Delete everything, I'm redesigning this.

I'll stop using a flat package and make things much more modular. And also experiment with streaming APIs so large blobs don't OOM us.
author: Runxi Yu 2026-02-20 19:06:13 +0800
committer: Runxi Yu 2026-02-20 19:07:14 +0800
commit: aa513c069c1418734aea894dc944e27c6a78a3bb (patch)
tree: 687f0a11bb550fa088fd82a98ceb8979bbc35f69 /internal
parent: Comment on prior reverts removing the pack writing API (diff)
signature: No signature
31 files changed, 0 insertions, 3207 deletions
diff --git a/internal/adler32/LICENSE b/internal/adler32/LICENSE
deleted file mode 100644
index 5cec357a..00000000
--- a/internal/adler32/LICENSE
+++ /dev/null
@@ -1,30 +0,0 @@
-Copyright (c) 2024, Michal Hruby
-Copyright (c) 2017 The Chromium Authors. All rights reserved.
-Copyright (c) 1995-2024 Mark Adler
-Copyright (c) 1995-2024 Jean-loup Gailly
-Copyright (c) 2022 Adam Stylinski
-
-BSD 2-Clause License
-
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are met:
-
-1. Redistributions of source code must retain the above copyright notice, this
-   list of conditions and the following disclaimer.
-
-2. Redistributions in binary form must reproduce the above copyright notice,
-   this list of conditions and the following disclaimer in the documentation
-   and/or other materials provided with the distribution.
-
-THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
-AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
-IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
-DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
-FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
-DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
-SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
-CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
-OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
diff --git a/internal/adler32/LICENSE.ZLIB b/internal/adler32/LICENSE.ZLIB
deleted file mode 100644
index c75c1568..00000000
--- a/internal/adler32/LICENSE.ZLIB
+++ /dev/null
@@ -1,17 +0,0 @@
-Copyright (C) 1995-2024 Jean-loup Gailly and Mark Adler
-
-This software is provided 'as-is', without any express or implied
-warranty.  In no event will the authors be held liable for any damages
-arising from the use of this software.
-
-Permission is granted to anyone to use this software for any purpose,
-including commercial applications, and to alter it and redistribute it
-freely, subject to the following restrictions:
-
-1. The origin of this software must not be misrepresented; you must not
-   claim that you wrote the original software. If you use this software
-   in a product, an acknowledgment in the product documentation would be
-   appreciated but is not required.
-2. Altered source versions must be plainly marked as such, and must not be
-   misrepresented as being the original software.
-3. This notice may not be removed or altered from any source distribution.
diff --git a/internal/adler32/README b/internal/adler32/README
deleted file mode 100644
index b80acd00..00000000
--- a/internal/adler32/README
+++ /dev/null
@@ -1 +0,0 @@
-This package was mostly copied from github.com/mhr3/adler32-simd.
diff --git a/internal/adler32/adler32_amd64.go b/internal/adler32/adler32_amd64.go
deleted file mode 100644
index 88a854ed..00000000
--- a/internal/adler32/adler32_amd64.go
+++ /dev/null
@@ -1,93 +0,0 @@
-//go:build amd64 && !purego
-
-package adler32
-
-import (
-	"encoding/binary"
-	"errors"
-	"hash"
-	"hash/adler32"
-
-	"golang.org/x/sys/cpu"
-)
-
-// The size of an Adler-32 checksum in bytes.
-const Size = 4
-
-var (
-	hasSSE3 = cpu.X86.HasSSE3
-	hasAVX2 = cpu.X86.HasAVX2
-)
-
-// digest represents the partial evaluation of a checksum.
-// The low 16 bits are s1, the high 16 bits are s2.
-type digest uint32
-
-func (d *digest) Reset() { *d = 1 }
-
-// New returns a new hash.Hash32 computing the Adler-32 checksum.
-func New() hash.Hash32 {
-	if !hasSSE3 {
-		return adler32.New()
-	}
-	d := new(digest)
-	d.Reset()
-	return d
-}
-
-func (d *digest) MarshalBinary() ([]byte, error) {
-	b := make([]byte, 0, marshaledSize)
-	b = append(b, magic...)
-	b = binary.BigEndian.AppendUint32(b, uint32(*d))
-	return b, nil
-}
-
-func (d *digest) UnmarshalBinary(b []byte) error {
-	if len(b) < len(magic) || string(b[:len(magic)]) != magic {
-		return errors.New("hash/adler32: invalid hash state identifier")
-	}
-	if len(b) != marshaledSize {
-		return errors.New("hash/adler32: invalid hash state size")
-	}
-	*d = digest(binary.BigEndian.Uint32(b[len(magic):]))
-	return nil
-}
-
-func (d *digest) Size() int { return Size }
-
-func (d *digest) BlockSize() int { return 4 }
-
-func (d *digest) Write(data []byte) (nn int, err error) {
-	if len(data) >= 64 {
-		var h uint32
-		if hasAVX2 {
-			h = adler32_avx2(uint32(*d), data)
-		} else {
-			h = adler32_sse3(uint32(*d), data)
-		}
-		*d = digest(h)
-	} else {
-		h := update(uint32(*d), data)
-		*d = digest(h)
-	}
-	return len(data), nil
-}
-
-func (d *digest) Sum32() uint32 { return uint32(*d) }
-
-func (d *digest) Sum(in []byte) []byte {
-	s := uint32(*d)
-	return append(in, byte(s>>24), byte(s>>16), byte(s>>8), byte(s))
-}
-
-// Checksum returns the Adler-32 checksum of data.
-func Checksum(data []byte) uint32 {
-	if !hasSSE3 || len(data) < 64 {
-		return update(1, data)
-	}
-
-	if hasAVX2 {
-		return adler32_avx2(1, data)
-	}
-	return adler32_sse3(1, data)
-}
diff --git a/internal/adler32/adler32_arm64.go b/internal/adler32/adler32_arm64.go
deleted file mode 100644
index ddf9cb5e..00000000
--- a/internal/adler32/adler32_arm64.go
+++ /dev/null
@@ -1,73 +0,0 @@
-//go:build arm64 && !purego
-
-package adler32
-
-import (
-	"encoding/binary"
-	"errors"
-	"hash"
-)
-
-// The size of an Adler-32 checksum in bytes.
-const Size = 4
-
-// digest represents the partial evaluation of a checksum.
-// The low 16 bits are s1, the high 16 bits are s2.
-type digest uint32
-
-func (d *digest) Reset() { *d = 1 }
-
-// New returns a new hash.Hash32 computing the Adler-32 checksum.
-func New() hash.Hash32 {
-	d := new(digest)
-	d.Reset()
-	return d
-}
-
-func (d *digest) MarshalBinary() ([]byte, error) {
-	b := make([]byte, 0, marshaledSize)
-	b = append(b, magic...)
-	b = binary.BigEndian.AppendUint32(b, uint32(*d))
-	return b, nil
-}
-
-func (d *digest) UnmarshalBinary(b []byte) error {
-	if len(b) < len(magic) || string(b[:len(magic)]) != magic {
-		return errors.New("hash/adler32: invalid hash state identifier")
-	}
-	if len(b) != marshaledSize {
-		return errors.New("hash/adler32: invalid hash state size")
-	}
-	*d = digest(binary.BigEndian.Uint32(b[len(magic):]))
-	return nil
-}
-
-func (d *digest) Size() int { return Size }
-
-func (d *digest) BlockSize() int { return 4 }
-
-func (d *digest) Write(data []byte) (nn int, err error) {
-	if len(data) >= 64 {
-		h := adler32_neon(uint32(*d), data)
-		*d = digest(h)
-	} else {
-		h := update(uint32(*d), data)
-		*d = digest(h)
-	}
-	return len(data), nil
-}
-
-func (d *digest) Sum32() uint32 { return uint32(*d) }
-
-func (d *digest) Sum(in []byte) []byte {
-	s := uint32(*d)
-	return append(in, byte(s>>24), byte(s>>16), byte(s>>8), byte(s))
-}
-
-// Checksum returns the Adler-32 checksum of data.
-func Checksum(data []byte) uint32 {
-	if len(data) >= 64 {
-		return adler32_neon(1, data)
-	}
-	return update(1, data)
-}
diff --git a/internal/adler32/adler32_avx2.go b/internal/adler32/adler32_avx2.go
deleted file mode 100644
index 042812b8..00000000
--- a/internal/adler32/adler32_avx2.go
+++ /dev/null
@@ -1,6 +0,0 @@
-//go:build !purego && amd64
-
-package adler32
-
-//go:noescape
-func adler32_avx2(in uint32, buf []byte) uint32
diff --git a/internal/adler32/adler32_avx2.s b/internal/adler32/adler32_avx2.s
deleted file mode 100644
index 1b9a1c50..00000000
--- a/internal/adler32/adler32_avx2.s
+++ /dev/null
@@ -1,263 +0,0 @@
-//go:build !purego && amd64
-
-#include "textflag.h"
-
-DATA weights_1_32<>+0x00(SB)/8, $0x191a1b1c1d1e1f20
-DATA weights_1_32<>+0x08(SB)/8, $0x1112131415161718
-DATA weights_1_32<>+0x10(SB)/8, $0x090a0b0c0d0e0f10
-DATA weights_1_32<>+0x18(SB)/8, $0x0102030405060708
-GLOBL weights_1_32<>(SB), (RODATA|NOPTR), $32
-
-DATA ones_u16<>+0x00(SB)/8, $0x0001000100010001
-DATA ones_u16<>+0x08(SB)/8, $0x0001000100010001
-DATA ones_u16<>+0x10(SB)/8, $0x0001000100010001
-DATA ones_u16<>+0x18(SB)/8, $0x0001000100010001
-GLOBL ones_u16<>(SB), (RODATA|NOPTR), $32
-
-DATA one_u16<>+0x00(SB)/2, $0x0001
-GLOBL one_u16<>(SB), (RODATA|NOPTR), $2
-
-TEXT ·adler32_avx2(SB), NOSPLIT, $0-36
-	MOVLQZX      in+0(FP), DI
-	MOVQ         buf_base+8(FP), SI
-	MOVQ         buf_len+16(FP), DX
-	MOVQ         buf_cap+24(FP), CX
-	WORD         $0x8548; BYTE $0xf6
-	JE           return_one
-	WORD         $0xf889
-	WORD         $0x8548; BYTE $0xd2
-	JE           return_result
-	NOP
-	NOP
-	NOP
-	WORD         $0xc189
-	WORD         $0xe9c1; BYTE $0x10
-	WORD         $0xb70f; BYTE $0xc0
-	CMPQ         DX, $0x20
-	JB           tail16_check
-	LONG         $0x078071bf; BYTE $0x80
-	LONG         $0xc0eff9c5
-	VMOVDQA      weights_1_32<>(SB), Y1
