Re: Is strip-nondeterminism causing performance regressions in your packages?
- Date: Wed, 4 Oct 2017 14:11:18 +0100
- From: Simon McVittie <smcv@xxxxxxxxxx>
- Subject: Re: Is strip-nondeterminism causing performance regressions in your packages?
On Wed, 04 Oct 2017 at 12:43:59 +0100, Chris Lamb wrote:
> Any which we do not ship compiled files cannot be affected (eg. we do not
> ship Python's .pyc files; they are generated at installation-time)
This is not generally the case for distros other than Debian, so is
worth being aware of for Python. I don't know what the rationale is for
byte-compiling in maintainer scripts rather than during dpkg-buildpackage
- perhaps it's because we want to support many Python versions in parallel
and we can't do Arch:all binNMUs yet?
libostree normalizes timestamps to (time_t) 0 to maximize identical file
reuse between OS snapshots and avoid timestamp skew, and Flatpak apps
and runtimes generally include (only one Python version and) pre-compiled
.pyc files, resulting in either libostree or Flatpak (I forget which one)
needing a special-case hack for .pyc files which sets them to
(time_t) 1 instead.