_SIX_BITS) as usize];
+                output_chunk[19] = self.encode_table[((input_u64 >> 40) & LOW_SIX_BITS) as usize];
+                output_chunk[20] = self.encode_table[((input_u64 >> 34) & LOW_SIX_BITS) as usize];
+                output_chunk[21] = self.encode_table[((input_u64 >> 28) & LOW_SIX_BITS) as usize];
+                output_chunk[22] = self.encode_table[((input_u64 >> 22) & LOW_SIX_BITS) as usize];
+                output_chunk[23] = self.encode_table[((input_u64 >> 16) & LOW_SIX_BITS) as usize];
+
+                let input_u64 = read_u64(&input_chunk[18..]);
+
+                output_chunk[24] = self.encode_table[((input_u64 >> 58) & LOW_SIX_BITS) as usize];
+                output_chunk[25] = self.encode_table[((input_u64 >> 52) & LOW_SIX_BITS) as usize];
+                output_chunk[26] = self.encode_table[((input_u64 >> 46) & LOW_SIX_BITS) as usize];
+                output_chunk[27] = self.encode_table[((input_u64 >> 40) & LOW_SIX_BITS) as usize];
+                output_chunk[28] = self.encode_table[((input_u64 >> 34) & LOW_SIX_BITS) as usize];
+                output_chunk[29] = self.encode_table[((input_u64 >> 28) & LOW_SIX_BITS) as usize];
+                output_chunk[30] = self.encode_table[((input_u64 >> 22) & LOW_SIX_BITS) as usize];
+                output_chunk[31] = self.encode_table[((input_u64 >> 16) & LOW_SIX_BITS) as usize];
+
+                output_index += BLOCKS_PER_FAST_LOOP * 8;
+                input_index += BLOCKS_PER_FAST_LOOP * 6;
+            }
+        }
+
+        // Encode what's left after the fast loop.
+
+        const LOW_SIX_BITS_U8: u8 = 0x3F;
+
+        let rem = input.len() % 3;
+        let start_of_rem = input.len() - rem;
+
+        // start at the first index not handled by fast loop, which may be 0.
+
+        while input_index < start_of_rem {
+            let input_chunk = &input[input_index..(input_index + 3)];
+            let output_chunk = &mut output[output_index..(output_index + 4)];
+
+            output_chunk[0] = self.encode_table[(input_chunk[0] >> 2) as usize];
+            output_chunk[1] = self.encode_table
+                [((input_chunk[0] << 4 | input_chunk[1] >> 4) & LOW_SIX_BITS_U8) as usize];
+            output_chunk[2] = self.encode_table
+                [((input_chunk[1] << 2 | input_chunk[2] >> 6) & LOW_SIX_BITS_U8) as usize];
+            output_chunk[3] = self.encode_table[(input_chunk[2] & LOW_SIX_BITS_U8) as usize];
+
+            input_index += 3;
+            output_index += 4;
+        }
+
+        if rem == 2 {
+            output[output_index] = self.encode_table[(input[start_of_rem] >> 2) as usize];
+            output[output_index + 1] =
+                self.encode_table[((input[start_of_rem] << 4 | input[start_of_rem + 1] >> 4)
+                    & LOW_SIX_BITS_U8) as usize];
+            output[output_index + 2] =
+                self.encode_table[((input[start_of_rem + 1] << 2) & LOW_SIX_BITS_U8) as usize];
+            output_index += 3;
+        } else if rem == 1 {
+            output[output_index] = self.encode_table[(input[start_of_rem] >> 2) as usize];
+            output[output_index + 1] =
+                self.encode_table[((input[start_of_rem] << 4) & LOW_SIX_BITS_U8) as usize];
+            output_index += 2;
+        }
+
+        output_index
+    }
+
+    fn internal_decoded_len_estimate(&self, input_len: usize) -> Self::DecodeEstimate {
+        GeneralPurposeEstimate::new(input_len)
+    }
+
+    fn internal_decode(
+        &self,
+        input: &[u8],
+        output: &mut [u8],
+        estimate: Self::DecodeEstimate,
+    ) -> Result<DecodeMetadata, DecodeSliceError> {
+        decode::decode_helper(
+            input,
+            estimate,
+            output,
+            &self.decode_table,
+            self.config.decode_allow_trailing_bits,
+            self.config.decode_padding_mode,
+        )
+    }
+
+    fn config(&self) -> &Self::Config {
+        &self.config
+    }
+}
+
+/// Returns a table mapping a 6-bit index to the ASCII byte encoding of the index
+pub(crate) const fn encode_table(alphabet: &Alphabet) -> [u8; 64] {
+    // the encode table is just the alphabet:
+    // 6-bit index lookup -> printable byte
+    let mut encode_table = [0_u8; 64];
+    {
+        let mut index = 0;
+        while index < 64 {
+            encode_table[index] = alphabet.symbols[index];
+            index += 1;
+        }
+    }
+
+    encode_table
+}
+
+/// Returns a table mapping base64 bytes as the lookup index to either:
+/// - [INVALID_VALUE] for bytes that aren't members of the alphabet
+/// - a byte whose lower 6 bits are the value that was encoded into the index byte
+pub(crate) const fn decode_table(alphabet: &Alphabet) -> [u8; 256] {
+    let mut decode_table = [INVALID_VALUE; 256];
+
+    // Since the table is full of `INVALID_VALUE` already, we only need to overwrite
+    // the parts that are valid.
+    let mut index = 0;
+    while index < 64 {
+        // The index in the alphabet is the 6-bit value we care about.
+        // Since the index is in 0-63, it is safe to cast to u8.