-	VPBROADCASTW one_u16<>(SB), Y2
-	JMP          block_loop_setup
-
-block_accum_init:
-	LONG $0xf46ffdc5
-	LONG $0xedefd1c5
-
-block_reduce:
-	SUBQ  AX, DX
-	LONG  $0xf572ddc5; BYTE $0x05
-	LONG  $0xdbfeddc5
-	LONG  $0x397de3c4; WORD $0x01f4
-	LONG  $0xecc6c8c5; BYTE $0x88
-	LONG  $0xe470f9c5; BYTE $0x88
-	LONG  $0xe4fed1c5
-	LONG  $0xec70f9c5; BYTE $0x55
-	LONG  $0xe4fed1c5
-	LONG  $0xe07ef9c5
-	MOVQ  AX, CX
-	IMULQ DI, CX
-	SHRQ  $0x2f, CX
-	LONG  $0xfff1c969; WORD $0x0000
-	WORD  $0xc829
-	LONG  $0x397de3c4; WORD $0x01dc
-	LONG  $0xdbfed9c5
-	LONG  $0xe370f9c5; BYTE $0xee
-	LONG  $0xdcfee1c5
-	LONG  $0xe370f9c5; BYTE $0x55
-	LONG  $0xdbfed9c5
-	LONG  $0xd97ef9c5
-	MOVQ  CX, R8
-	IMULQ DI, R8
-	SHRQ  $0x2f, R8
-	LONG  $0xf1c06945; WORD $0x00ff; BYTE $0x00
-	WORD  $0x2944; BYTE $0xc1
-	CMPQ  DX, $0x1f
-	JBE   tail_check
-
-block_loop_setup:
-	LONG $0xe06ef9c5
-	LONG $0xd96ef9c5
-	CMPQ DX, $0x15b0
-	LONG $0x15b0b841; WORD $0x0000
-	LONG $0xc2420f4c
-	WORD $0x8944; BYTE $0xc0
-	LONG $0x001fe025; BYTE $0x00
-	JE   block_accum_init
-	ADDQ $-0x20, R8
-	LONG $0xedefd1c5
-	LONG $0x20c0f641
-	JNE  block_loop_entry
-	LONG $0x2e6ffec5
-	ADDQ $0x20, SI
-	LEAQ -0x20(AX), CX
-	LONG $0xf0f6d5c5
-	LONG $0xf4fecdc5
-	LONG $0x0455e2c4; BYTE $0xe9
-	LONG $0xeaf5d5c5
-	LONG $0xdbfed5c5
-	LONG $0xec6ffdc5
-	LONG $0xe66ffdc5
-	CMPQ R8, $0x20
-	JAE  block_loop_64
-	JMP  block_reduce
-
-block_loop_entry:
-	MOVQ AX, CX
-	CMPQ R8, $0x20
-	JB   block_reduce
-
-block_loop_64:
-	LONG $0x366ffec5
-	LONG $0x7e6ffec5; BYTE $0x20
-	LONG $0xc0f64dc5
-	LONG $0xc4fe3dc5
-	LONG $0xecfed5c5
-	LONG $0x044de2c4; BYTE $0xe1
-	LONG $0xe2f5ddc5
-	LONG $0xdbfeddc5
-	ADDQ $0x40, SI
-	LONG $0xe0f6c5c5
-	LONG $0xe4febdc5
-	LONG $0xedfebdc5
-	LONG $0x0445e2c4; BYTE $0xf1
-	LONG $0xf2f5cdc5
-	LONG $0xdbfecdc5
-	ADDQ $-0x40, CX
-	JNE  block_loop_64
-	LONG $0xf46ffdc5
-	JMP  block_reduce
-
-return_one:
-	LONG $0x000001b8; BYTE $0x00
-
-return_result:
-	MOVL AX, ret+32(FP)
-	RET
-
-tail_check:
-	WORD $0x8548; BYTE $0xd2
-	JE   return_no_tail
-
-tail16_check:
-	CMPQ DX, $0x10
-	JB   tail_bytes_setup
-	WORD $0xb60f; BYTE $0x3e
-	WORD $0xf801
-	WORD $0xc101
-	LONG $0x017eb60f
-	WORD $0xc701
-	WORD $0xf901
-	LONG $0x0246b60f
-	WORD $0xf801
-	WORD $0xc101
-	LONG $0x037eb60f
-	WORD $0xc701
-	WORD $0xf901
-	LONG $0x0446b60f
-	WORD $0xf801
-	WORD $0xc101
-	LONG $0x057eb60f
-	WORD $0xc701
-	WORD $0xf901
-	LONG $0x0646b60f
-	WORD $0xf801
-	WORD $0xc101
-	LONG $0x077eb60f
-	WORD $0xc701
-	WORD $0xf901
-	LONG $0x0846b60f
-	WORD $0xf801
-	WORD $0xc101
-	LONG $0x097eb60f
-	WORD $0xc701
-	WORD $0xf901
-	LONG $0x0a46b60f
-	WORD $0xf801
-	WORD $0xc101
-	LONG $0x0b7eb60f
-	WORD $0xc701
-	WORD $0xf901
-	LONG $0x0c46b60f
-	WORD $0xf801
-	WORD $0xc101
-	LONG $0x0d7eb60f
-	WORD $0xc701
-	WORD $0xf901
-	LONG $0x46b60f44; BYTE $0x0e
-	WORD $0x0141; BYTE $0xf8
-	WORD $0x0144; BYTE $0xc1
-	LONG $0x0f46b60f
-	WORD $0x0144; BYTE $0xc0
-	WORD $0xc101
-	ADDQ $-0x10, DX
-	JE   final_reduce
-	ADDQ $0x10, SI
-
-tail_bytes_setup:
-	LEAQ -0x1(DX), DI
-	MOVQ DX, R9
-	ANDQ $0x3, R9
-	JE   tail_dword_setup
-	XORL R8, R8
-
-tail_byte_loop:
-	LONG $0x14b60f46; BYTE $0x06
-	WORD $0x0144; BYTE $0xd0
-	WORD $0xc101
-	INCQ R8
-	CMPQ R9, R8
-	JNE  tail_byte_loop
-	ADDQ R8, SI
-	SUBQ R8, DX
-
-tail_dword_setup:
-	CMPQ DI, $0x3
-	JB   final_reduce
-	XORL DI, DI
-
-tail_dword_loop:
-	LONG $0x04b60f44; BYTE $0x3e
-	WORD $0x0141; BYTE $0xc0
-	WORD $0x0144; BYTE $0xc1
-	LONG $0x3e44b60f; BYTE $0x01
-	WORD $0x0144; BYTE $0xc0
-	WORD $0xc101
-	LONG $0x44b60f44; WORD $0x023e
-	WORD $0x0141; BYTE $0xc0
-	WORD $0x0144; BYTE $0xc1
-	LONG $0x3e44b60f; BYTE $0x03
-	WORD $0x0144; BYTE $0xc0
-	WORD $0xc101
-	ADDQ $0x4, DI
-	CMPQ DX, DI
-	JNE  tail_dword_loop
-
-final_reduce:
-	LONG  $0x000f908d; WORD $0xffff
-	CMPL  AX, $0xfff1
-	WORD  $0x420f; BYTE $0xd0
-	WORD  $0xc889
-	LONG  $0x078071be; BYTE $0x80
-	IMULQ AX, SI
-	SHRQ  $0x2f, SI
-	LONG  $0xfff1c669; WORD $0x0000
-	WORD  $0xc129
-	WORD  $0xe1c1; BYTE $0x10
-	WORD  $0xd109
-	WORD  $0xc889
-	NOP
-	NOP
-	VZEROUPPER
-	MOVL  AX, ret+32(FP)
-	RET
-
-return_no_tail:
-	WORD $0xe1c1; BYTE $0x10
-	WORD $0xc809
-	NOP
-	NOP
-	VZEROUPPER
-	MOVL AX, ret+32(FP)
-	RET
diff --git a/internal/adler32/adler32_fallback.go b/internal/adler32/adler32_fallback.go
deleted file mode 100644
index c213c3c1..00000000
--- a/internal/adler32/adler32_fallback.go
+++ /dev/null
@@ -1,19 +0,0 @@
-//go:build (!arm64 && !amd64) || purego
-
-package adler32
-
-import (
-	"hash"
-	"hash/adler32"
-)
-
-// The size of an Adler-32 checksum in bytes.
-const Size = 4
-
-// New returns a new hash.Hash32 computing the Adler-32 checksum.
-func New() hash.Hash32 {
-	return adler32.New()
-}
-
-// Checksum returns the Adler-32 checksum of data.
-func Checksum(data []byte) uint32 { return adler32.Checksum(data) }
diff --git a/internal/adler32/adler32_generic.go b/internal/adler32/adler32_generic.go
deleted file mode 100644
index f33e0f9b..00000000
--- a/internal/adler32/adler32_generic.go
+++ /dev/null
@@ -1,45 +0,0 @@
-// Package adler32 implements the Adler-32 checksum.
-package adler32
-
-const (
-	// mod is the largest prime that is less than 65536.
-	mod = 65521
-	// nmax is the largest n such that
-	// 255 * n * (n+1) / 2 + (n+1) * (mod-1) <= 2^32-1.
-	// It is mentioned in RFC 1950 (search for "5552").
-	nmax = 5552
-
-	// binary representation compatible with standard library.
-	magic         = "adl\x01"
-	marshaledSize = len(magic) + 4
-)
-
-// Add p to the running checksum d.
-func update(d uint32, p []byte) uint32 {
-	s1, s2 := d&0xffff, d>>16
-	for len(p) > 0 {
-		var q []byte
-		if len(p) > nmax {
-			p, q = p[:nmax], p[nmax:]
-		}
-		for len(p) >= 4 {
-			s1 += uint32(p[0])
-			s2 += s1
-			s1 += uint32(p[1])
-			s2 += s1
-			s1 += uint32(p[2])
-			s2 += s1
-			s1 += uint32(p[3])
-			s2 += s1
-			p = p[4:]
-		}
-		for _, x := range p {
-			s1 += uint32(x)
-			s2 += s1
-		}
-		s1 %= mod
-		s2 %= mod
-		p = q
-	}
-	return s2<<16 | s1
-}
diff --git a/internal/adler32/adler32_neon.go b/internal/adler32/adler32_neon.go
deleted file mode 100644
index 521b71e0..00000000
--- a/internal/adler32/adler32_neon.go
+++ /dev/null
@@ -1,6 +0,0 @@
-//go:build !purego && arm64
-
-package adler32
-
-//go:noescape
-func adler32_neon(in uint32, buf []byte) uint32
diff --git a/internal/adler32/adler32_neon.s b/internal/adler32/adler32_neon.s
deleted file mode 100644
index 08b170bd..00000000
--- a/internal/adler32/adler32_neon.s
+++ /dev/null
@@ -1,208 +0,0 @@
-//go:build !purego && arm64
-
-#include "textflag.h"
-
-DATA mult_table<>+0x00(SB)/8, $0x001d001e001f0020
-DATA mult_table<>+0x08(SB)/8, $0x0019001a001b001c
-DATA mult_table<>+0x10(SB)/8, $0x0015001600170018
-DATA mult_table<>+0x18(SB)/8, $0x0011001200130014
-DATA mult_table<>+0x20(SB)/8, $0x000d000e000f0010
-DATA mult_table<>+0x28(SB)/8, $0x0009000a000b000c
-DATA mult_table<>+0x30(SB)/8, $0x0005000600070008
-DATA mult_table<>+0x38(SB)/8, $0x0001000200030004
-GLOBL mult_table<>(SB), (RODATA|NOPTR), $64
-
-TEXT ·adler32_neon(SB), NOSPLIT, $0-36
-	MOVW in+0(FP), R0
-	MOVD buf_base+8(FP), R1
-	MOVD buf_len+16(FP), R2
-	MOVD buf_cap+24(FP), R3
-	NOP
-	ANDS $15, R1, R10
-	ANDW $65535, R0, R8
-	LSRW $16, R0, R9
-	NOP
-	BEQ  vector_loop_setup
-	ADD  $1, R1, R11
-	MOVD R1, R12
-
-align_loop:
-	WORD  $0x3840158d
-	SUB   $1, R2, R2
-	TST   $15, R11
-	ADD   $1, R11, R11
-	ADDW  R13, R8, R8
-	ADDW  R9, R8, R9
-	BNE   align_loop
-	MOVW  $32881, R11
-	MOVW  $65521, R13
-	MOVKW $(32775<<16), R11
-	MOVW  $4294901775, R12
-	MOVW  $65520, R14
-	SUB   R10, R1, R10
-	UMULL R11, R9, R11
-	ADDW  R12, R8, R12
-	CMPW  R14, R8
-	ADD   $16, R10, R1
-	LSR   $47, R11, R11
-	CSELW HI, R12, R8, R8
-	MSUBW R13, R9, R11, R9
-
-vector_loop_setup:
-	AND   $31, R2, R10
-	CMP   $32, R2
-	BCC   tail_entry
-	MOVD  $mult_table<>(SB), R11
-	ADD   $0, R11, R11
-	MOVW  $32881, R14
-	MOVW  $173, R12
-	MOVD  $137438953440, R13
-	MOVKW $(32775<<16), R14
-	VLD1  (R11), [V0.H8, V1.H8, V2.H8, V3.H8]
-	LSR   $5, R2, R11
-	MOVW  $65521, R15
-	VEXT  $8, V0.B16, V0.B16, V4.B16
-	VEXT  $8, V1.B16, V1.B16, V5.B16
-	VEXT  $8, V2.B16, V2.B16, V6.B16
-	VEXT  $8, V3.B16, V3.B16, V7.B16
-
-vector_outer_loop:
-	CMP  $173, R11
-	MOVD R1, R2
-	CSEL LO, R11, R12, R16
-	WORD $0x6f00e414
-	MULW R16, R8, R0
-	ADD  R16<<5, R13, R17
-	WORD $0x6f00e410
-	AND  $137438953440, R17, R17
-	WORD $0x6f00e412
-	WORD $0x6f00e413
-	WORD $0x6f00e415
-	VMOV R0, V20.S[3]
-	MOVW R16, R0
-	WORD $0x6f00e411
-
-vector_inner_loop:
-	WORD  $0xacc15857
-	SUBSW $1, R0, R0
-	VADD  V17.S4, V20.S4, V20.S4
-	WORD  $0x2e3712b5
-	WORD  $0x6e371273
-	WORD  $0x6e202ad8
-	WORD  $0x2e361252
-	WORD  $0x6e361210
-	WORD  $0x6e206af8
-	WORD  $0x6e606b11
-	BNE   vector_inner_loop
-	VSHL  $5, V20.S4, V20.S4
-	ADD   R17, R1, R17
-	SUBS  R16, R11, R11
-	ADD   $32, R17, R1
-	WORD  $0x2e6082b4
-	VEXT  $8, V21.B16, V21.B16, V21.B16
-	WORD  $0x2e6482b4
-	VEXT  $8, V19.B16, V19.B16, V21.B16
-	WORD  $0x2e618274
-	VEXT  $8, V18.B16, V18.B16, V19.B16
-	WORD  $0x2e6582b4
-	WORD  $0x2e628254
-	WORD  $0x2e668274
-	WORD  $0x2e638214
-	VEXT  $8, V16.B16, V16.B16, V16.B16
