chore(benchmarks): re-measure the committed datasets and the deck that renders them#466
Open
DemchaAV wants to merge 1 commit into
Open
chore(benchmarks): re-measure the committed datasets and the deck that renders them#466DemchaAV wants to merge 1 commit into
DemchaAV wants to merge 1 commit into
Conversation
…t renders them Both committed result files predated the 2.x line. The comparative snapshot the flagship deck renders had one commit in its entire history, taken on the 1.7.1 codebase, and the perf baseline was captured a week before that — while the deck PDFs built from them were re-committed at every release, so freshly dated assets carried figures from a codebase two minor versions behind. Re-measured on one quiet machine: five current-speed runs aggregated to a median for the baseline, and one comparative run for the deck. The ratios moved, and unlike absolute timings they compare across machines because both libraries are measured in the same run. At 1000 rows the advantage over iText 9 reads 4.54x on time and 3.56x on allocation, where the published snapshot said 5.38x and 4.19x; part of that is iText moving 9.6.0 -> 9.7.0 after the snapshot was taken. Against JasperReports, whose version is unchanged, the gap at 1000 rows is now 1.1%. The deck's own footnotes are corrected while they are being regenerated: one still asked for a refresh "on the final 2.0 commit" two releases after 2.0 shipped, and the summary line still spoke for 2.0 alone. Both now say what the data is — one machine, one run, ratios that travel and timings that do not. Layout snapshot re-recorded for the reworded footnotes; the figures alone did not move it, because every value kept its digit count and the auto-width columns kept their widths.
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.
Why
Both committed benchmark result files predated the 2.x line.
examples/src/main/resources/benchmarks/comparative.json— the file the flagship deck renders — had one commit in its entire history, taken on the 1.7.1 codebase, before the module split.baselines/current-speed-full.jsonwas captured a week earlier. Meanwhile the deck PDFs built from them are re-committed at every release, most recently by the v2.1.0 release commit, so freshly dated assets carried figures from a codebase two minor versions behind. The deck's own slide printed2026-06-15and asked for a refresh before release claims; that never happened across two releases.What
Re-measured on one quiet machine: five
CurrentSpeedBenchmarkruns in thefullprofile aggregated to a median for the baseline (matching how the previous one was captured), and oneComparativeBenchmarkrun for the deck.The ratios moved. Unlike absolute timings these compare across machines, because both libraries are measured in the same run on the same hardware:
Part of the iText movement is the dependency itself: it went 9.6.0 -> 9.7.0 in
4e148a99, after the snapshot was taken. JasperReports is unchanged at 7.0.7, and the advantage over it narrowed too — most at the small end, 2.02x -> 1.74x at 40 rows.Two deck footnotes are corrected while the deck is being regenerated. One still asked to "refresh both data and provenance on the final 2.0 commit before release claims" two releases after 2.0 shipped; the other spoke for 2.0 alone. Both now state what the data is: one machine, one run, ratios that travel between machines and timings that do not.
Tests
Full eight-module reactor
clean verify— BUILD SUCCESS.examplessuite 41/41. The CI performance gate (CurrentSpeedBenchmarkin thesmokeprofile, gate enabled) passes.The layout snapshot is re-recorded for the reworded footnotes only. The figures alone did not move it: every value kept its digit count, so the auto-width columns kept their widths — verified by running the snapshot test against the fresh data before touching any prose. That is luck rather than robustness; a future refresh that crosses a digit boundary will shift the geometry.
Not addressed here
The baseline still records no machine identity — only profile, iteration counts and a timestamp. That is what makes any comparison against it a measurement of the hardware rather than the code, and it is why this refresh does not, by itself, make the file portable. Worth a follow-up that stamps CPU/OS/JDK/commit into the report writer.