Parshred

High-performance SIMD-accelerated XML parser for C++20 and Python.

Parshred is built from the ground up for modern CPUs: it scans XML structure with AVX-512/AVX2/SSE4.2/NEON, keeps data in zero-copy string_view ranges, and offers a compact 32-byte DOM node. The result is a small, fast, standards-focused library that works well for both C++ and Python applications.

Feature Highlights

2.5–2.7 GB/s DOM parsing on x86-64 with AVX2; matches RapidXML at small/mid sizes (within ~5%) and is 1.5–2.4x faster at 64 MB+, where RapidXML's allocation pattern thrashes. Parshred's compact 32-byte DOM node keeps the large-file advantage consistent across machines.
5–8x faster than lxml in Python, with competitive or better memory efficiency.
Multiple APIs: DOM, SAX, Pull Parser (StAX-style), and a streaming pipeline for files larger than memory.
XPath 1.0 with full operator, function, and predicate support.
XML Namespace 1.0, internal DTD validation, and XSD simple type validation.
SIMD acceleration: AVX-512, AVX2, SSE4.2, and ARM NEON with runtime dispatch.
32-byte compact DOM nodes — roughly 3x less memory than traditional node-based parsers.
573 unit tests, fuzz-tested, and run against the official W3C XML Conformance Test Suite in CI (results published as a workflow artifact; see Conformance for the current pass/fail breakdown by category).

Performance

All numbers below are measured, not projected. They vary by CPU, memory subsystem, and compiler; rerun bench/bench_dom and bench/bench_python.py on your target hardware for accurate figures. See docs/performance.md for methodology and the benchmark-regression CI job that guards against perf regressions.

C++ DOM throughput (x86-64, AVX2)

Reference machine: AMD Ryzen 5 7640U, 12 threads, GCC 16.1 -O3, single thread. Median of repeated runs; bench/bench_dom.

Size	Parshred (FastDOM)	RapidXML	pugixml	vs RapidXML
64 KB	2572 MB/s	2477	1465	1.04x
1 MB	2722 MB/s	2757	821	0.99x
16 MB	2519 MB/s	2421	1585	1.04x
64 MB	1678 MB/s	915	641	1.83x
128 MB	1317 MB/s	866	712	1.52x

Reading the table: at 64 KB–16 MB both parsers are bound by the same scalar setup and run within ~5% of each other — Parshred does not claim a win there. The gap opens at 64 MB+ because RapidXML allocates per-node on the heap while Parshred packs nodes into a single malloc'd array, so cache behavior and allocator pressure diverge. On machines with more cache or a faster allocator the large-file ratio trends higher (we have observed up to 2.4x); on this laptop it is 1.5–1.8x.

Python throughput

Reference machine: as above, Python 3.14, lxml 6.1.1, single thread. Median of 3 runs; bench/bench_python.py.

Size	Parshred	lxml (DOM)	lxml (iterparse)	vs lxml (DOM)
10 MB	527 MB/s	91	106	5.8x
100 MB	575 MB/s	90	101	6.4x

lxml's iterparse streaming mode is its fastest path for large files; Parshred's full-DOM path still beats it because the DOM build is zero-copy into string_view ranges with no per-node allocation.

Quick Start

CMake (FetchContent)

include(FetchContent)
FetchContent_Declare(parshred
    GIT_REPOSITORY https://github.com/parshred/parshred.git
    GIT_TAG v0.1.0
)
FetchContent_MakeAvailable(parshred)
target_link_libraries(myapp PRIVATE parshred::parshred)

Build from Source

git clone https://github.com/parshred/parshred.git
cd parshred
cmake --preset release
cmake --build build/release
sudo cmake --install build/release

Python

pip install parshred

Usage Examples

DOM Parsing (C++)

#include <parshred/parshred.hpp>

std::string xml = "<catalog><book id=\"1\">Title</book></catalog>";
auto dom = parshred::fast_dom_parse<0>(xml.data(), xml.size());

// XPath queries
auto title = parshred::xpath::evaluate_string(dom, "/catalog/book/text()");
auto id = parshred::xpath::evaluate_string(dom, "/catalog/book/@id");

Pull Parser (C++)

#include <parshred/pull_parser.hpp>

parshred::XmlReader reader(xml);
while (reader.next()) {
    if (reader.is_start_element("book")) {
        auto id = reader.attribute("id");
        auto text = reader.read_element_text();
    }
}

Python

import parshred

doc = parshred.parse_file("catalog.xml")
for book in doc.xpath("//book"):
    print(book.attr("id"), book.text)