-	WORD  $0x2e678214
-	WORD  $0x4eb1be30
-	WORD  $0x4eb4be91
-	WORD  $0x0eb1be10
-	VMOV  V16.S[1], R0
-	FMOVS F16, R2
-	ADDW  R8, R2, R8
-	ADDW  R9, R0, R9
-	UMULL R14, R8, R0
-	UMULL R14, R9, R2
-	LSR   $47, R0, R0
-	LSR   $47, R2, R2
-	MSUBW R15, R8, R0, R8
-	MSUBW R15, R9, R2, R9
-	BNE   vector_outer_loop
-
-tail_entry:
-	CBZ  R10, return_result
-	CMP  $16, R10
-	BCC  tail_byte_loop
-	WORD $0x3940002b
-	SUBS $16, R10, R10
-	WORD $0x3940042c
-	WORD $0x3940082d
-	ADDW R11, R8, R8
-	WORD $0x39400c2b
-	ADDW R9, R8, R9
-	ADDW R12, R8, R8
-	WORD $0x3940102c
-	ADDW R8, R9, R9
-	ADDW R13, R8, R8
-	WORD $0x3940142d
-	ADDW R8, R9, R9
-	ADDW R11, R8, R8
-	WORD $0x3940182b
-	ADDW R8, R9, R9
-	ADDW R12, R8, R8
-	WORD $0x39401c2c
-	ADDW R8, R9, R9
-	ADDW R13, R8, R8
-	ADDW R8, R9, R9
-	ADDW R11, R8, R8
-	WORD $0x3940202b
-	ADDW R8, R9, R9
-	ADDW R12, R8, R8
-	WORD $0x3940242c
-	ADDW R8, R9, R9
-	WORD $0x3940382d
-	ADDW R11, R8, R8
-	WORD $0x3940282b
-	ADDW R8, R9, R9
-	ADDW R12, R8, R8
-	WORD $0x39402c2c
-	ADDW R8, R9, R9
-	ADDW R11, R8, R8
-	WORD $0x3940302b
-	ADDW R8, R9, R9
-	ADDW R12, R8, R8
-	WORD $0x3940342c
-	ADDW R8, R9, R9
-	ADDW R11, R8, R8
-	WORD $0x39403c2b
-	ADDW R8, R9, R9
-	ADDW R12, R8, R8
-	ADDW R8, R9, R9
-	ADDW R13, R8, R8
-	ADDW R8, R9, R9
-	ADDW R11, R8, R8
-	ADDW R8, R9, R9
-	BEQ  final_reduce
-	ADD  $16, R1, R1
-
-tail_byte_loop:
-	WORD $0x3840142b
-	SUBS $1, R10, R10
-	ADDW R11, R8, R8
-	ADDW R9, R8, R9
-	BNE  tail_byte_loop
-
-final_reduce:
-	MOVW  $32881, R10
-	MOVW  $65521, R12
-	MOVKW $(32775<<16), R10
-	MOVW  $4294901775, R11
-	MOVW  $65520, R13
-	ADDW  R11, R8, R11
-	UMULL R10, R9, R10
-	CMPW  R13, R8
-	CSELW HI, R11, R8, R8
-	LSR   $47, R10, R10
-	MSUBW R12, R9, R10, R9
-
-return_result:
-	ORRW R9<<16, R8, R0
-	NOP
-	MOVW R0, ret+32(FP)
-	RET
diff --git a/internal/adler32/adler32_sse3.go b/internal/adler32/adler32_sse3.go
deleted file mode 100644
index 8e8c8a9b..00000000
--- a/internal/adler32/adler32_sse3.go
+++ /dev/null
@@ -1,6 +0,0 @@
-//go:build !purego && amd64
-
-package adler32
-
-//go:noescape
-func adler32_sse3(in uint32, buf []byte) uint32
diff --git a/internal/adler32/adler32_sse3.s b/internal/adler32/adler32_sse3.s
deleted file mode 100644
index 5880bab8..00000000
--- a/internal/adler32/adler32_sse3.s
+++ /dev/null
@@ -1,214 +0,0 @@
-//go:build !purego && amd64
-
-#include "textflag.h"
-
-DATA weights_17_32<>+0x00(SB)/8, $0x191a1b1c1d1e1f20
-DATA weights_17_32<>+0x08(SB)/8, $0x1112131415161718
-GLOBL weights_17_32<>(SB), (RODATA|NOPTR), $16
-
-DATA ones_u16<>+0x00(SB)/8, $0x0001000100010001
-DATA ones_u16<>+0x08(SB)/8, $0x0001000100010001
-GLOBL ones_u16<>(SB), (RODATA|NOPTR), $16
-
-DATA weights_1_16<>+0x00(SB)/8, $0x090a0b0c0d0e0f10
-DATA weights_1_16<>+0x08(SB)/8, $0x0102030405060708
-GLOBL weights_1_16<>(SB), (RODATA|NOPTR), $16
-
-TEXT ·adler32_sse3(SB), NOSPLIT, $0-36
-	MOVLQZX in+0(FP), DI
-	MOVQ    buf_base+8(FP), SI
-	MOVQ    buf_len+16(FP), DX
-	MOVQ    buf_cap+24(FP), CX
-	NOP
-	NOP
-	NOP
-	WORD    $0xf889
-	LONG    $0xc8b70f44
-	WORD    $0xe8c1; BYTE $0x10
-	WORD    $0xd189
-	WORD    $0xe183; BYTE $0x1f
-	CMPQ    DX, $0x20
-	JAE     block_loop_setup
-	WORD    $0x8944; BYTE $0xcf
-	JMP     tail_entry
-
-block_loop_setup:
-	SHRQ $0x5, DX
-	LONG $0xc0ef0f66
-	MOVO weights_17_32<>(SB), X1
-	MOVO ones_u16<>(SB), X2
-	MOVO weights_1_16<>(SB), X3
-	LONG $0x8071b841; WORD $0x8007
-
-block_outer_loop:
-	CMPQ DX, $0xad
-	LONG $0x00adba41; WORD $0x0000
-	LONG $0xd2420f4c
-	WORD $0x8944; BYTE $0xcf
-	LONG $0xfaaf0f41
-	LONG $0xef6e0f66
-	LONG $0xe06e0f66
-	WORD $0x8944; BYTE $0xd0
-	LONG $0xf6ef0f66
-
-block_inner_loop:
-	LONG  $0x3e6f0ff3
-	LONG  $0x6f0f4466; BYTE $0xc7
-	LONG  $0x04380f66; BYTE $0xf9
-	LONG  $0xfaf50f66
-	LONG  $0xfcfe0f66
-	LONG  $0x666f0ff3; BYTE $0x10
-	LONG  $0xeefe0f66
-	LONG  $0xf60f4466; BYTE $0xc0
-	LONG  $0xfe0f4466; BYTE $0xc6
-	LONG  $0xf46f0f66
-	LONG  $0xf0f60f66
-	LONG  $0xfe0f4166; BYTE $0xf0
-	LONG  $0x04380f66; BYTE $0xe3
-	LONG  $0xe2f50f66
-	LONG  $0xe7fe0f66
-	ADDQ  $0x20, SI
-	WORD  $0xc8ff
-	JNE   block_inner_loop
-	LONG  $0xf5720f66; BYTE $0x05
-	LONG  $0xe5fe0f66
-	LONG  $0xee700f66; BYTE $0xb1
-	LONG  $0xeefe0f66
-	LONG  $0xf5700f66; BYTE $0xee
-	LONG  $0xf5fe0f66
-	LONG  $0xf77e0f66
-	WORD  $0x0144; BYTE $0xcf
-	LONG  $0xec700f66; BYTE $0xb1
-	LONG  $0xecfe0f66
-	LONG  $0xe5700f66; BYTE $0xee
-	LONG  $0xe5fe0f66
-	LONG  $0xe07e0f66
-	MOVQ  DI, R9
-	IMULQ R8, R9
-	SHRQ  $0x2f, R9
-	LONG  $0xf1c96945; WORD $0x00ff; BYTE $0x00
-	WORD  $0x2944; BYTE $0xcf
-	MOVQ  AX, R9
-	IMULQ R8, R9
-	SHRQ  $0x2f, R9
-	LONG  $0xf1c96945; WORD $0x00ff; BYTE $0x00
-	WORD  $0x2944; BYTE $0xc8
-	WORD  $0x8941; BYTE $0xf9
-	SUBQ  R10, DX
-	JNE   block_outer_loop
-
-tail_entry:
-	WORD $0x8548; BYTE $0xc9
-	JE   return_result
-	CMPL CX, $0x10
-	JB   tail_bytes_setup
-	WORD $0xb60f; BYTE $0x16
-	WORD $0xd701
-	WORD $0xf801
-	LONG $0x0156b60f
-	WORD $0xfa01
-	WORD $0xd001
-	LONG $0x027eb60f
-	WORD $0xd701
-	WORD $0xf801
-	LONG $0x0356b60f
-	WORD $0xfa01
-	WORD $0xd001
-	LONG $0x047eb60f
-	WORD $0xd701
-	WORD $0xf801
-	LONG $0x0556b60f
-	WORD $0xfa01
-	WORD $0xd001
-	LONG $0x067eb60f
-	WORD $0xd701
-	WORD $0xf801
-	LONG $0x0756b60f
-	WORD $0xfa01
-	WORD $0xd001
-	LONG $0x087eb60f
-	WORD $0xd701
-	WORD $0xf801
-	LONG $0x0956b60f
-	WORD $0xfa01
-	WORD $0xd001
-	LONG $0x0a7eb60f
-	WORD $0xd701
-	WORD $0xf801
-	LONG $0x0b56b60f
-	WORD $0xfa01
-	WORD $0xd001
-	LONG $0x0c7eb60f
-	WORD $0xd701
-	WORD $0xf801
-	LONG $0x0d56b60f
-	WORD $0xfa01
-	WORD $0xd001
-	LONG $0x46b60f44; BYTE $0x0e
-	WORD $0x0141; BYTE $0xd0
-	WORD $0x0144; BYTE $0xc0
-	LONG $0x0f7eb60f
-	WORD $0x0144; BYTE $0xc7
-	WORD $0xf801
-	ADDQ $-0x10, CX
-	JE   final_reduce
-	ADDQ $0x10, SI
-
-tail_bytes_setup:
-	LEAQ -0x1(CX), DX
-	MOVQ CX, R9
-	ANDQ $0x3, R9
-	JE   tail_dword_setup
-	XORL R8, R8
-
-tail_byte_loop:
-	LONG $0x14b60f46; BYTE $0x06
-	WORD $0x0144; BYTE $0xd7
-	WORD $0xf801
-	INCQ R8
-	CMPQ R9, R8
-	JNE  tail_byte_loop
-	ADDQ R8, SI
-	SUBQ R8, CX
-
-tail_dword_setup:
-	CMPQ DX, $0x3
-	JB   final_reduce
-	XORL DX, DX
-
-tail_dword_loop:
-	LONG $0x04b60f44; BYTE $0x16
-	WORD $0x0141; BYTE $0xf8
-	WORD $0x0144; BYTE $0xc0
-	LONG $0x167cb60f; BYTE $0x01
-	WORD $0x0144; BYTE $0xc7
-	WORD $0xf801
-	LONG $0x44b60f44; WORD $0x0216
-	WORD $0x0141; BYTE $0xf8
-	WORD $0x0144; BYTE $0xc0
-	LONG $0x167cb60f; BYTE $0x03
-	WORD $0x0144; BYTE $0xc7
-	WORD $0xf801
-	ADDQ $0x4, DX
-	CMPQ CX, DX
-	JNE  tail_dword_loop
-
-final_reduce:
-	LONG  $0x000f8f8d; WORD $0xffff
-	CMPL  DI, $0xfff1
-	WORD  $0x420f; BYTE $0xcf
-	WORD  $0xc289
-	LONG  $0x078071be; BYTE $0x80
-	IMULQ DX, SI
-	SHRQ  $0x2f, SI
-	LONG  $0xfff1d669; WORD $0x0000
-	WORD  $0xd029
-	WORD  $0xcf89
-
-return_result:
-	WORD $0xe0c1; BYTE $0x10
-	WORD $0xf809
-	NOP
-	NOP
-	MOVL AX, ret+32(FP)
-	RET
diff --git a/internal/adler32/bench_test.go b/internal/adler32/bench_test.go
deleted file mode 100644
index 7744b903..00000000
--- a/internal/adler32/bench_test.go
+++ /dev/null
@@ -1,22 +0,0 @@
-package adler32
-
-import (
-	"testing"
-)
-
-const benchmarkSize = 64 * 1024
-
-var data = make([]byte, benchmarkSize)
-
-func init() {
-	for i := range benchmarkSize {
-		data[i] = byte(i % 256)
-	}
-}
-
-func BenchmarkChecksum(b *testing.B) {
-	b.ReportAllocs()
-	for range b.N {
-		Checksum(data)
-	}
-}
diff --git a/internal/bloom/bloom.go b/internal/bloom/bloom.go
deleted file mode 100644
index c1a116d6..00000000
--- a/internal/bloom/bloom.go
+++ /dev/null
@@ -1,236 +0,0 @@
-// Package bloom provides a bloom filter implementation used for changed-path
-// filters in Git commit graphs.
-package bloom
-
-import "encoding/binary"
-
-const (
-	// DataHeaderSize is the size of the BDAT header in commit-graph files.
-	DataHeaderSize = 3 * 4
-	// DefaultMaxChange matches Git's default max-changed-paths behavior.
-	DefaultMaxChange = 512
-)
-
-// Settings describe the changed-paths Bloom filter parameters stored in
-// commit-graph BDAT chunks.
-//
-// Obviously, they must match the repository's commit-graph settings to
-// interpret filters correctly.
-type Settings struct {
-	HashVersion    uint32
-	NumHashes      uint32
-	BitsPerEntry   uint32
-	MaxChangePaths uint32
-}
-
-// Filter represents a changed-paths Bloom filter associated with a commit.
-//
-// The filter encodes which paths changed between a commit and its first
-// parent. Paths are expected to be in Git's slash-separated form and
-// are queried using a path and its prefixes (e.g. "a/b/c", "a/b", "a").
-type Filter struct {
-	Data    []byte
-	Version uint32
-}
-
-// ParseSettings reads Bloom filter settings from a BDAT chunk header.
-func ParseSettings(bdat []byte) (*Settings, error) {
-	if len(bdat) < DataHeaderSize {
-		return nil, ErrInvalid
-	}
-	settings := &Settings{
-		HashVersion:    binary.BigEndian.Uint32(bdat[0:4]),
-		NumHashes:      binary.BigEndian.Uint32(bdat[4:8]),
-		BitsPerEntry:   binary.BigEndian.Uint32(bdat[8:12]),
-		MaxChangePaths: DefaultMaxChange,
-	}
-	return settings, nil
-}
-
-// MightContain reports whether the Bloom filter may contain the given path.
