[Commented On] D11616: rhg: stop manifest traversal when no more files are needed

aalekseyev (Arseniy Alekseyev) phabricator at mercurial-scm.org
Tue Oct 12 12:34:27 UTC 2021


aalekseyev added a comment.


  In D11616#178125 <https://phab.mercurial-scm.org/D11616#178125>, @martinvonz wrote:
  
  > I wonder why we see so different results. Are those the times reported by `hyperfine` in your case?
  
  No, I collected the times with a custom OCaml script, but I've been seeing similar results when measuring by hand with `time`.
  I haven't tried using `hyperfine`.
  
  > rustc version: rustc 1.57.0-nightly (8f8092cc3 2021-09-28)
  > CPU: Intel(R) Xeon(R) Gold 6154 CPU @ 3.00GHz
  > OS: Debian
  
  I've got:
  
  rustc 1.55.0 (Red Hat 1.55.0-1.el7)
  CPU: AMD EPYC 7702P
  OS: CentOS 7
  
  I found one more difference: my all-moz.build benchmark is only concatenating 42 files, instead of 1897, because my glob was not recursive. Whoops.
  Correcting for that, I'm still seeing a 15-20ms advantage of merge-based algorithm over a hashtable-based. 
  The absolute time is 220ms vs 240ms, by the way, which is surprisingly far off from 600+ms you're seeing.

REPOSITORY
  rHG Mercurial

CHANGES SINCE LAST ACTION
  https://phab.mercurial-scm.org/D11616/new/

REVISION DETAIL
  https://phab.mercurial-scm.org/D11616

To: aalekseyev, #hg-reviewers, martinvonz
Cc: Alphare, SimonSapin, martinvonz, mercurial-patches
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mercurial-scm.org/pipermail/mercurial-patches/attachments/20211012/1dd5387a/attachment-0002.html>


More information about the Mercurial-patches mailing list