Build XML Programmatically

#include <parshred/writer.hpp>

parshred::DomBuilder builder;
builder.start_element("catalog");
builder.start_element("book");
builder.add_attribute("id", "1");
builder.add_text("Title");
builder.end_element();
builder.end_element();

auto dom = builder.build();
parshred::XmlWriter writer;
std::string xml = writer.serialize(dom);

Documentation

Requirements

C++20 compiler: GCC 12+, Clang 15+, or MSVC 2022+
CMake 3.20+
x86-64 with SSE4.2 minimum (AVX2/AVX-512 recommended) or ARM64 with NEON
Python 3.9+ (for the optional Python bindings)

License

Parshred is licensed under the GNU Affero General Public License v3.0 or later (AGPL-3.0-or-later) for open-source use. See LICENSE for the full license text.

A commercial license is available for proprietary use. See COMMERCIAL_LICENSE.md for details.

Contributing

Contributions are welcome. Please see CONTRIBUTING.md for guidelines on code style, testing, and submitting pull requests.

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

parshred-0.1.5.tar.gz (232.5 kB view details)

Uploaded Jul 29, 2026 Source

Built Distributions

If you're not sure about the file name format, learn more about wheel file names.

The dropdown lists show the available interpreters, ABIs, and platforms. Enable javascript to be able to filter the list of wheel files.

parshred-0.1.5-cp313-cp313-musllinux_1_2_x86_64.whl (1.2 MB view details)

Uploaded Jul 29, 2026 CPython 3.13musllinux: musl 1.2+ x86-64

parshred-0.1.5-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.whl (188.1 kB view details)

Uploaded Jul 29, 2026 CPython 3.13manylinux: glibc 2.17+ x86-64

parshred-0.1.5-cp312-cp312-win_amd64.whl (342.4 kB view details)

Uploaded Jul 29, 2026 CPython 3.12Windows x86-64

parshred-0.1.5-cp312-cp312-musllinux_1_2_x86_64.whl (1.2 MB view details)

Uploaded Jul 29, 2026 CPython 3.12musllinux: musl 1.2+ x86-64

parshred-0.1.5-cp312-cp312-musllinux_1_2_aarch64.whl (1.1 MB view details)

Uploaded Jul 29, 2026 CPython 3.12musllinux: musl 1.2+ ARM64

parshred-0.1.5-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.whl (188.1 kB view details)

Uploaded Jul 29, 2026 CPython 3.12manylinux: glibc 2.17+ x86-64

parshred-0.1.5-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.whl (177.4 kB view details)

Uploaded Jul 29, 2026 CPython 3.12manylinux: glibc 2.17+ ARM64

parshred-0.1.5-cp312-cp312-macosx_11_0_arm64.whl (141.1 kB view details)

Uploaded Jul 29, 2026 CPython 3.12macOS 11.0+ ARM64

parshred-0.1.5-cp312-cp312-macosx_10_15_x86_64.whl (154.5 kB view details)

Uploaded Jul 29, 2026 CPython 3.12macOS 10.15+ x86-64

parshred-0.1.5-cp311-cp311-musllinux_1_2_x86_64.whl (1.2 MB view details)

Uploaded Jul 29, 2026 CPython 3.11musllinux: musl 1.2+ x86-64

parshred-0.1.5-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.whl (188.8 kB view details)

Uploaded Jul 29, 2026 CPython 3.11manylinux: glibc 2.17+ x86-64

parshred-0.1.5-cp310-cp310-musllinux_1_2_x86_64.whl (1.2 MB view details)

Uploaded Jul 29, 2026 CPython 3.10musllinux: musl 1.2+ x86-64

parshred-0.1.5-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.whl (188.1 kB view details)

Uploaded Jul 29, 2026 CPython 3.10manylinux: glibc 2.17+ x86-64

parshred-0.1.5-cp39-cp39-win_amd64.whl (343.5 kB view details)

Uploaded Jul 29, 2026 CPython 3.9Windows x86-64

parshred-0.1.5-cp39-cp39-musllinux_1_2_x86_64.whl (1.2 MB view details)

Uploaded Jul 29, 2026 CPython 3.9musllinux: musl 1.2+ x86-64

parshred-0.1.5-cp39-cp39-musllinux_1_2_aarch64.whl (1.1 MB view details)

Uploaded Jul 29, 2026 CPython 3.9musllinux: musl 1.2+ ARM64

parshred-0.1.5-cp39-cp39-manylinux2014_x86_64.manylinux_2_17_x86_64.whl (188.5 kB view details)