-//
-// Evaluated against the full path and each of its directory prefixes. A true
-// result indicates a possible match; false means the path definitely did not
-// change.
-func (f *Filter) MightContain(path []byte, settings *Settings) bool {
-	if f == nil || settings == nil {
-		return false
-	}
-	if len(f.Data) == 0 {
-		return false
-	}
-	keys := keyvec(path, settings)
-	for i := range keys {
-		if filterContainsKey(f, &keys[i], settings) {
-			return true
-		}
-	}
-	return false
-}
-
-type key struct {
-	hashes []uint32
-}
-
-func keyvec(path []byte, settings *Settings) []key {
-	if len(path) == 0 {
-		return nil
-	}
-	count := 1
-	for _, b := range path {
-		if b == '/' {
-			count++
-		}
-	}
-	keys := make([]key, 0, count)
-	keys = append(keys, keyFill(path, settings))
-	for i := len(path) - 1; i >= 0; i-- {
-		if path[i] == '/' {
-			keys = append(keys, keyFill(path[:i], settings))
-		}
-	}
-	return keys
-}
-
-func keyFill(path []byte, settings *Settings) key {
-	const seed0 = 0x293ae76f
-	const seed1 = 0x7e646e2c
-	var h0, h1 uint32
-	if settings.HashVersion == 2 {
-		h0 = murmur3SeededV2(seed0, path)
-		h1 = murmur3SeededV2(seed1, path)
-	} else {
-		h0 = murmur3SeededV1(seed0, path)
-		h1 = murmur3SeededV1(seed1, path)
-	}
-	hashes := make([]uint32, settings.NumHashes)
-	for i := uint32(0); i < settings.NumHashes; i++ {
-		hashes[i] = h0 + i*h1
-	}
-	return key{hashes: hashes}
-}
-
-func filterContainsKey(filter *Filter, key *key, settings *Settings) bool {
-	if filter == nil || key == nil || settings == nil {
-		return false
-	}
-	if len(filter.Data) == 0 {
-		return false
-	}
-	mod := uint64(len(filter.Data)) * 8
-	for _, h := range key.hashes {
-		idx := uint64(h) % mod
-		bytePos := idx / 8
-		bit := byte(1 << (idx & 7))
-		if filter.Data[bytePos]&bit == 0 {
-			return false
-		}
-	}
-	return true
-}
-
-func murmur3SeededV2(seed uint32, data []byte) uint32 {
-	const (
-		c1 = 0xcc9e2d51
-		c2 = 0x1b873593
-		r1 = 15
-		r2 = 13
-		m  = 5
-		n  = 0xe6546b64
-	)
-	h := seed
-	nblocks := len(data) / 4
-	for i := 0; i < nblocks; i++ {
-		k := uint32(data[4*i]) |
-			(uint32(data[4*i+1]) << 8) |
-			(uint32(data[4*i+2]) << 16) |
-			(uint32(data[4*i+3]) << 24)
-		k *= c1
-		k = (k << r1) | (k >> (32 - r1))
-		k *= c2
-
-		h ^= k
-		h = (h << r2) | (h >> (32 - r2))
-		h = h*m + n
-	}
-
-	var k1 uint32
-	tail := data[nblocks*4:]
-	switch len(tail) & 3 {
-	case 3:
-		k1 ^= uint32(tail[2]) << 16
-		fallthrough
-	case 2:
-		k1 ^= uint32(tail[1]) << 8
-		fallthrough
-	case 1:
-		k1 ^= uint32(tail[0])
-		k1 *= c1
-		k1 = (k1 << r1) | (k1 >> (32 - r1))
-		k1 *= c2
-		h ^= k1
-	}
-
-	h ^= uint32(len(data))
-	h ^= h >> 16
-	h *= 0x85ebca6b
-	h ^= h >> 13
-	h *= 0xc2b2ae35
-	h ^= h >> 16
-	return h
-}
-
-func murmur3SeededV1(seed uint32, data []byte) uint32 {
-	const (
-		c1 = 0xcc9e2d51
-		c2 = 0x1b873593
-		r1 = 15
-		r2 = 13
-		m  = 5
-		n  = 0xe6546b64
-	)
-	h := seed
-	nblocks := len(data) / 4
-	for i := 0; i < nblocks; i++ {
-		b0 := int8(data[4*i])
-		b1 := int8(data[4*i+1])
-		b2 := int8(data[4*i+2])
-		b3 := int8(data[4*i+3])
-		k := uint32(b0) |
-			(uint32(b1) << 8) |
-			(uint32(b2) << 16) |
-			(uint32(b3) << 24)
-		k *= c1
-		k = (k << r1) | (k >> (32 - r1))
-		k *= c2
-
-		h ^= k
-		h = (h << r2) | (h >> (32 - r2))
-		h = h*m + n
-	}
-
-	var k1 uint32
-	tail := data[nblocks*4:]
-	switch len(tail) & 3 {
-	case 3:
-		k1 ^= uint32(int8(tail[2])) << 16
-		fallthrough
-	case 2:
-		k1 ^= uint32(int8(tail[1])) << 8
-		fallthrough
-	case 1:
-		k1 ^= uint32(int8(tail[0]))
-		k1 *= c1
-		k1 = (k1 << r1) | (k1 >> (32 - r1))
-		k1 *= c2
-		h ^= k1
-	}
-
-	h ^= uint32(len(data))
-	h ^= h >> 16
-	h *= 0x85ebca6b
-	h ^= h >> 13
-	h *= 0xc2b2ae35
-	h ^= h >> 16
-	return h
-}
diff --git a/internal/bloom/errors.go b/internal/bloom/errors.go
deleted file mode 100644
index fe38d1bc..00000000
--- a/internal/bloom/errors.go
+++ /dev/null
@@ -1,5 +0,0 @@
-package bloom
-
-import "errors"
-
-var ErrInvalid = errors.New("bloom: invalid data")
diff --git a/internal/bufpool/buffers.go b/internal/bufpool/buffers.go
deleted file mode 100644
index 439e7e04..00000000
--- a/internal/bufpool/buffers.go
+++ /dev/null
@@ -1,189 +0,0 @@
-// Package bufpool provides a lightweight byte-buffer type with optional
-// pooling.
-package bufpool
-
-import "sync"
-
-const (
-	// DefaultBufferCap is the minimum capacity a borrowed buffer will have.
-	// Borrow() will allocate or retrieve a buffer with at least this capacity.
-	DefaultBufferCap = 32 * 1024
-
-	// maxPooledBuffer defines the maximum capacity of a buffer that may be
-	// returned to the pool. Buffers larger than this will not be pooled to
-	// avoid unbounded memory usage.
-	maxPooledBuffer = 8 << 20
-)
-
-// Buffer is a growable byte container that optionally participates in a
-// memory pool. A Buffer may be obtained through Borrow() or constructed
-// directly from owned data via FromOwned().
-//
-// A Buffer's underlying slice may grow as needed. When finished with a
-// pooled buffer, the caller should invoke Release() to return it to the pool.
-//
-// Buffers must not be copied after first use; doing so can cause double-returns
-// to the pool and data races.
-//
-//go:nocopy
-type Buffer struct {
-	_    struct{} // for nocopy
-	buf  []byte
-	pool poolIndex
-}
-
-type poolIndex int8
-
-const (
-	unpooled poolIndex = -1
-)
-
-var sizeClasses = [...]int{
-	DefaultBufferCap,
-	64 << 10,
-	128 << 10,
-	256 << 10,
-	512 << 10,
-	1 << 20,
-	2 << 20,
-	4 << 20,
-	maxPooledBuffer,
-}
-
-var bufferPools = func() []sync.Pool {
-	pools := make([]sync.Pool, len(sizeClasses))
-	for i, classCap := range sizeClasses {
-		capCopy := classCap
-		pools[i].New = func() any {
-			buf := make([]byte, 0, capCopy)
-			return &buf
-		}
-	}
-	return pools
-}()
-
-// Borrow retrieves a Buffer suitable for storing up to capHint bytes.
-// The returned Buffer may come from an internal sync.Pool.
-//
-// If capHint is smaller than DefaultBufferCap, it is automatically raised
-// to DefaultBufferCap. If no pooled buffer has sufficient capacity, a new
-// unpooled buffer is allocated.
-//
-// The caller must call Release() when finished using the returned Buffer.
-func Borrow(capHint int) Buffer {
-	if capHint < DefaultBufferCap {
-		capHint = DefaultBufferCap
-	}
-	classIdx, classCap, pooled := classFor(capHint)
-	if !pooled {
-		newBuf := make([]byte, 0, capHint)
-		return Buffer{buf: newBuf, pool: unpooled}
-	}
-	buf := bufferPools[classIdx].Get().(*[]byte)
-	if cap(*buf) < classCap {
-		*buf = make([]byte, 0, classCap)
-	}
-	slice := (*buf)[:0]
-	return Buffer{buf: slice, pool: poolIndex(classIdx)}
-}
-
-// FromOwned constructs a Buffer from a caller-owned byte slice. The resulting
-// Buffer does not participate in pooling and will never be returned to the
-// internal pool when released.
-func FromOwned(buf []byte) Buffer {
-	return Buffer{buf: buf, pool: unpooled}
-}
-
-// Resize adjusts the length of the buffer to n bytes. If n exceeds the current
-// capacity, the underlying storage is grown. If n is negative, it is treated
-// as zero.
-//
-// The buffer's new contents beyond the previous length are undefined.
-func (buf *Buffer) Resize(n int) {
-	if n < 0 {
-		n = 0
-	}
-	buf.ensureCapacity(n)
-	buf.buf = buf.buf[:n]
-}
-
-// Append copies the provided bytes onto the end of the buffer, growing its
-// capacity if required. If src is empty, the method does nothing.
-//
-// The receiver retains ownership of the data; the caller may reuse src freely.
-func (buf *Buffer) Append(src []byte) {
-	if len(src) == 0 {
-		return
-	}
-	start := len(buf.buf)
-	buf.ensureCapacity(start + len(src))
-	buf.buf = buf.buf[:start+len(src)]
-	copy(buf.buf[start:], src)
-}
-
-// Bytes returns the underlying byte slice that represents the current contents
-// of the buffer. Modifying the returned slice modifies the Buffer itself.
-func (buf *Buffer) Bytes() []byte {
-	return buf.buf
-}
-
-// Release returns the buffer to the global pool if it originated from the
-// pool and its capacity is no larger than maxPooledBuffer. After release, the
-// Buffer becomes invalid and should not be used further.
-//
-// Releasing a non-pooled buffer has no effect beyond clearing its internal
-// storage.
-func (buf *Buffer) Release() {
-	if buf.buf == nil {
-		return
-	}
-	buf.returnToPool()
-	buf.buf = nil
-	buf.pool = unpooled
-}
-
-// ensureCapacity grows the underlying buffer to accommodate the requested
-// number of bytes. Growth doubles the capacity by default unless a larger
-// expansion is needed. If the previous storage was pooled and not oversized,
-// it is returned to the pool.
-func (buf *Buffer) ensureCapacity(needed int) {
-	if cap(buf.buf) >= needed {
-		return
-	}
-	classIdx, classCap, pooled := classFor(needed)
-	var newBuf []byte
-	if pooled {
-		raw := bufferPools[classIdx].Get().(*[]byte)
-		if cap(*raw) < classCap {
-			*raw = make([]byte, 0, classCap)
-		}
-		newBuf = (*raw)[:len(buf.buf)]
-	} else {
-		newBuf = make([]byte, len(buf.buf), classCap)
-	}
-	copy(newBuf, buf.buf)
-	buf.returnToPool()
-	buf.buf = newBuf
-	if pooled {
-		buf.pool = poolIndex(classIdx)
-	} else {
-		buf.pool = unpooled
-	}
-}
-
-func classFor(size int) (idx int, classCap int, ok bool) {
-	for i, class := range sizeClasses {
-		if size <= class {
-			return i, class, true
-		}
-	}
-	return -1, size, false
-}
-
-func (buf *Buffer) returnToPool() {
-	if buf.pool == unpooled {
-		return
-	}
-	tmp := buf.buf[:0]
-	bufferPools[int(buf.pool)].Put(&tmp)
-}
diff --git a/internal/bufpool/buffers_test.go b/internal/bufpool/buffers_test.go
deleted file mode 100644
index f5c006da..00000000
--- a/internal/bufpool/buffers_test.go
+++ /dev/null
@@ -1,77 +0,0 @@
-package bufpool
-
-import "testing"
-
-func TestBorrowBufferResizeAndAppend(t *testing.T) {
-	b := Borrow(1)
-	defer b.Release()
-
-	if cap(b.buf) < DefaultBufferCap {
-		t.Fatalf("expected capacity >= %d, got %d", DefaultBufferCap, cap(b.buf))
-	}
-
-	b.Append([]byte("alpha"))
-	b.Append([]byte("beta"))
-	if got := string(b.Bytes()); got != "alphabeta" {
-		t.Fatalf("unexpected contents: %q", got)
-	}
-
-	b.Resize(3)
-	if got := string(b.Bytes()); got != "alp" {
-		t.Fatalf("resize shrink mismatch: %q", got)
-	}
-
-	b.Resize(8)
-	if len(b.Bytes()) != 8 {
-		t.Fatalf("expected len 8 after grow, got %d", len(b.Bytes()))
-	}
-	if prefix := string(b.Bytes()[:3]); prefix != "alp" {
-		t.Fatalf("prefix lost after grow: %q", prefix)
-	}
-}
-
-func TestBorrowBufferRelease(t *testing.T) {
-	b := Borrow(DefaultBufferCap / 2)
-	b.Append([]byte("data"))
-	b.Release()
-	if b.buf != nil {
-		t.Fatal("expected buffer cleared after release")
-	}
-}
-
-func TestBorrowUsesLargerPools(t *testing.T) {
-	const request = DefaultBufferCap * 4
-
-	classIdx, classCap, pooled := classFor(request)
-	if !pooled {
-		t.Fatalf("expected %d to map to a pooled class", request)
-	}
-
-	b := Borrow(request)
-	if b.pool != poolIndex(classIdx) {
-		t.Fatalf("expected pooled buffer in class %d, got %d", classIdx, b.pool)
-	}
-	if cap(b.buf) != classCap {
-		t.Fatalf("expected capacity %d, got %d", classCap, cap(b.buf))
-	}
-	b.Release()
-
-	b2 := Borrow(request)
-	defer b2.Release()
-	if b2.pool != poolIndex(classIdx) {
-		t.Fatalf("expected pooled buffer in class %d on reuse, got %d", classIdx, b2.pool)
-	}
-	if cap(b2.buf) != classCap {
-		t.Fatalf("expected capacity %d on reuse, got %d", classCap, cap(b2.buf))
-	}
-}
-
-func TestGrowingBufferStaysPooled(t *testing.T) {
-	b := Borrow(DefaultBufferCap)
-	defer b.Release()
-
-	b.Append(make([]byte, DefaultBufferCap*3))
-	if b.pool == unpooled {
-		t.Fatal("buffer should stay pooled after growth within limit")
-	}
-}
diff --git a/internal/flatex/LICENSE b/internal/flatex/LICENSE
deleted file mode 100644
index 2a7cf70d..00000000
--- a/internal/flatex/LICENSE
+++ /dev/null
@@ -1,27 +0,0 @@
-Copyright 2009 The Go Authors.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are
-met:
-
-   * Redistributions of source code must retain the above copyright
-notice, this list of conditions and the following disclaimer.
