[racket-dev] Windows test bundle 29M vs. 47M for other platforms?

From: Eli Barzilay (eli at barzilay.org)
Date: Mon Jul 19 19:55:15 EDT 2010

On Jul 19, John Clements wrote:
> On this page
> http://pre.racket-lang.org/release/installers/
> the reported size of the Windows installer is 29M, vs. about 47 or
> 48 M for other platforms.  I see that this was true of the 5.0
> release as well, so this is probably expected, but I'm curious: why
> is this? Is it just that the other platforms are distributed
> uncompressed and rely upon dynamic compression?

* The Windows installers use some LZMA variant, so they have a really
  good compression rate.

* On OSX, there is a flag for the dmg creation utility which does a
  much poorer gzip compression.  On a newer version there is a bzip2
  compression flag too, and I *did* ask about using it -- and several
  OSX users said that it's not a good idea.  You can probably find
  that exchange somewhere in the archives.  (This would be easy to
  change if it's fine now.)

* The linux installers use a gzipped tarball, hence the bad
  compression factor.  It can't rely even on bzip2 to be present since
  we had an installer for Solaris too -- but perhaps it's time to
  require one.

  There's also a new compression utility called "xy" that looks very
  good: it's compressing at rates that are better than the windows
  installer (IIRC, it's some LZMA variant that is optimized somehow),
  and while compression is slower and more memory intensive,
  uncompressing is faster than with bunzip2.  I've looked for some
  statically linked and easy-to-turn-into-an-installer variant, but
  didn't see anything promising.

          ((lambda (x) (x x)) (lambda (x) (x x)))          Eli Barzilay:
                    http://barzilay.org/                   Maze is Life!

Posted on the dev mailing list.