Uploaded Jul 29, 2026 CPython 3.9manylinux: glibc 2.17+ x86-64

parshred-0.1.5-cp39-cp39-manylinux2014_aarch64.manylinux_2_17_aarch64.whl (178.0 kB view details)

Uploaded Jul 29, 2026 CPython 3.9manylinux: glibc 2.17+ ARM64

parshred-0.1.5-cp39-cp39-macosx_11_0_arm64.whl (139.4 kB view details)

Uploaded Jul 29, 2026 CPython 3.9macOS 11.0+ ARM64

parshred-0.1.5-cp39-cp39-macosx_10_15_x86_64.whl (152.4 kB view details)

Uploaded Jul 29, 2026 CPython 3.9macOS 10.15+ x86-64

File details

Details for the file parshred-0.1.5.tar.gz.

File metadata

Download URL: parshred-0.1.5.tar.gz
Upload date: Jul 29, 2026
Size: 232.5 kB
Tags: Source
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for parshred-0.1.5.tar.gz
Algorithm	Hash digest
SHA256	`9e7fc0b81a91c7be6eeae32a919cf47ca3b411493fb9a5be7a067a0d84c99c80`
MD5	`8409f651f35072cf929a2b9c4bf14ba5`
BLAKE2b-256	`dfc0b219c94d9d01873286020531c4603de5af779d3028e82ad289010fa4d61b`

Algorithm	Hash digest
SHA256	`c423f0cf3f4e18c59590968279135239e25735a85f12f23dec0b03de1ad209fa`
MD5	`29121647b3051bdd393ae09aa2a63511`
BLAKE2b-256	`c50003b3c3a9e39aa7fdfdc0eeb26d3a39e927d3c5f09e09910299ab36f604a4`

Algorithm	Hash digest
SHA256	`8a7d32c2af8f7afbd04b9c37b5f4e2a7029b4a9bde3a6e8cd0b135c6ba5cc6d0`
MD5	`0ef04da504fc1946c749f3921093fa37`
BLAKE2b-256	`5748f19482b8103c465630989b195eada71f9d357a3bf16ae6822753926b6c8c`

Algorithm	Hash digest
SHA256	`6e639fc9071433369833bd2fc7d52a39317bc6324c2668cc99a4052bda1edab9`
MD5	`80b4b869efbf3fe49861a2d881d5e355`
BLAKE2b-256	`a6e14e1c681b44a2c0e3de816983423b4482f89c25f906d283bb0233df3f834a`

Algorithm	Hash digest
SHA256	`c3ca8c5bdbf923c7a641030e6c8c86048150b44aff78271cad4ee0b75894d90a`
MD5	`25199659fafdd40d42b87a13ce81b260`
BLAKE2b-256	`00fc20bac32b492e821c43945318471fe382580836517f4a099ddbccd6a58c12`

Algorithm	Hash digest
SHA256	`a18d51eff0c738eb9949645410977fd77eb5aaed954bdd4d4f38ced70348bddc`
MD5	`42c5d4c814aa83853380fab2a3ba01a2`
BLAKE2b-256	`a4196a47997715d337b4e612c303f42dba419043a93b796ca563209a569fb08f`

Algorithm	Hash digest
SHA256	`1bf5fa6d97317ed18b8f6e3cae0d05805e88f00817330f34ab4335016cbdf7e5`
MD5	`bc6782dfa0e6fbaf95029da72bc6d356`
BLAKE2b-256	`b048303fa0ed3a3d5a6b6d731f5a4d665c73e1ecdc4c5fb700e3398d3fceb1ba`

Algorithm	Hash digest
SHA256	`3c89a4cbd2574665020582ccbcfc4e38247b4e8ec40ddddc69deaf0b1f4a139a`
MD5	`da7cbc1fdbfc117085ca43b62825e9a5`
BLAKE2b-256	`2a7dfdc61a0c2ea15b6d0ad1fbfd21aa48591e2e3605f95fd6c3d2a8628e4c22`

Algorithm	Hash digest
SHA256	`9aebbc2defbfe6da7c4c19bff7a3dec2c3505a36f83fd50265684189a102abc5`
MD5	`64781624d0af0d3c0b0004fd4a798037`
BLAKE2b-256	`14039294f6bf40a44eec6c077406d269fd47a6470ba54f83038a4ca3583ad448`

Algorithm	Hash digest
SHA256	`9fff7919edf518c02b081b103604abe2171392f71eb532563d76655bb7b7b2bb`
MD5	`41d97ca62c2b052d549420aa5502fee9`
BLAKE2b-256	`2ad49179faca6ebcc3cd7dbad5f5dd66fc1360fe609005d7b8edc503b21487b3`