-   * Redistributions in binary form must reproduce the above
-copyright notice, this list of conditions and the following disclaimer
-in the documentation and/or other materials provided with the
-distribution.
-   * Neither the name of Google LLC nor the names of its
-contributors may be used to endorse or promote products derived from
-this software without specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
diff --git a/internal/flatex/decompress.go b/internal/flatex/decompress.go
deleted file mode 100644
index 065e23f3..00000000
--- a/internal/flatex/decompress.go
+++ /dev/null
@@ -1,38 +0,0 @@
-package flatex
-
-import (
-	"io"
-
-	"codeberg.org/lindenii/furgit/internal/bufpool"
-)
-
-func DecompressSized(src []byte, sizeHint int) (bufpool.Buffer, int, error) {
-	d := sliceInflaterPool.Get().(*sliceInflater)
-	defer sliceInflaterPool.Put(d)
-
-	if err := d.reset(src); err != nil {
-		return bufpool.Buffer{}, 0, err
-	}
-
-	out := bufpool.Borrow(sizeHint)
-	out.Resize(0)
-
-	for {
-		if len(d.toRead) > 0 {
-			out.Append(d.toRead)
-			d.toRead = nil
-			continue
-		}
-		if d.err != nil {
-			if d.err == io.EOF {
-				return out, d.pos, nil
-			}
-			out.Release()
-			return bufpool.Buffer{}, 0, d.err
-		}
-		d.step(d)
-		if d.err != nil && len(d.toRead) == 0 {
-			d.toRead = d.window.readFlush()
-		}
-	}
-}
diff --git a/internal/flatex/decompress_test.go b/internal/flatex/decompress_test.go
deleted file mode 100644
index e53a6581..00000000
--- a/internal/flatex/decompress_test.go
+++ /dev/null
@@ -1,57 +0,0 @@
-package flatex
-
-import (
-	"bytes"
-	stdflate "compress/flate"
-	"testing"
-)
-
-func compressDeflate(t *testing.T, payload []byte) []byte {
-	t.Helper()
-	var buf bytes.Buffer
-	w, err := stdflate.NewWriter(&buf, stdflate.DefaultCompression)
-	if err != nil {
-		t.Fatalf("NewWriter: %v", err)
-	}
-	if _, err := w.Write(payload); err != nil {
-		t.Fatalf("Write: %v", err)
-	}
-	if err := w.Close(); err != nil {
-		t.Fatalf("Close: %v", err)
-	}
-	return buf.Bytes()
-}
-
-func TestDecompressSized(t *testing.T) {
-	payload := bytes.Repeat([]byte("golang"), 32)
-	compressed := compressDeflate(t, payload)
-
-	out, _, err := DecompressSized(compressed, 0)
-	if err != nil {
-		t.Fatalf("DecompressSized: %v", err)
-	}
-	defer out.Release()
-
-	if !bytes.Equal(out.Bytes(), payload) {
-		t.Fatalf("unexpected payload: got %q", out.Bytes())
-	}
-}
-
-func TestDecompressSizedUsesHint(t *testing.T) {
-	payload := []byte("short")
-	compressed := compressDeflate(t, payload)
-
-	const hint = 1 << 19
-	out, _, err := DecompressSized(compressed, hint)
-	if err != nil {
-		t.Fatalf("DecompressSized: %v", err)
-	}
-	defer out.Release()
-
-	if !bytes.Equal(out.Bytes(), payload) {
-		t.Fatalf("unexpected payload: got %q", out.Bytes())
-	}
-	if cap(out.Bytes()) < hint {
-		t.Fatalf("expected capacity >= %d, got %d", hint, cap(out.Bytes()))
-	}
-}
diff --git a/internal/flatex/huffman.go b/internal/flatex/huffman.go
deleted file mode 100644
index 32172dbb..00000000
--- a/internal/flatex/huffman.go
+++ /dev/null
@@ -1,245 +0,0 @@
-// Copyright 2009 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-// Package flatex implements the DEFLATE compressed data format, described in
-// RFC 1951.  The [compress/gzip] and [compress/zlib] packages implement access
-// to DEFLATE-based file formats.
-package flatex
-
-import (
-	"math/bits"
-	"strconv"
-	"sync"
-)
-
-const (
-	// The special code used to mark the end of a block.
-	endBlockMarker = 256
-	maxCodeLen     = 16      // max length of Huffman code
-	maxMatchOffset = 1 << 15 // The largest match offset
-	// The next three numbers come from the RFC section 3.2.7, with the
-	// additional proviso in section 3.2.5 which implies that distance codes
-	// 30 and 31 should never occur in compressed data.
-	maxNumLit  = 286
-	maxNumDist = 30
-	numCodes   = 19 // number of codes in Huffman meta-code
-)
-
-// A CorruptInputError reports the presence of corrupt input at a given offset.
-type CorruptInputError int64
-
-func (e CorruptInputError) Error() string {
-	return "flate: corrupt input before offset " + strconv.FormatInt(int64(e), 10)
-}
-
-// The data structure for decoding Huffman tables is based on that of
-// zlib. There is a lookup table of a fixed bit width (huffmanChunkBits),
-// For codes smaller than the table width, there are multiple entries
-// (each combination of trailing bits has the same value). For codes
-// larger than the table width, the table contains a link to an overflow
-// table. The width of each entry in the link table is the maximum code
-// size minus the chunk width.
-//
-// Note that you can do a lookup in the table even without all bits
-// filled. Since the extra bits are zero, and the DEFLATE Huffman codes
-// have the property that shorter codes come before longer ones, the
-// bit length estimate in the result is a lower bound on the actual
-// number of bits.
-//
-// See the following:
-//	https://github.com/madler/zlib/raw/master/doc/algorithm.txt
-
-// chunk & 15 is number of bits
-// chunk >> 4 is value, including table link
-
-const (
-	huffmanChunkBits  = 9
-	huffmanNumChunks  = 1 << huffmanChunkBits
-	huffmanCountMask  = 15
-	huffmanValueShift = 4
-)
-
-type huffmanDecoder struct {
-	min      int                      // the minimum code length
-	chunks   [huffmanNumChunks]uint32 // chunks as described above
-	links    [][]uint32               // overflow links
-	linkMask uint32                   // mask the width of the link table
-}
-
-// Initialize Huffman decoding tables from array of code lengths.
-// Following this function, h is guaranteed to be initialized into a complete
-// tree (i.e., neither over-subscribed nor under-subscribed). The exception is a
-// degenerate case where the tree has only a single symbol with length 1. Empty
-// trees are permitted.
-func (h *huffmanDecoder) init(lengths []int) bool {
-	// Sanity enables additional runtime tests during Huffman
-	// table construction. It's intended to be used during
-	// development to supplement the currently ad-hoc unit tests.
-	const sanity = false
-
-	if h.min != 0 {
-		*h = huffmanDecoder{}
-	}
-
-	// Count number of codes of each length,
-	// compute min and max length.
-	var count [maxCodeLen]int
-	var min, max int
-	for _, n := range lengths {
-		if n == 0 {
-			continue
-		}
-		if min == 0 || n < min {
-			min = n
-		}
-		if n > max {
-			max = n
-		}
-		count[n]++
-	}
-
-	// Empty tree. The decompressor.huffSym function will fail later if the tree
-	// is used. Technically, an empty tree is only valid for the HDIST tree and
-	// not the HCLEN and HLIT tree. However, a stream with an empty HCLEN tree
-	// is guaranteed to fail since it will attempt to use the tree to decode the
-	// codes for the HLIT and HDIST trees. Similarly, an empty HLIT tree is
-	// guaranteed to fail later since the compressed data section must be
-	// composed of at least one symbol (the end-of-block marker).
-	if max == 0 {
-		return true
-	}
-
-	code := 0
-	var nextcode [maxCodeLen]int
-	for i := min; i <= max; i++ {
-		code <<= 1
-		nextcode[i] = code
-		code += count[i]
-	}
-
-	// Check that the coding is complete (i.e., that we've
-	// assigned all 2-to-the-max possible bit sequences).
-	// Exception: To be compatible with zlib, we also need to
-	// accept degenerate single-code codings. See also
-	// TestDegenerateHuffmanCoding.
-	if code != 1<<uint(max) && (code != 1 || max != 1) {
-		return false
-	}
-
-	h.min = min
-	if max > huffmanChunkBits {
-		numLinks := 1 << (uint(max) - huffmanChunkBits)
-		h.linkMask = uint32(numLinks - 1)
-
-		// create link tables
-		link := nextcode[huffmanChunkBits+1] >> 1
-		h.links = make([][]uint32, huffmanNumChunks-link)
-		for j := uint(link); j < huffmanNumChunks; j++ {
-			reverse := int(bits.Reverse16(uint16(j)))
-			reverse >>= uint(16 - huffmanChunkBits)
-			off := j - uint(link)
-			if sanity && h.chunks[reverse] != 0 {
-				panic("impossible: overwriting existing chunk")
-			}
-			h.chunks[reverse] = uint32(off<<huffmanValueShift | (huffmanChunkBits + 1))
-			h.links[off] = make([]uint32, numLinks)
-		}
-	}
-
-	for i, n := range lengths {
-		if n == 0 {
-			continue
-		}
-		code := nextcode[n]
-		nextcode[n]++
-		chunk := uint32(i<<huffmanValueShift | n)
-		reverse := int(bits.Reverse16(uint16(code)))
-		reverse >>= uint(16 - n)
-		if n <= huffmanChunkBits {
-			for off := reverse; off < len(h.chunks); off += 1 << uint(n) {
-				// We should never need to overwrite
-				// an existing chunk. Also, 0 is
-				// never a valid chunk, because the
-				// lower 4 "count" bits should be
-				// between 1 and 15.
-				if sanity && h.chunks[off] != 0 {
-					panic("impossible: overwriting existing chunk")
-				}
-				h.chunks[off] = chunk
-			}
-		} else {
-			j := reverse & (huffmanNumChunks - 1)
-			if sanity && h.chunks[j]&huffmanCountMask != huffmanChunkBits+1 {
-				// Longer codes should have been
-				// associated with a link table above.
-				panic("impossible: not an indirect chunk")
-			}
-			value := h.chunks[j] >> huffmanValueShift
-			linktab := h.links[value]
-			reverse >>= huffmanChunkBits
-			for off := reverse; off < len(linktab); off += 1 << uint(n-huffmanChunkBits) {
-				if sanity && linktab[off] != 0 {
-					panic("impossible: overwriting existing chunk")
-				}
-				linktab[off] = chunk
-			}
-		}
-	}
-
-	if sanity {
-		// Above we've sanity checked that we never overwrote
-		// an existing entry. Here we additionally check that
-		// we filled the tables completely.
-		for i, chunk := range h.chunks {
-			if chunk == 0 {
-				// As an exception, in the degenerate
-				// single-code case, we allow odd
-				// chunks to be missing.
-				if code == 1 && i%2 == 1 {
-					continue
-				}
-				panic("impossible: missing chunk")
-			}
-		}
-		for _, linktab := range h.links {
-			for _, chunk := range linktab {
-				if chunk == 0 {
-					panic("impossible: missing chunk")
-				}
-			}
-		}
-	}
-
-	return true
-}
-
-// RFC 1951 section 3.2.7.
-// Compression with dynamic Huffman codes
-var codeOrder = [...]int{16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15}
-
-var (
-	// Initialize the fixedHuffmanDecoder only once upon first use.
-	fixedOnce           sync.Once
-	fixedHuffmanDecoder huffmanDecoder
-)
-
-func fixedHuffmanDecoderInit() {
-	fixedOnce.Do(func() {
-		// These come from the RFC section 3.2.6.