+        decode_table[alphabet.symbols[index] as usize] = index as u8;
+        index += 1;
+    }
+
+    decode_table
+}
+
+#[inline]
+fn read_u64(s: &[u8]) -> u64 {
+    u64::from_be_bytes(s[..8].try_into().unwrap())
+}
+
+/// Contains configuration parameters for base64 encoding and decoding.
+///
+/// ```
+/// # use base64::engine::GeneralPurposeConfig;
+/// let config = GeneralPurposeConfig::new()
+///     .with_encode_padding(false);
+///     // further customize using `.with_*` methods as needed
+/// ```
+///
+/// The constants [PAD] and [NO_PAD] cover most use cases.
+///
+/// To specify the characters used, see [Alphabet].
+#[derive(Clone, Copy, Debug)]
+pub struct GeneralPurposeConfig {
+    encode_padding: bool,
+    decode_allow_trailing_bits: bool,
+    decode_padding_mode: DecodePaddingMode,
+}
+
+impl GeneralPurposeConfig {
+    /// Create a new config with `padding` = `true`, `decode_allow_trailing_bits` = `false`, and
+    /// `decode_padding_mode = DecodePaddingMode::RequireCanonicalPadding`.
+    ///
+    /// This probably matches most people's expectations, but consider disabling padding to save
+    /// a few bytes unless you specifically need it for compatibility with some legacy system.
+    pub const fn new() -> Self {
+        Self {
+            // RFC states that padding must be applied by default
+            encode_padding: true,
+            decode_allow_trailing_bits: false,
+            decode_padding_mode: DecodePaddingMode::RequireCanonical,
+        }
+    }
+
+    /// Create a new config based on `self` with an updated `padding` setting.
+    ///
+    /// If `padding` is `true`, encoding will append either 1 or 2 `=` padding characters as needed
+    /// to produce an output whose length is a multiple of 4.
+    ///
+    /// Padding is not needed for correct decoding and only serves to waste bytes, but it's in the
+    /// [spec](https://datatracker.ietf.org/doc/html/rfc4648#section-3.2).
+    ///
+    /// For new applications, consider not using padding if the decoders you're using don't require
+    /// padding to be present.
+    pub const fn with_encode_padding(self, padding: bool) -> Self {
+        Self {
+            encode_padding: padding,
+            ..self
+        }
+    }
+
+    /// Create a new config based on `self` with an updated `decode_allow_trailing_bits` setting.
+    ///
+    /// Most users will not need to configure this. It's useful if you need to decode base64
+    /// produced by a buggy encoder that has bits set in the unused space on the last base64
+    /// character as per [forgiving-base64 decode](https://infra.spec.whatwg.org/#forgiving-base64-decode).
+    /// If invalid trailing bits are present and this is `true`, those bits will
+    /// be silently ignored, else `DecodeError::InvalidLastSymbol` will be emitted.
+    pub const fn with_decode_allow_trailing_bits(self, allow: bool) -> Self {
+        Self {
+            decode_allow_trailing_bits: allow,
+            ..self
+        }
+    }
+
+    /// Create a new config based on `self` with an updated `decode_padding_mode` setting.
+    ///
+    /// Padding is not useful in terms of representing encoded data -- it makes no difference to
+    /// the decoder if padding is present or not, so if you have some un-padded input to decode, it
+    /// is perfectly fine to use `DecodePaddingMode::Indifferent` to prevent errors from being
+    /// emitted.
+    ///
+    /// However, since in practice
+    /// [people who learned nothing from BER vs DER seem to expect base64 to have one canonical encoding](https://eprint.iacr.org/2022/361),
+    /// the default setting is the stricter `DecodePaddingMode::RequireCanonicalPadding`.
+    ///
+    /// Or, if "canonical" in your circumstance means _no_ padding rather than padding to the
+    /// next multiple of four, there's `DecodePaddingMode::RequireNoPadding`.
+    pub const fn with_decode_padding_mode(self, mode: DecodePaddingMode) -> Self {
+        Self {
+            decode_padding_mode: mode,
+            ..self
+        }
+    }
+}
+
+impl Default for GeneralPurposeConfig {
+    /// Delegates to [GeneralPurposeConfig::new].
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+impl Config for GeneralPurposeConfig {
+    fn encode_padding(&self) -> bool {
+        self.encode_padding
+    }
+}
+
+/// A [GeneralPurpose] engine using the [alphabet::STANDARD] base64 alphabet and [PAD] config.
+pub const STANDARD: GeneralPurpose = GeneralPurpose::new(&alphabet::STANDARD, PAD);
+
+/// A [GeneralPurpose] engine using the [alphabet::STANDARD] base64 alphabet and [NO_PAD] config.
+pub const STANDARD_NO_PAD: GeneralPurpose = GeneralPurpose::new(&alphabet::STANDARD, NO_PAD);
+
+/// A [GeneralPurpose] engine using the [alphabet::URL_SAFE] base64 alphabet and [PAD] config.
+pub const URL_SAFE: GeneralPurpose = GeneralPurpose::new(&alphabet::URL_SAFE, PAD);
+
+/// A [GeneralPurpose] engine using the [alphabet::URL_SAFE] base64 alphabet and [NO_PAD] config.
+pub const URL_SAFE_NO_PAD: GeneralPurpose = GeneralPurpose::new(&alphabet::URL_SAFE, NO_PAD);
+
+/// Include padding bytes when encoding, and require that they be present when decoding.
+///
+/// This is the standard per the base64 RFC, but consider using [NO_PAD] instead as padding serves
+/// little purpose in practice.
+pub const PAD: GeneralPurposeConfig = GeneralPurposeConfig::new();
+
+/// Don't add padding when encoding, and require no padding when decoding.
+pub const NO_PAD: GeneralPurposeConfig = GeneralPurposeConfig::new()
+    .with_encode_padding(false)
+    .with_decode_padding_mode(DecodePaddingMode::RequireNone);