The latest MozillaBuild pre-release includes vlad's winrm tool as rm.exe, which is fast at deleting things: $ cp -R debug-mozilla-central debug-mozilla-copy $ cd mozilla-central/ $ time ./mach clobber real 0m18.126s user 0m0.030s sys 0m0.046s $ time rm -rf ../debug-mozilla-copy/ real 0m7.461s user 0m0.015s sys 0m0.015s We should use that if it's available. We could detect it during configure or in mach, whatever: $ rm -h winrm version 0.4 ... $ rm-msys.exe -h rm-msys.exe: invalid option -- h Try `rm-msys.exe --help' for more information.

winrm has been shipping in mozillabuild since version 1.10, released in late July of last year.

bug 1162519 - use winrm for mach clobber on Windows. r?gps

Comment on attachment 8628291 [details] MozReview Request: bug 1162519 - use winrm for mach clobber on Windows. r?gps https://reviewboard.mozilla.org/r/12369/#review10873 Cool. FWIW, I think further optimizing mozfile to reduce redundant stat() could bring its implementation within spitting distance of winrm. We could also leverage some ctypes magic to basically inline winrm into mozfile. But that's all for other bugs.

Here's what I don't understand - we replace rm.exe with winrm as part of the packaging process: http://hg.mozilla.org/mozilla-build/file/050424653d4e/packageit.sh#l24 Shouldn't this already be working without any extra patches?

`mach clobber` uses mozfile.remove(), which does some chmod stuff and then uses shutil.rmtree, all in Python. It doesn't shell out to `rm`.

I'm still seeing long clobber times even with this patch. Ted, can you confirm that things are better for you?

I tested it locally but I haven't been doing regular Windows builds. Are you saying you're seeing long clobber times on your local builds?

Yeah. $ time ./mach clobber real 2m26.866s user 0m0.045s sys 0m0.046s

Does `winrm -h` in your shell produce help text starting with "winrm version 0.4" ?

I think I've got a more general perf issue here. This is a screenshot of shift+deleting my objdir. You can see that it sees some pretty horrible perf dropoffs throughout. The HDD activity indicator is solidly lit the entire time.

My times are from a machine with an SSD, so if you have a spinning disk it's not going to be comparable.

Eh, will fix in the blocking bug.

I have a Crucial m4 512GB SSD. I've tried reinstalling the Rapid Storage drivers in case that helps.