../

svabench Implementation moc

2026-08-14T02:05:49

  • GLIBC is the bane of my existence
  • claude_gen -> post_proc -> jasper_eval -> aggregate_results
  • I want to split up the entire benchmark process into these smaller programs
  • This was things can be asynchronous and I can reuse these

2026-08-03T11:22:10

  • assert_decl_from_sva
  • Deciding between stdin and -f FILE
    • /dev/stdin is not portable. I cannot use std::filesystem::path.
    • read() is portable. I can read both std::cin and std::ifstream
    • SourceManager needs to be used I think

2026-08-01T10:51:46

2026-07-31T11:56:05

  • Moved the uuid logic to small_bench
  • This seems to be the way to go
  • claude_infer tries to mimic an API call and thats it

2026-07-31T11:08:37

  • In order for me use the usage tracker, I need wrestle with bash’s limitations
    • claude_infer, which is the function that wraps claude with the appropriate options, outputs the .sva files
    • I cannot output any diagnostic information with this
    • I can send in an uuid as the third arg and I can retrieve the session files from the outside. But then I may not be able to clear them before the next command.
    • claude_infer just has to do the inferring then, I have to handle all the other things outside
    • claude_infer’s purpose is to just run claude with the given model, prompt and given uuid

2026-07-31T04:46:13