Conversation
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k
Cover all main-event sample sites, zero off-CPU weights under --weight-by-period, keep compare_profiles _ms count-based, define the weighted flag and the Perf events grammar, and make the samply weight logic unit-testable. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k
Record the sample weight mode in the Perf events section, keep the two samply pull requests independent, weight summary rankings, and spell out accumulators, RawMeta.extra, and exact test comparisons. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k
Also settle two spec gaps the plan surfaced: attributes that do not sample get `no sampling`, and a marker period of 0 falls back to the event's fixed period. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k
…n tests Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k
aggregate_grouped's doc paragraph still pointed at the removed event::stack_indices helper and its old Option<usize> yield shape. compare_profiles's delta computations used `weight as i64`; switch to i64::try_from with a saturating fallback. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k
Extract build_listing (asm.rs) and build_output (compare_functions.rs) sync tails so weighted-flag tests use hand-built decoded/listing data instead of depending on the test binary's own layout at a hardcoded address, which isn't guaranteed to decode and doesn't hold on macOS. Also adds a source.rs regression test for reading sample weights by original (not compacted) index across a null-stack sample. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k
Marker events are always weighted by their perf event period, but the samples track only counts events when --weight-by-period is passed. A fixed period (-c N) does not make the two comparable on its own; it was wrong to say it gives exact cross-event comparisons without the flag. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k
…JSLB samply's default output is .jslb.gz, but the only period-weighted end-to-end fixture was JSON. Extend regenerate.sh to also import the recording with the period-support samply build as multi_period.jslb.gz, scrubbed like the other fixtures. Regenerating from a single new recording refreshes every existing fixture, so multi_v49/multi_v75 change too. All four surviving cross-fixture tests still pass. Add a test asserting multi_period.json.gz and multi_period.jslb.gz give identical top_functions output and weighted flags across the samples track and every marker event, modeled on json_and_jslb_from_one_build_match. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k
Tally::add/merge in top_functions.rs already saturate on overflow, but call_tree, folded, stacks_containing, and summary accumulated per-event weights with plain +=. Switch those weight accumulations to saturating_add so every tool follows the same overflow policy. No behavior change for realistic values. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k
antiguru
added this pull request to stack #120
September 26, 2026 13:50
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
1 participant
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.
pollard counted every sample and every perf event marker as 1. In frequency mode (
perf record -F), perf varies the period per sample, so count shares are biased. samply can now record each event's period and name the events in aPerf eventssection ofmeta.extra, with sample weights undersamply import --weight-by-period.pollard parses that section and weights
top_functions,call_tree,stacks_containing,folded_stacks,top_groups,compare_profiles,compare_functions, the source and asm listings, and thesummaryrankings by period. Each of these reports aweightedflag, andcompare_profilesadds a note when only one side is weighted. Totals insummaryanddescribe_profilestay raw counts,compare_profilesmillisecond columns stay count-based, andlist_eventsnames the main event and each event's sampling. Profiles without period information produce the same output as before, apart from the new flags.Stacked on #118. The samply side is mstange/samply#888 (periods and event metadata) and mstange/samply#889 (CPU delta for clock events only). Spec and plan are in
docs/superpowers/.Posted by Claude Code.
🤖 Generated with Claude Code
https://claude.ai/code/session_016aEc8uf7GZbHQHXmoX8o2k