Averaging no longers shortens the output CSV to the minimum CSV length. Instead, stat frame values are averaged across the CSVs we have frames for. This allows for meaningful comparisons of large sets of CSVs
TODO: process the CSVs one at a time to reduce peak memory use
#ROBOMERGE-SOURCE: CL 8069626 via CL 8095248
#ROBOMERGE-BOT: (v401-8057353)
[CL 8095363 by ben woodhouse in Main branch]
Bulk query improvements:
- Add -beginEvent and -endEvent arguments to generate perf reports only within particular events (e.g skydiving begin/skydiving end)
- Add -summaryTableFilename arg so we can override the summary table filename
[FYI] andrew.ladenberger
#ROBOMERGE-SOURCE: CL 7984408 via CL 7996448
#ROBOMERGE-BOT: (v396-7974030)
[CL 7996657 by ben woodhouse in Main branch]
- Fix an exception when a CSV has no metadata (e.g LLM)
- Improve peak summary. Add support for the <stats> element (no longer need to rely on stats being referenced in graphs)
- Add support for wildcards, e.g <stats>*</stats>
- Remove duplicate stats automatically
- Refactor peak summary to avoid separate arrays for stat attributes - everything is now wrapped up in a PeakStatInfo struct
- Modify CsvStats csv parsing to skip stats with only spaces in the name (for some reason LLM does this)
[FYI] ben.salem
#ROBOMERGE-SOURCE: CL 7669228 via CL 7673482
#ROBOMERGE-BOT: (v380-7672804)
[CL 7673532 by ben woodhouse in Main branch]
A nicer solution would be to strip commas at source in the CSV profiler, but that won't work with existing data
[FYI] chris.bunner
#ROBOMERGE-SOURCE: CL 7650431 via CL 7654000
#ROBOMERGE-BOT: (v372-7473910)
[CL 7654072 by ben woodhouse in Main branch]
