Fix naturalsize() rounding rollover at unit boundaries - #329
Merged
hugovk merged 1 commit intoJun 30, 2026
Merged
Conversation
Copy link
Copy Markdown
Sorry, something went wrong.
naturalsize() chooses the suffix from the unrounded byte count via
int(min(log(abs_bytes, base), ...)), then rounds the mantissa with the
format string afterward. When rounding pushes the mantissa up to the base,
the already-chosen suffix is left stale:
>>> naturalsize(999999)
'1000.0 kB' # expected '1.0 MB'
>>> naturalsize(999999999)
'1000.0 MB' # expected '1.0 GB'
>>> naturalsize(1024 ** 2 - 1, binary=True)
'1024.0 KiB' # expected '1.0 MiB'
This is distinct from the ZB->YB top-boundary fix in python-humanize#206 (which added
larger suffixes but did not touch the rounding) and from the metric() fix
in python-humanize#328 (a different function). Here the rollover happens at every small
unit too.
Fix: after rounding, if the mantissa has reached the base and a larger
suffix is available, step up one suffix. Added regression cases to
test_naturalsize (they fail before this change, pass after); all existing
assertions and documented examples are unchanged.
Copy link
Copy Markdown
Codecov Report✅ All modified and coverable lines are covered by tests. @@ Coverage Diff @@
## main #329 +/- ##
=======================================
Coverage 99.55% 99.55%
=======================================
Files 12 12
Lines 900 902 +2
=======================================
+ Hits 896 898 +2
Misses 4 4
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Harness.
|
Sorry, something went wrong.
hugovk
enabled auto-merge (squash)
June 30, 2026 15:34
Copy link
Copy Markdown
Member
|
Thanks! |
Sorry, something went wrong.
patchwright
added a commit
to patchwright/aibug-gate
that referenced
this pull request
Jul 14, 2026
…I integrity gate Three new provenance-pinned rules, each distilled from a *merged* gift-PR fleet fix (stronger provenance than the open-PR rules in AB003-005): - AB006 radix-from-source-not-override [high/default] skorokithakis/shortuuid#115 (merged) — len/radix derived from the original parameter after an override parameter is established as the source of truth; alpha_len = len(alphabet) instead of len(alphabet_index). - AB007 log-exp-rounded-mantissa-no-carry [medium/pedantic] python-humanize/humanize#329 (merged) — magnitude exponent chosen from the unrounded value, then the mantissa formatted with rounding; no carry-guard when rounding rolls the mantissa to the base ('1000.0 kB'). - AB008 magnitude-bin-inclusive-boundary-break [medium/pedantic] mahmoud/boltons#403 (merged) — consecutive-bin loop breaks on '<=' against the next bin's lower bound; bytes2human(1024) -> '1024B'. Corpus now: 8 rules, 5 merged-PR-backed (AB001/AB002/AB006/AB007/AB008), 3 open-PR (AB003/AB004/AB005). All 8 upstream URLs verified against the GitHub API (state + merged flag) at commit time. CI hardening (.github/workflows/ci.yml): - New 'Corpus integrity gate' step runs the corpus-size, URL-format + generator-attribution, and merged>=open invariant tests verbosely as a named gate (not just under the general test step). - Added pedantic-tier self-scan of own source (both tiers must stay clean). - Integrity test strengthened: URL must match ^https://github.com/[^/]+/[^/]+/pull/\d+$; generator field checked against an 11-name denylist (copilot/chatgpt/gpt/claude/gemini/cursor/ codex/codewhisperer/deepseek/llama/mistral). 31 tests green (20 -> 31). Self-scan clean at both default and pedantic tiers. Version bumped to 0.2.0. Optional ast-grep multi-lang port (goal #3) deferred: ast-grep binary is not installed locally, and shipping an unverified .sgy would violate the build-honesty discipline that is the product's premise. Co-Authored-By: claude-flow <ruv@ruv.net>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
naturalsize() picks the suffix from the unrounded byte count (exp = int(min(log(abs_bytes, base), ...))), then rounds the mantissa with the format string afterward. When rounding pushes the mantissa up to the base, the already-chosen suffix is left stale:
999999 bytes is 999.999 kB; "%.1f" rounds that to 1000.0, but the suffix was already fixed at kB, so the output reads 1000.0 kB instead of 1.0 MB. This happens at every decimal, binary, and GNU boundary.
This is distinct from the two related changes already in the repo:
Fix
After rounding, if the mantissa has reached the base (1000 decimal / 1024 binary) and a larger suffix is available, step up one suffix. Values already at the largest suffix (QB/QiB) are unaffected (e.g. the documented naturalsize(10**34 * 3) → '30000.0 QB' still holds), as are all values that don't round across a boundary.
Tests
Added regression assertions to test_naturalsize covering decimal, binary, and GNU boundaries. They fail on the current code and pass with this change; every existing assertion and documented example is unchanged (70 → 76 passing).