-		var bits [288]int
-		for i := 0; i < 144; i++ {
-			bits[i] = 8
-		}
-		for i := 144; i < 256; i++ {
-			bits[i] = 9
-		}
-		for i := 256; i < 280; i++ {
-			bits[i] = 7
-		}
-		for i := 280; i < 288; i++ {
-			bits[i] = 8
-		}
-		fixedHuffmanDecoder.init(bits[:])
-	})
-}
diff --git a/internal/flatex/slice_inflate.go b/internal/flatex/slice_inflate.go
deleted file mode 100644
index f9120143..00000000
--- a/internal/flatex/slice_inflate.go
+++ /dev/null
@@ -1,472 +0,0 @@
-package flatex
-
-import (
-	"io"
-	"math/bits"
-	"sync"
-)
-
-// sliceInflater is a specialized DEFLATE decoder that reads directly from an
-// in-memory byte slice. It mirrors the main decompressor but avoids the
-// overhead of the Reader interfaces, enabling faster byte-slice decoding.
-type sliceInflater struct {
-	input   []byte
-	pos     int
-	roffset int64
-
-	b  uint32
-	nb uint
-
-	h1, h2 huffmanDecoder
-
-	bits     *[maxNumLit + maxNumDist]int
-	codebits *[numCodes]int
-
-	window windowDecoder
-
-	toRead    []byte
-	step      func(*sliceInflater)
-	stepState int
-	final     bool
-	err       error
-	hl, hd    *huffmanDecoder
-	copyLen   int
-	copyDist  int
-}
-
-var sliceInflaterPool = sync.Pool{
-	New: func() any {
-		fixedHuffmanDecoderInit()
-		return &sliceInflater{
-			bits:     new([maxNumLit + maxNumDist]int),
-			codebits: new([numCodes]int),
-		}
-	},
-}
-
-func (f *sliceInflater) reset(src []byte) error {
-	bits := f.bits
-	codebits := f.codebits
-	windowState := f.window
-	*f = sliceInflater{
-		input:    src,
-		bits:     bits,
-		codebits: codebits,
-		window:   windowState,
-		step:     (*sliceInflater).nextBlock,
-	}
-	f.window.init(maxMatchOffset)
-	return nil
-}
-
-func (f *sliceInflater) nextBlock() {
-	for f.nb < 1+2 {
-		if err := f.moreBits(); err != nil {
-			f.err = err
-			return
-		}
-	}
-	f.final = f.b&1 == 1
-	f.b >>= 1
-	typ := f.b & 3
-	f.b >>= 2
-	f.nb -= 1 + 2
-	switch typ {
-	case 0:
-		f.dataBlock()
-	case 1:
-		f.hl = &fixedHuffmanDecoder
-		f.hd = nil
-		f.huffmanBlock()
-	case 2:
-		if err := f.readHuffman(); err != nil {
-			f.err = err
-			return
-		}
-		f.hl = &f.h1
-		f.hd = &f.h2
-		f.huffmanBlock()
-	default:
-		f.err = CorruptInputError(f.roffset)
-	}
-}
-
-func (f *sliceInflater) huffmanBlock() {
-	const (
-		stateInit = iota
-		stateDict
-	)
-	switch f.stepState {
-	case stateInit:
-		goto readLiteral
-	case stateDict:
-		goto copyHistory
-	}
-
-readLiteral:
-	{
-		v, err := f.huffSym(f.hl)
-		if err != nil {
-			f.err = err
-			return
-		}
-		var n uint
-		var length int
-		switch {
-		case v < 256:
-			f.window.writeByte(byte(v))
-			if f.window.availWrite() == 0 {
-				f.toRead = f.window.readFlush()
-				f.step = (*sliceInflater).huffmanBlock
-				f.stepState = stateInit
-				return
-			}
-			goto readLiteral
-		case v == 256:
-			f.finishBlock()
-			return
-		case v < 265:
-			length = v - (257 - 3)
-			n = 0
-		case v < 269:
-			length = v*2 - (265*2 - 11)
-			n = 1
-		case v < 273:
-			length = v*4 - (269*4 - 19)
-			n = 2
-		case v < 277:
-			length = v*8 - (273*8 - 35)
-			n = 3
-		case v < 281:
-			length = v*16 - (277*16 - 67)
-			n = 4
-		case v < 285:
-			length = v*32 - (281*32 - 131)
-			n = 5
-		case v < maxNumLit:
-			length = 258
-			n = 0
-		default:
-			f.err = CorruptInputError(f.roffset)
-			return
-		}
-		if n > 0 {
-			for f.nb < n {
-				if err = f.moreBits(); err != nil {
-					f.err = err
-					return
-				}
-			}
-			length += int(f.b & uint32(1<<n-1))
-			f.b >>= n
-			f.nb -= n
-		}
-
-		var dist int
-		if f.hd == nil {
-			for f.nb < 5 {
-				if err = f.moreBits(); err != nil {
-					f.err = err
-					return
-				}
-			}
-			dist = int(bits.Reverse8(uint8(f.b & 0x1F << 3)))
-			f.b >>= 5
-			f.nb -= 5
-		} else {
-			if dist, err = f.huffSym(f.hd); err != nil {
-				f.err = err
-				return
-			}
-		}
-
-		switch {
-		case dist < 4:
-			dist++
-		case dist < maxNumDist:
-			nb := uint(dist-2) >> 1
-			extra := (dist & 1) << nb
-			for f.nb < nb {
-				if err = f.moreBits(); err != nil {
-					f.err = err
-					return
-				}
-			}
-			extra |= int(f.b & uint32(1<<nb-1))
-			f.b >>= nb
-			f.nb -= nb
-			dist = 1<<(nb+1) + 1 + extra
-		default:
-			f.err = CorruptInputError(f.roffset)
-			return
-		}
-
-		if dist > f.window.histSize() {
-			f.err = CorruptInputError(f.roffset)
-			return
-		}
-
-		f.copyLen, f.copyDist = length, dist
-		goto copyHistory
-	}
-
-copyHistory:
-	{
-		cnt := f.window.tryWriteCopy(f.copyDist, f.copyLen)
-		if cnt == 0 {
-			cnt = f.window.writeCopy(f.copyDist, f.copyLen)
-		}
-		f.copyLen -= cnt
-
-		if f.window.availWrite() == 0 || f.copyLen > 0 {
-			f.toRead = f.window.readFlush()
-			f.step = (*sliceInflater).huffmanBlock
-			f.stepState = stateDict
-			return
-		}
-		goto readLiteral
-	}
-}
-
-func (f *sliceInflater) dataBlock() {
-	f.nb = 0
-	f.b = 0
-
-	if f.pos+4 > len(f.input) {
-		f.pos = len(f.input)
-		f.err = io.ErrUnexpectedEOF
-		return
-	}
-	hdr := f.input[f.pos : f.pos+4]
-	f.pos += 4
-	f.roffset += 4
-	n := int(hdr[0]) | int(hdr[1])<<8
-	nn := int(hdr[2]) | int(hdr[3])<<8
-	if uint16(nn) != uint16(^n) {
-		f.err = CorruptInputError(f.roffset)
-		return
-	}
-
-	if n == 0 {
-		f.toRead = f.window.readFlush()
-		f.finishBlock()
-		return
-	}
-
-	f.copyLen = n
-	f.copyData()
-}
-
-func (f *sliceInflater) copyData() {
-	for {
-		if f.copyLen == 0 {
-			f.finishBlock()
-			return
-		}
-		buf := f.window.writeSlice()
-		if len(buf) == 0 {
-			f.toRead = f.window.readFlush()
-			f.step = (*sliceInflater).copyData
-			return
-		}
-		n := f.copyLen
-		if n > len(buf) {
-			n = len(buf)
-		}
-		if f.pos+n > len(f.input) {
-			f.err = io.ErrUnexpectedEOF
-			return
-		}
-		copy(buf[:n], f.input[f.pos:f.pos+n])
-		f.pos += n
-		f.roffset += int64(n)
-		f.copyLen -= n
-		f.window.writeMark(n)
-		if f.window.availWrite() == 0 {
-			f.toRead = f.window.readFlush()
-			f.step = (*sliceInflater).copyData
-			return
-		}
-	}
-}
-
-func (f *sliceInflater) finishBlock() {
-	if f.final {
-		if f.window.availRead() > 0 {
-			f.toRead = f.window.readFlush()
-		}
-		f.err = io.EOF
-	}
-	f.step = (*sliceInflater).nextBlock
-	f.stepState = 0
-}
-
-func (f *sliceInflater) moreBits() error {
-	if f.pos >= len(f.input) {
-		return io.ErrUnexpectedEOF
-	}
-	c := f.input[f.pos]
-	f.pos++
-	f.roffset++
-	f.b |= uint32(c) << (f.nb & 31)
-	f.nb += 8
-	return nil
-}
-
-func (f *sliceInflater) huffSym(h *huffmanDecoder) (int, error) {
-	n := uint(h.min)
-	nb, b := f.nb, f.b
-	for {
-		for nb < n {
-			if f.pos >= len(f.input) {
-				f.b = b
-				f.nb = nb
-				return 0, io.ErrUnexpectedEOF
-			}
-			c := f.input[f.pos]
-			f.pos++
-			f.roffset++
-			b |= uint32(c) << (nb & 31)
-			nb += 8
-		}
-		chunk := h.chunks[b&(huffmanNumChunks-1)]
-		n = uint(chunk & huffmanCountMask)
-		if n > huffmanChunkBits {
-			chunk = h.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&h.linkMask]
-			n = uint(chunk & huffmanCountMask)
-		}
-		if n <= nb {
-			if n == 0 {
-				f.b = b
-				f.nb = nb
-				f.err = CorruptInputError(f.roffset)
-				return 0, f.err
-			}
-			f.b = b >> (n & 31)
-			f.nb = nb - n
-			return int(chunk >> huffmanValueShift), nil
-		}
-	}
-}
-
-func (f *sliceInflater) readHuffman() error {
-	for f.nb < 5+5+4 {
-		if err := f.moreBits(); err != nil {
-			return err
-		}
-	}
-	nlit := int(f.b&0x1F) + 257
-	if nlit > maxNumLit {
-		return CorruptInputError(f.roffset)
-	}
-	f.b >>= 5
-	ndist := int(f.b&0x1F) + 1
-	if ndist > maxNumDist {
-		return CorruptInputError(f.roffset)
-	}
-	f.b >>= 5
-	nclen := int(f.b&0xF) + 4
-	f.b >>= 4
-	f.nb -= 5 + 5 + 4
-	codebits := f.codebits[:]
-	bits := f.bits[:]
-	clear(codebits)
-	clear(bits)
-	for i := 0; i < nclen; i++ {
-		for f.nb < 3 {
-			if err := f.moreBits(); err != nil {
-				return err
-			}
-		}
-		codebits[codeOrder[i]] = int(f.b & 0x7)
-		f.b >>= 3
-		f.nb -= 3
-	}
-	if !f.h1.init(codebits) {
-		return CorruptInputError(f.roffset)
-	}
-	for i := range bits {
-		bits[i] = 0
-	}
-	i := 0
-	for i < nlit+ndist {
-		x, err := f.huffSym(&f.h1)
-		if err != nil {
-			return err
-		}
-		switch {
-		case x < 16:
-			bits[i] = x
-			i++
-		case x == 16:
-			if i == 0 {
-				return CorruptInputError(f.roffset)
-			}
-			repeat := 3
-			for f.nb < 2 {
-				if err := f.moreBits(); err != nil {
-					return err
-				}
-			}
-			repeat += int(f.b & 0x3)
-			f.b >>= 2
-			f.nb -= 2
-			for repeat > 0 {
-				if i >= len(bits) {
-					return CorruptInputError(f.roffset)
-				}
-				bits[i] = bits[i-1]
-				i++
-				repeat--
-			}
-		case x == 17:
-			repeat := 3
-			for f.nb < 3 {
-				if err := f.moreBits(); err != nil {
-					return err
-				}
-			}
-			repeat += int(f.b & 0x7)
-			f.b >>= 3
-			f.nb -= 3
-			for repeat > 0 {
-				if i >= len(bits) {
-					return CorruptInputError(f.roffset)
-				}
-				bits[i] = 0
-				i++
-				repeat--
-			}
-		case x == 18:
-			repeat := 11
-			for f.nb < 7 {
-				if err := f.moreBits(); err != nil {
-					return err
-				}
-			}
-			repeat += int(f.b & 0x7F)
-			f.b >>= 7
-			f.nb -= 7
-			for repeat > 0 {
-				if i >= len(bits) {
-					return CorruptInputError(f.roffset)
-				}
-				bits[i] = 0
-				i++
-				repeat--
-			}
-		default:
-			return CorruptInputError(f.roffset)
-		}
-	}
-	if !f.h1.init(bits[:nlit]) {
-		return CorruptInputError(f.roffset)
-	}
-	if !f.h2.init(bits[nlit : nlit+ndist]) {
-		return CorruptInputError(f.roffset)
-	}
-	if f.h1.min < bits[endBlockMarker] {
-		f.h1.min = bits[endBlockMarker]
-	}
-	return nil
-}
diff --git a/internal/flatex/window_decoder.go b/internal/flatex/window_decoder.go
deleted file mode 100644
index 492c6a96..00000000
--- a/internal/flatex/window_decoder.go
+++ /dev/null
@@ -1,101 +0,0 @@
-package flatex
-
-// windowDecoder implements the sliding window used in decompression.
