v0.3.0

0.3.0 (2026-06-16)

⚠ BREAKING CHANGES

• store memory as per-repeat series; drop peak_max and the memory alias (#75)
• drop the gross metric and the now-obsolete mode tag (#67)

Features

• --benchmark-memory-profile — keep the memray .bin for regressions (closes #24) (#100) (85fb0fd)
• --free-axes to unmatch faceted plot axes (x|y|both) (#95) (2570988)
• --where row filter for plots (KEY=VALUE, repeatable) (#93) (4a1220b)
• benchmem sweep CLI — cross-version sweeps without a harness (#87) (8230ddb)
• action-scoped absolute memory ceilings via @pytest.mark.benchmem(max_*) (#86) (9afac1f)
• actionable error when memray's Tracker is already active (#89) (bdd24e9)
• adaptive memory sampling instead of a fixed repeat count (#97) (1e79ef1)
• benchmem compare --csv and --sort (#74) (5da39dc)
• bring compare and sweeps onto the combined-table guidelines (#68) (0caf794)
• capture per-repeat memory series and report distributions via --stat (#72) (36f8856)
• let callers label snapshots, decoupling series names from filenames (#57) (10228e1)
• one table for timing + memory (combined by default) (#65) (2f6e8f8)
• rebuild benchmem compare columns on a metric × stat grid (#101) (160e4af)
• rebuild benchmem compare on pytest-benchmark's table model (#98) (761c116)
• show min…max spread whiskers on scaling plots (#99) (a993c98)
• spread varying memory metrics into min/mean/max columns (#76) (01377ef)
• suite-wide memory repeats + column/stat selection (#79) (88251a7)

Bug Fixes

• anchor memory column scaling on the smallest value, not the largest (#92) (aea8a33)
• compare no longer crashes when two runs share a file stem (#64) (60d7e7b)
• hoist byte units into the header in the standalone memory table (#73) (8ad5a73)

Code Refactoring

• drop the gross metric and the now-obsolete mode tag (#67) (90c7045)
• store memory as per-repeat series; drop peak_max and the memory alias (#75) (bc04805)