Algorithm	Hash digest
SHA256	`e2bf5a400d125c8cf16ffb1bf0f2cd6ce61f65226327751396f84124042e900e`
MD5	`4295e6b102cb100a556f9245747b172c`
BLAKE2b-256	`e183699ba6c58bdfe72bdea874b53b55cb5e130b231b93303d2d9abaf5741597`

Algorithm	Hash digest
SHA256	`2c87d9ef7a7992493f9a536ebceaacf98a6cbf143f95ac96cca3338d6f382dd8`
MD5	`0480c3315f2adb7071e7b417135442dd`
BLAKE2b-256	`952de6d05680b9d9e09df9756bc57731675a761b1eb9d5dc26d89e12fb2db503`

Algorithm	Hash digest
SHA256	`f22e9e861fb75145a5c79a45b584898055325d60f1011e26cc03454e4d838fe0`
MD5	`240ef965d8ca54b3d85b1aba370336b9`
BLAKE2b-256	`c67a2e56284263161f82ac8cd05d7bca28d3690178dd938df9a591020eaad235`

Algorithm	Hash digest
SHA256	`83cdfb15decc8f92fb86bdaff87efce479f153907f1afff46dc78649a8cff631`
MD5	`414a7f5b1547e75532f3beaa8ba19132`
BLAKE2b-256	`1f0eec30121b0981e3d487c40282444f7907638af3cd479ea27b20c525d3b491`

Algorithm	Hash digest
SHA256	`855d4d92aaa36d3a6bb80b1a4e0a58d11c8d8fd9083249ae328bee647cfc70be`
MD5	`6370691b23c338b1f707e07074df35e1`
BLAKE2b-256	`48f206fb2c90a8bf553d92891998e05103fe4f08b1ddcc17f3701ffde1cf114a`

Algorithm	Hash digest
SHA256	`7deb7cf08d448618bb816540a9557459865ce4082909a106a64389463d236f63`
MD5	`61da5e6d103008943722f11767bb6f62`
BLAKE2b-256	`ef9c29e931ed34bfe3c453e2b5e2c3348bce3ff5f1c03de00edf6d3698a33951`

Algorithm	Hash digest
SHA256	`c2377a05bc396c926c6f7e1a1e2167b96ed579a023b6da8016e55f1a8e7af980`
MD5	`602409f03e1ec65c5b9b865acaa22b12`
BLAKE2b-256	`be5e19c40e8318c674dae9a1589322f478582a41fef1e730cb4efa487f30d411`

Algorithm	Hash digest
SHA256	`8558d01b7cc92039fd0bbef23efd2071543df78fb717505ec1c6cad05c181fe2`
MD5	`041abee249ab5da49f44aff53a51088f`
BLAKE2b-256	`83afe1c84ea4882f573ab9e3d516f536076ed3c2b805d44e989f4f80e942794b`

Algorithm	Hash digest
SHA256	`d3e94ecaf77d875499713871a240748bfa01e2027ce6fd544806c653cc50f11f`
MD5	`2446eec9e27164021eacf323f8f2888f`
BLAKE2b-256	`d819404f2e7ff05bf32f5a9b6dcc63ccc5bddf80ac95f55eec3ac2208ac76e6b`

Algorithm	Hash digest
SHA256	`316522ba50da8ecdb3de1682ea39dc1e67ce41fc45c2e43c4fa8fd6f44d22480`
MD5	`67b8536b0b556b4ad90feddd3503987a`
BLAKE2b-256	`619f1a2ead49a2ab672aabb0dce5580920c8d8b86bbf045e496fd5c60e278697`

Algorithm	Hash digest
SHA256	`a9f39ebcd71a53e10b0db0f4ae8b8c4e5c9f1fa55060ab8b20f19208fa4495b1`
MD5	`0c4ad3bc9a890ec430a30ab067e28141`
BLAKE2b-256	`ec2282257c2749fbdb413d316e9686b80d423d7420dd9ec0288eee8d5c2e942f`

parshred 0.1.5

Parshred

Feature Highlights

Performance

C++ DOM throughput (x86-64, AVX2)

Python throughput

Quick Start

CMake (FetchContent)

Build from Source

Python

Usage Examples

DOM Parsing (C++)

Pull Parser (C++)

Python

Build XML Programmatically

Documentation

Requirements

License

Contributing

Project links

Key dates

1 maintainer

Credits

GitHub Statistics

License

Requires

Tags

Classifiers

Download files

Source Distribution

Built Distributions

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details