-type windowDecoder struct {
-	hist []byte
-
-	wrPos int
-	rdPos int
-	full  bool
-}
-
-func (wd *windowDecoder) init(size int) {
-	*wd = windowDecoder{hist: wd.hist}
-
-	if cap(wd.hist) < size {
-		wd.hist = make([]byte, size)
-	}
-	wd.hist = wd.hist[:size]
-
-	wd.wrPos = 0
-	wd.rdPos = 0
-	wd.full = false
-}
-
-func (wd *windowDecoder) histSize() int {
-	if wd.full {
-		return len(wd.hist)
-	}
-	return wd.wrPos
-}
-
-func (wd *windowDecoder) availRead() int {
-	return wd.wrPos - wd.rdPos
-}
-
-func (wd *windowDecoder) availWrite() int {
-	return len(wd.hist) - wd.wrPos
-}
-
-func (wd *windowDecoder) writeSlice() []byte {
-	return wd.hist[wd.wrPos:]
-}
-
-func (wd *windowDecoder) writeMark(cnt int) {
-	wd.wrPos += cnt
-}
-
-func (wd *windowDecoder) writeByte(c byte) {
-	wd.hist[wd.wrPos] = c
-	wd.wrPos++
-}
-
-func (wd *windowDecoder) writeCopy(dist, length int) int {
-	dstBase := wd.wrPos
-	dstPos := dstBase
-	srcPos := dstPos - dist
-	endPos := dstPos + length
-	if endPos > len(wd.hist) {
-		endPos = len(wd.hist)
-	}
-
-	if srcPos < 0 {
-		srcPos += len(wd.hist)
-		dstPos += copy(wd.hist[dstPos:endPos], wd.hist[srcPos:])
-		srcPos = 0
-	}
-
-	for dstPos < endPos {
-		dstPos += copy(wd.hist[dstPos:endPos], wd.hist[srcPos:dstPos])
-	}
-
-	wd.wrPos = dstPos
-	return dstPos - dstBase
-}
-
-func (wd *windowDecoder) tryWriteCopy(dist, length int) int {
-	dstPos := wd.wrPos
-	endPos := dstPos + length
-	if dstPos < dist || endPos > len(wd.hist) {
-		return 0
-	}
-	dstBase := dstPos
-	srcPos := dstPos - dist
-
-	for dstPos < endPos {
-		dstPos += copy(wd.hist[dstPos:endPos], wd.hist[srcPos:dstPos])
-	}
-
-	wd.wrPos = dstPos
-	return dstPos - dstBase
-}
-
-func (wd *windowDecoder) readFlush() []byte {
-	toRead := wd.hist[wd.rdPos:wd.wrPos]
-	wd.rdPos = wd.wrPos
-	if wd.wrPos == len(wd.hist) {
-		wd.wrPos, wd.rdPos = 0, 0
-		wd.full = true
-	}
-	return toRead
-}
diff --git a/internal/zlib/LICENSE b/internal/zlib/LICENSE
deleted file mode 100644
index 2a7cf70d..00000000
--- a/internal/zlib/LICENSE
+++ /dev/null
@@ -1,27 +0,0 @@
-Copyright 2009 The Go Authors.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are
-met:
-
-   * Redistributions of source code must retain the above copyright
-notice, this list of conditions and the following disclaimer.
-   * Redistributions in binary form must reproduce the above
-copyright notice, this list of conditions and the following disclaimer
-in the documentation and/or other materials provided with the
-distribution.
-   * Neither the name of Google LLC nor the names of its
-contributors may be used to endorse or promote products derived from
-this software without specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
diff --git a/internal/zlib/reader.go b/internal/zlib/reader.go
deleted file mode 100644
index 2234e7e0..00000000
--- a/internal/zlib/reader.go
+++ /dev/null
@@ -1,199 +0,0 @@
-// Copyright 2009 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-/*
-Package zlib implements reading and writing of zlib format compressed data,
-as specified in RFC 1950.
-
-This package differs from the standard library's compress/zlib package
-in that it pools readers and writers to reduce allocations.
-
-Note that closing a reader or writer causes it to be returned to a pool
-for reuse. Therefore, the caller must not retain references to a
-reader or writer after closing it; in the standard library's
-compress/zlib package, it is legal to Reset a closed reader or writer
-and continue using it; that is not allowed here, so there is simply no
-Resetter interface.
-
-The implementation provides filters that uncompress during reading
-and compress during writing.  For example, to write compressed data
-to a buffer:
-
-	var b bytes.Buffer
-	w := zlib.NewWriter(&b)
-	w.Write([]byte("hello, world\n"))
-	w.Close()
-
-and to read that data back:
-
-	r, err := zlib.NewReader(&b)
-	io.Copy(os.Stdout, r)
-	r.Close()
-*/
-package zlib
-
-import (
-	"bufio"
-	"compress/flate"
-	"encoding/binary"
-	"errors"
-	"hash"
-	"io"
-	"sync"
-
-	"codeberg.org/lindenii/furgit/internal/adler32"
-)
-
-const (
-	zlibDeflate   = 8
-	zlibMaxWindow = 7
-)
-
-var (
-	// ErrChecksum is returned when reading ZLIB data that has an invalid checksum.
-	ErrChecksum = errors.New("zlib: invalid checksum")
-	// ErrDictionary is returned when reading ZLIB data that has an invalid dictionary.
-	ErrDictionary = errors.New("zlib: invalid dictionary")
-	// ErrHeader is returned when reading ZLIB data that has an invalid header.
-	ErrHeader = errors.New("zlib: invalid header")
-)
-
-var readerPool = sync.Pool{
-	New: func() any {
-		r := new(reader)
-		return r
-	},
-}
-
-type reader struct {
-	r            flate.Reader
-	decompressor io.ReadCloser
-	digest       hash.Hash32
-	err          error
-	scratch      [4]byte
-}
-
-// NewReader creates a new ReadCloser.
-// Reads from the returned ReadCloser read and decompress data from r.
-// If r does not implement [io.ByteReader], the decompressor may read more
-// data than necessary from r.
-// It is the caller's responsibility to call Close on the ReadCloser when done.
-func NewReader(r io.Reader) (io.ReadCloser, error) {
-	return NewReaderDict(r, nil)
-}
-
-// NewReaderDict is like [NewReader] but uses a preset dictionary.
-// NewReaderDict ignores the dictionary if the compressed data does not refer to it.
-// If the compressed data refers to a different dictionary, NewReaderDict returns [ErrDictionary].
-func NewReaderDict(r io.Reader, dict []byte) (io.ReadCloser, error) {
-	v := readerPool.Get()
-	z, ok := v.(*reader)
-	if !ok {
-		panic("zlib: pool returned unexpected type")
-	}
-	err := z.Reset(r, dict)
-	if err != nil {
-		return nil, err
-	}
-	return z, nil
-}
-
-func (z *reader) Read(p []byte) (int, error) {
-	if z.err != nil {
-		return 0, z.err
-	}
-
-	var n int
-	n, z.err = z.decompressor.Read(p)
-	z.digest.Write(p[0:n])
-	if z.err != io.EOF {
-		// In the normal case we return here.
-		return n, z.err
-	}
-
-	// Finished file; check checksum.
-	if _, err := io.ReadFull(z.r, z.scratch[0:4]); err != nil {
-		if err == io.EOF {
-			err = io.ErrUnexpectedEOF
-		}
-		z.err = err
-		return n, z.err
-	}
-	// ZLIB (RFC 1950) is big-endian, unlike GZIP (RFC 1952).
-	checksum := binary.BigEndian.Uint32(z.scratch[:4])
-	if checksum != z.digest.Sum32() {
-		z.err = ErrChecksum
-		return n, z.err
-	}
-	return n, io.EOF
-}
-
-// Calling Close does not close the wrapped [io.Reader] originally passed to [NewReader].
-// In order for the ZLIB checksum to be verified, the reader must be
-// fully consumed until the [io.EOF].
-func (z *reader) Close() error {
-	if z.err != nil && z.err != io.EOF {
-		return z.err
-	}
-	z.err = z.decompressor.Close()
-	if z.err != nil {
-		return z.err
-	}
-
-	readerPool.Put(z)
-	return nil
-}
-
-func (z *reader) Reset(r io.Reader, dict []byte) error {
-	*z = reader{decompressor: z.decompressor}
-	if fr, ok := r.(flate.Reader); ok {
-		z.r = fr
-	} else {
-		z.r = bufio.NewReader(r)
-	}
-
-	// Read the header (RFC 1950 section 2.2.).
-	_, z.err = io.ReadFull(z.r, z.scratch[0:2])
-	if z.err != nil {
-		if z.err == io.EOF {
-			z.err = io.ErrUnexpectedEOF
-		}
-		return z.err
-	}
-	h := binary.BigEndian.Uint16(z.scratch[:2])
-	if (z.scratch[0]&0x0f != zlibDeflate) || (z.scratch[0]>>4 > zlibMaxWindow) || (h%31 != 0) {
-		z.err = ErrHeader
-		return z.err
-	}
-	haveDict := z.scratch[1]&0x20 != 0
-	if haveDict {
-		_, z.err = io.ReadFull(z.r, z.scratch[0:4])
-		if z.err != nil {
-			if z.err == io.EOF {
-				z.err = io.ErrUnexpectedEOF
-			}
-			return z.err
-		}
-		checksum := binary.BigEndian.Uint32(z.scratch[:4])
-		if checksum != adler32.Checksum(dict) {
-			z.err = ErrDictionary
-			return z.err
-		}
-	}
-
-	if z.decompressor == nil {
-		if haveDict {
-			z.decompressor = flate.NewReaderDict(z.r, dict)
-		} else {
-			z.decompressor = flate.NewReader(z.r)
-		}
-	} else {
-		z.err = z.decompressor.(flate.Resetter).Reset(z.r, dict)
-		if z.err != nil {
-			return z.err
-		}
-	}
-	z.digest = adler32.New()
-	return nil
-}
diff --git a/internal/zlib/writer.go b/internal/zlib/writer.go
deleted file mode 100644
index 81c57f55..00000000
--- a/internal/zlib/writer.go
+++ /dev/null
@@ -1,228 +0,0 @@
-// Copyright 2009 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-package zlib
-
-import (
-	"compress/flate"
-	"encoding/binary"
-	"fmt"
-	"hash"
-	"io"
-	"sync"
-
-	"codeberg.org/lindenii/furgit/internal/adler32"
-)
-
-// These constants are copied from the [flate] package, so that code that imports
-// [compress/zlib] does not also have to import [compress/flate].
-const (
-	NoCompression      = flate.NoCompression
-	BestSpeed          = flate.BestSpeed
-	BestCompression    = flate.BestCompression
-	DefaultCompression = flate.DefaultCompression
-	HuffmanOnly        = flate.HuffmanOnly
-)
-
-// A Writer takes data written to it and writes the compressed
-// form of that data to an underlying writer (see [NewWriter]).
-type Writer struct {
-	w           io.Writer
-	level       int
-	dict        []byte
-	compressor  *flate.Writer
-	digest      hash.Hash32
-	err         error
-	scratch     [4]byte
-	wroteHeader bool
-}
-
-var writerPool = sync.Pool{
-	New: func() any {
-		return new(Writer)
-	},
-}
-
-// NewWriter creates a new [Writer].
-// Writes to the returned Writer are compressed and written to w.
-//
-// It is the caller's responsibility to call Close on the Writer when done.
-// Writes may be buffered and not flushed until Close.
-func NewWriter(w io.Writer) *Writer {
-	z, _ := NewWriterLevelDict(w, DefaultCompression, nil)
-	return z
-}
-
-// NewWriterLevel is like [NewWriter] but specifies the compression level instead
-// of assuming [DefaultCompression].
-//
-// The compression level can be [DefaultCompression], [NoCompression], [HuffmanOnly]
-// or any integer value between [BestSpeed] and [BestCompression] inclusive.
-// The error returned will be nil if the level is valid.
-func NewWriterLevel(w io.Writer, level int) (*Writer, error) {
-	return NewWriterLevelDict(w, level, nil)
-}
-
-// NewWriterLevelDict is like [NewWriterLevel] but specifies a dictionary to
-// compress with.
-//
-// The dictionary may be nil. If not, its contents should not be modified until
-// the Writer is closed.
-func NewWriterLevelDict(w io.Writer, level int, dict []byte) (*Writer, error) {
-	if level < HuffmanOnly || level > BestCompression {
-		return nil, fmt.Errorf("zlib: invalid compression level: %d", level)
-	}
-	v := writerPool.Get()
-	z, ok := v.(*Writer)
-	if !ok {
-		panic("zlib: pool returned unexpected type")
-	}
-
-	// flate.Writer can only be Reset with the same level/dictionary mode.
-	// Reuse it only when the configuration is unchanged and dictionary-free.
-	reuseCompressor := z.compressor != nil && z.level == level && z.dict == nil && dict == nil
-	if !reuseCompressor {
-		z.compressor = nil
-	}
-	if z.digest != nil {
-		z.digest.Reset()
-	}
-
-	*z = Writer{
-		w:          w,
-		level:      level,
-		dict:       dict,
-		compressor: z.compressor,
-		digest:     z.digest,
-	}
-	if z.compressor != nil {
-		z.compressor.Reset(w)
-	}
-	return z, nil
-}
-
-// Reset clears the state of the [Writer] z such that it is equivalent to its
-// initial state from [NewWriterLevel] or [NewWriterLevelDict], but instead writing
-// to w.
-func (z *Writer) Reset(w io.Writer) {
-	z.w = w
-	// z.level and z.dict left unchanged.
-	if z.compressor != nil {
-		z.compressor.Reset(w)
-	}
-	if z.digest != nil {
-		z.digest.Reset()
-	}
-	z.err = nil
-	z.scratch = [4]byte{}
-	z.wroteHeader = false
-}
-
-// writeHeader writes the ZLIB header.
