Skip to content

Releases

v0.1.3

Choose a tag to compare

@ferinazumaDEV ferinazumaDEV released this 14 Sep 05:40
6598c38

Built from 6598c38308fb9d49676edd23c0c6c2a8d1ea6d81 by run 34810499021, never from a working tree.
Provenance is attested; verify with:

gh attestation verify <file> --repo ferinazumaDEV/webhook-replay

SHA-256

01ad972ce888168fec5118520d244d0428f23c7694c6094276e2738eb842a903  webhook_replay-0.1.3-py3-none-any.whl
97265f39cb21d016e054f46d0b52bb3703beb988370cd2608f49bbba3fad6950  webhook_replay-0.1.3.tar.gz

sbom.cyclonedx.json describes the package and its runtime dependencies as installed from this wheel.

How this release was assembled. The tag and this release were created by hand before the workflow ran, so the run's own "GitHub Release" step failed with "a release with the same tag name already exists" after PyPI had already accepted the upload. Per RELEASING.md the run was not repeated; the assets above are the run's dist artefact — the same bytes PyPI holds — and were attached on 2026-09-14 after checking SHA256SUMS, the attestations, the PyPI digests and the Integrity API provenance by hand.

Changelog

Deprecated

  • Python below 3.11 is compatibility, not support. The floor becomes 3.11 in 0.2.0 (3.10 reaches
    end of life in October 2026). Nothing changes within 0.1.x; CI keeps running on every version the
    package still declares.

Fixed

  • A Content-Length with more than 4300 digits no longer kills the handler. int() refuses
    strings that long with a ValueError that escaped as a traceback on stderr and a connection
    closed without any response (also through Expect: 100-continue). Leading zeros are now dropped
    before conversion, so 000…01 of any length is the one byte the grammar says it is, and a numeral
    of more than 20 significant digits is a 400 malformed Content-Length. That is what RFC 9110 §8.6
    asks for: anticipate very large decimal numerals and prevent integer-conversion errors.
  • A body shorter than its Content-Length is no longer stored as a capture. When the sender
    closed early, the partial body was saved as if complete and answered 200. RFC 9112 §6.3 says
    such a message MUST be treated as incomplete; serve now answers 400 body truncated: N of M bytes received and stores nothing. A sender that stalls instead still gets the 408.
  • --max-body, --since and --header fail as usage errors instead of tracebacks. A size
    past float range (--max-body 1 followed by 400 zeros) raised OverflowError from int(inf); an
    age past timedelta's range (--since 99999999999d) or an ISO timestamp with no UTC form
    (0001-01-01T00:00:00+14:00) raised OverflowError from the date arithmetic; and a header with
    an empty name or a line break in its value (--header ':x', --header $'X-A: v\nInjected: y')
    raised ValueError from http.client in the middle of the replay. All three are now
    ArgumentTypeErrors with a message. Sizes are parsed exactly (no float rounding) and capped at
    2⁶³−1; header names must be RFC 9110 tokens.

Security

  • Redaction covers more credential-bearing header names. X-Auth-Key (Cloudflare),
    Ocp-Apim-Subscription-Key (Azure), X-Forwarded-Authorization, X-Password, X-Credential,
    X-Access-Key, X-Private-Key, X-Shared-Key, X-Client-Key and X-Session-Key were printed in
    clear by show, list --json and curl. The name rule now also matches password, passwd,
    credential, authorization, an auth segment, and the *-key compounds that name a
    credential; Idempotency-Key, X-Request-Id, X-Session-Id and X-Author stay readable.

All of the above were found by tests/test_adversarial.py, a new suite that feeds serve, the
CLI parsers and the redaction rule the inputs a hostile or merely broken sender produces. Every
test in it failed against 0.1.2 before the fix was written.

Archived at Software Heritage

Resolved on 2026-09-14 against the archive, after a Save Code Now visit that followed this tag.

v0.1.2

Choose a tag to compare

@github-actions github-actions released this 13 Sep 20:30
181cc70

See CHANGELOG.md for what changed.

Built from 181cc702bc32d86cc2643f543b980f3ad50c1d7f by this run, never from a working tree.
Provenance is attested; verify with:

gh attestation verify <file> --repo ferinazumaDEV/webhook-replay

SHA-256

fcd5f57fc5fb910190dd1f3c7394b8cd2c6f2945f7024ed485f4c60b3c268b87  webhook_replay-0.1.2-py3-none-any.whl
7da238434a0cf74f1e33a46d70c1cfec2ebd56628573875a2c29cafe9f3627cd  webhook_replay-0.1.2.tar.gz

sbom.cyclonedx.json describes the package and its runtime dependencies as installed from this wheel.

Archived at Software Heritage

Resolved on 2026-09-14 against the archive, after a Save Code Now visit that followed this tag.

v0.1.1 — no longer drops webhooks under a burst

Choose a tag to compare

@ferinazumaDEV ferinazumaDEV released this 06 Sep 16:47
0174151

Upgrade if you are on 0.1.0. That release drops webhooks under a burst.

pip install --upgrade webhook-replay

Fixed: the listen backlog was 5, so bursts were lost

socketserver defaults request_queue_size to 5. The kernel holds five pending connections and refuses or resets everything past that before any handler thread sees it. The server was already threaded and could have served them all — it never got the chance to accept them.

Measured on a 2-core machine, six rounds each, counting failed connections:

burst backlog=5 backlog=SOMAXCONN
20 6 0
60 173 0
150 649 0
300 1448 0

Twenty simultaneous senders already lost data, and a fan-out or a provider working through a retry backlog is exactly that. For a tool whose entire job is not losing captures, this was the failure that mattered.

Fixed with a CaptureServer subclass using request_queue_size = socket.SOMAXCONN — as deep a queue as the OS permits rather than a guessed number; the kernel caps it where its own maximum is smaller.

Added

  • Two regression tests, both checked to fail against the old value: the contract asserted directly, and a 50-sender burst released from a threading.Barrier. The barrier is the point — the pre-existing concurrency test starts its threads one at a time, and that stagger hid the bug locally while it still failed on CI.
  • CI on every push and pull request across Python 3.9–3.13, running the suite the README documents (pytest) and as a module (python -m pytest), because the two resolve imports differently and had diverged.
  • SECURITY.md.
  • A changelog.

Provenance

Built from a clean clone of this tag — not a working tree — with python -m build; twine check passed on both artefacts.

2dd735b9d7fd3eb162717ad8939040a4b4bb1b430863170cf2b43acabf2bccd2  webhook_replay-0.1.1-py3-none-any.whl
9eb689452048420dbc86c8ef16bfb2ae7cf518522a12fc1d00190996d06b5309  webhook_replay-0.1.1.tar.gz

Verified against the published package, not just the repository: installed webhook-replay==0.1.1 from PyPI into a clean venv and fired 100 simultaneous senders at it — 0 failures, 100 captures stored.

webhook-replay v0.1.0

Choose a tag to compare

@ferinazumaDEV ferinazumaDEV released this 05 Sep 01:27
c250457

First tagged release, and the first published to PyPI: pip install webhook-replay.

It ships the state of main after the verified audit of 2026-09-04 — every fix in it was reproduced before being applied and its verification command recorded in the pull request.

See the repository README for what this does and its limits.