-func (z *Writer) writeHeader() (err error) {
-	z.wroteHeader = true
-	// ZLIB has a two-byte header (as documented in RFC 1950).
-	// The first four bits is the CINFO (compression info), which is 7 for the default deflate window size.
-	// The next four bits is the CM (compression method), which is 8 for deflate.
-	z.scratch[0] = 0x78
-	// The next two bits is the FLEVEL (compression level). The four values are:
-	// 0=fastest, 1=fast, 2=default, 3=best.
-	// The next bit, FDICT, is set if a dictionary is given.
-	// The final five FCHECK bits form a mod-31 checksum.
-	switch z.level {
-	case -2, 0, 1:
-		z.scratch[1] = 0 << 6
-	case 2, 3, 4, 5:
-		z.scratch[1] = 1 << 6
-	case 6, -1:
-		z.scratch[1] = 2 << 6
-	case 7, 8, 9:
-		z.scratch[1] = 3 << 6
-	default:
-		panic("unreachable")
-	}
-	if z.dict != nil {
-		z.scratch[1] |= 1 << 5
-	}
-	z.scratch[1] += uint8(31 - binary.BigEndian.Uint16(z.scratch[:2])%31)
-	if _, err = z.w.Write(z.scratch[0:2]); err != nil {
-		return err
-	}
-	if z.dict != nil {
-		// The next four bytes are the Adler-32 checksum of the dictionary.
-		binary.BigEndian.PutUint32(z.scratch[:], adler32.Checksum(z.dict))
-		if _, err = z.w.Write(z.scratch[0:4]); err != nil {
-			return err
-		}
-	}
-	if z.compressor == nil {
-		// Initialize deflater unless the Writer is being reused
-		// after a Reset call.
-		z.compressor, err = flate.NewWriterDict(z.w, z.level, z.dict)
-		if err != nil {
-			return err
-		}
-		z.digest = adler32.New()
-	}
-	return nil
-}
-
-// Write writes a compressed form of p to the underlying [io.Writer]. The
-// compressed bytes are not necessarily flushed until the [Writer] is closed or
-// explicitly flushed.
-func (z *Writer) Write(p []byte) (n int, err error) {
-	if !z.wroteHeader {
-		z.err = z.writeHeader()
-	}
-	if z.err != nil {
-		return 0, z.err
-	}
-	if len(p) == 0 {
-		return 0, nil
-	}
-	n, err = z.compressor.Write(p)
-	if err != nil {
-		z.err = err
-		return
-	}
-	z.digest.Write(p)
-	return
-}
-
-// Flush flushes the Writer to its underlying [io.Writer].
-func (z *Writer) Flush() error {
-	if !z.wroteHeader {
-		z.err = z.writeHeader()
-	}
-	if z.err != nil {
-		return z.err
-	}
-	z.err = z.compressor.Flush()
-	return z.err
-}
-
-// Close closes the Writer, flushing any unwritten data to the underlying
-// [io.Writer], but does not close the underlying io.Writer.
-func (z *Writer) Close() error {
-	if !z.wroteHeader {
-		z.err = z.writeHeader()
-	}
-	if z.err != nil {
-		return z.err
-	}
-	z.err = z.compressor.Close()
-	if z.err != nil {
-		return z.err
-	}
-	checksum := z.digest.Sum32()
-	// ZLIB (RFC 1950) is big-endian, unlike GZIP (RFC 1952).
-	binary.BigEndian.PutUint32(z.scratch[:], checksum)
-	_, z.err = z.w.Write(z.scratch[0:4])
-	if z.err != nil {
-		return z.err
-	}
-
-	writerPool.Put(z)
-	return nil
-}
diff --git a/internal/zlibx/LICENSE b/internal/zlibx/LICENSE
deleted file mode 100644
index 2a7cf70d..00000000
--- a/internal/zlibx/LICENSE
+++ /dev/null
@@ -1,27 +0,0 @@
-Copyright 2009 The Go Authors.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are
-met:
-
-   * Redistributions of source code must retain the above copyright
-notice, this list of conditions and the following disclaimer.
-   * Redistributions in binary form must reproduce the above
-copyright notice, this list of conditions and the following disclaimer
-in the documentation and/or other materials provided with the
-distribution.
-   * Neither the name of Google LLC nor the names of its
-contributors may be used to endorse or promote products derived from
-this software without specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
diff --git a/internal/zlibx/constants.go b/internal/zlibx/constants.go
deleted file mode 100644
index 161e3458..00000000
--- a/internal/zlibx/constants.go
+++ /dev/null
@@ -1,52 +0,0 @@
-// Copyright 2009 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-/*
-Package zlibx implements reading of zlib format compressed data,
-as specified in RFC 1950.
-
-This package differs from the standard library's compress/zlib package
-in that it pools readers to reduce allocations. Writing is unsupported.
-
-THis package will likely be refactorered much more for our specific
-use case of only doing full decompressions to byte slices.
-
-Note that closing the reader causes it to be returned to a pool for
-reuse. Therefore, the caller must not retain references to the
-reader after closing it; in the standard library's compress/zlib package,
-it is legal to Reset a closed reader and continue using it; that is
-not allowed here, so there is simply no Resetter interface.
-
-The implementation provides filters that uncompress during reading
-and compress during writing.  For example, to write compressed data
-to a buffer:
-
-	var b bytes.Buffer
-	w := zlib.NewWriter(&b)
-	w.Write([]byte("hello, world\n"))
-	w.Close()
-
-and to read that data back:
-
-	r, err := zlib.NewReader(&b)
-	io.Copy(os.Stdout, r)
-	r.Close()
-*/
-package zlibx
-
-import (
-	"errors"
-)
-
-const (
-	zlibDeflate   = 8
-	zlibMaxWindow = 7
-)
-
-var (
-	// ErrChecksum is returned when reading ZLIB data that has an invalid checksum.
-	ErrChecksum = errors.New("zlib: invalid checksum")
-	// ErrHeader is returned when reading ZLIB data that has an invalid header.
-	ErrHeader = errors.New("zlib: invalid header")
-)
diff --git a/internal/zlibx/decompress.go b/internal/zlibx/decompress.go
deleted file mode 100644
index 126c1fcb..00000000
--- a/internal/zlibx/decompress.go
+++ /dev/null
@@ -1,54 +0,0 @@
-package zlibx
-
-import (
-	"encoding/binary"
-	"io"
-
-	"codeberg.org/lindenii/furgit/internal/adler32"
-	"codeberg.org/lindenii/furgit/internal/bufpool"
-	"codeberg.org/lindenii/furgit/internal/flatex"
-)
-
-func Decompress(src []byte) (bufpool.Buffer, error) {
-	out, _, err := DecompressSized(src, 0)
-	return out, err
-}
-
-func DecompressSized(src []byte, sizeHint int) (buf bufpool.Buffer, consumed int, err error) {
-	if len(src) < 6 {
-		return bufpool.Buffer{}, 0, io.ErrUnexpectedEOF
-	}
-
-	cmf := src[0]
-	flg := src[1]
-	if (cmf&0x0f != zlibDeflate) || (cmf>>4 > zlibMaxWindow) || (binary.BigEndian.Uint16(src[:2])%31 != 0) {
-		return bufpool.Buffer{}, 0, ErrHeader
-	}
-
-	offset := 2
-	if flg&0x20 != 0 {
-		return bufpool.Buffer{}, 0, ErrHeader
-	}
-
-	if len(src[offset:]) < 4 {
-		return bufpool.Buffer{}, 0, io.ErrUnexpectedEOF
-	}
-
-	deflateData := src[offset:]
-	out, consumed, err := flatex.DecompressSized(deflateData, sizeHint)
-	if err != nil {
-		return bufpool.Buffer{}, 0, err
-	}
-
-	checksumPos := offset + consumed
-	if checksumPos+4 > len(src) {
-		out.Release()
-		return bufpool.Buffer{}, 0, io.ErrUnexpectedEOF
-	}
-	expected := binary.BigEndian.Uint32(src[checksumPos : checksumPos+4])
-	if expected != adler32.Checksum(out.Bytes()) {
-		out.Release()
-		return bufpool.Buffer{}, 0, ErrChecksum
-	}
-	return out, checksumPos + 4, nil
-}
diff --git a/internal/zlibx/decompress_test.go b/internal/zlibx/decompress_test.go
deleted file mode 100644
index bea348d2..00000000
--- a/internal/zlibx/decompress_test.go
+++ /dev/null
@@ -1,170 +0,0 @@
-package zlibx
-
-import (
-	"bytes"
-	stdzlib "compress/zlib"
-	"crypto/rand"
-	"testing"
-)
-
-func compressZlib(t *testing.T, payload []byte) []byte {
-	t.Helper()
-	var buf bytes.Buffer
-	w := stdzlib.NewWriter(&buf)
-	if _, err := w.Write(payload); err != nil {
-		t.Fatalf("Write: %v", err)
-	}
-	if err := w.Close(); err != nil {
-		t.Fatalf("Close: %v", err)
-	}
-	return buf.Bytes()
-}
-
-func TestDecompress(t *testing.T) {
-	makeRand := func(n int) []byte {
-		b := make([]byte, n)
-		if _, err := rand.Read(b); err != nil {
-			t.Fatalf("rand.Read: %v", err)
-		}
-		return b
-	}
-
-	type tc struct {
-		name    string
-		payload []byte
-	}
-
-	tests := []tc{
-		{
-			name:    "simple-hello",
-			payload: []byte("hello, zlib world!"),
-		},
-		{
-			name:    "empty",
-			payload: []byte{},
-		},
-		{
-			name:    "single-byte",
-			payload: []byte{0x42},
-		},
-		{
-			name:    "all-zero-1k",
-			payload: bytes.Repeat([]byte{0}, 1024),
-		},
-		{
-			name:    "all-FF-1k",
-			payload: bytes.Repeat([]byte{0xFF}, 1024),
-		},
-		{
-			name:    "ascii-repeated-pattern",
-			payload: bytes.Repeat([]byte("ABC123!"), 500),
-		},
-		{
-			name: "binary-structured",
-			payload: []byte{
-				0x00, 0x01, 0x02, 0x03,
-				0x10, 0x20, 0x30, 0x40,
-				0xFF, 0xEE, 0xDD, 0xCC,
-			},
-		},
-		{
-			name:    "1k-crypto-random",
-			payload: makeRand(1024),
-		},
-		{
-			name:    "32k-crypto-random",
-			payload: makeRand(32 * 1024),
-		},
-		{
-			name:    "256k-crypto-random",
-			payload: makeRand(256 * 1024),
-		},
-		{
-			name:    "highly-compressible-large",
-			payload: bytes.Repeat([]byte("AAAAAAAAAAAAAAAAAAAA"), 50_000),
-		},
-		{
-			name:    "json",
-			payload: []byte(`{"name":"test","values":[1,2,3,4],"deep":{"x":123,"y":"abc"}}`),
-		},
-		{
-			name:    "html",
-			payload: []byte("<html><body><h1>Title</h1><p>Paragraph</p></body></html>"),
-		},
-		{
-			name: "alternating-binary-pattern",
-			payload: func() []byte {
-				b := make([]byte, 4096)
-				for i := 0; i < len(b); i++ {
-					if i%2 == 0 {
-						b[i] = 0xAA
-					} else {
-						b[i] = 0x55
-					}
-				}
-				return b
-			}(),
-		},
-		{
-			name:    "large-repetitive-words",
-			payload: bytes.Repeat([]byte("the quick brown fox jumps over the lazy dog\n"), 4000),
-		},
-		{
-			name:    "unicode",
-			payload: []byte("我不知道该说点啥就随便打点字吧🤷‍♀️"),
-		},
-		{
-			name:    "multi-meg-random-2MB",
-			payload: makeRand(2 * 1024 * 1024),
-		},
-		{
-			name:    "multi-meg-random-16MB",
-			payload: makeRand(16 * 1024 * 1024),
-		},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			compressed := compressZlib(t, tt.payload)
-
-			out, err := Decompress(compressed)
-			if err != nil {
-				t.Fatalf("Decompress: %v", err)
-			}
-			defer out.Release()
-
-			if !bytes.Equal(out.Bytes(), tt.payload) {
-				t.Fatalf("payload mismatch: got %d bytes, want %d", len(out.Bytes()), len(tt.payload))
-			}
-		})
-	}
-}
-
-func TestDecompressChecksumError(t *testing.T) {
-	payload := []byte("checksum check")
-	compressed := compressZlib(t, payload)
-	compressed[len(compressed)-1] ^= 0xff
-
-	if _, err := Decompress(compressed); err != ErrChecksum {
-		t.Fatalf("expected ErrChecksum, got %v", err)
-	}
-}
-
-func TestDecompressSizedUsesHint(t *testing.T) {
-	payload := []byte("tiny payload")
-	compressed := compressZlib(t, payload)
-
-	const hint = 1 << 20
-	out, _, err := DecompressSized(compressed, hint)
-	if err != nil {
-		t.Fatalf("DecompressSized: %v", err)
-	}
-	defer out.Release()
-
-	if !bytes.Equal(out.Bytes(), payload) {
-		t.Fatalf("unexpected payload %q", out.Bytes())
-	}
-	if cap(out.Bytes()) < hint {
-		t.Fatalf("expected capacity >= %d, got %d", hint, cap(out.Bytes()))
-	}
-}
author	Runxi Yu	2026-02-20 19:06:13 +0800
committer	Runxi Yu	2026-02-20 19:07:14 +0800
commit	aa513c069c1418734aea894dc944e27c6a78a3bb (patch)
tree	687f0a11bb550fa088fd82a98ceb8979bbc35f69 /internal
parent	Comment on prior reverts removing the pack writing API (diff)
signature	No signature