Patches to make GNU gzip and BusyBox gzip produce identical compression results

Daniel Edgecumbe email at
Thu Sep 5 12:48:39 UTC 2019

Well, not just the default behaviour.

It's probably nonsensical to assume that a specific compression level will
produce identical (or even similar) results across different implementations because
there are other possible variables, like the GZIP_FAST flag.

There is though a slight complication in that due to BusyBox being embedded
software we skip some checks. For example, with GZIP_LEVELS off, -1..9 flags are
just ignored transparently. So in that case scripts that rely on the compression
levels will fail anyway.

The patch set was more aimed at fixing some behaviours that I think aren't intuitive.
Specifically I think it's very strange that the default would go from -9 to -6 when
levels are enabled. Much head scratching was had.


On 05/09/2019 14.32, Kang-Che Sung wrote:
> Well, it seems that Denys merged the changes before I have the time to
> respond to comments, but anyway:
> I actually disliked the argument about "de-facto standard" on compression
> levels. I think scripts should not depend on the default compression level,
> and should instead specify it explicitly. The reason is that you can not
> guarantee the 'gzip' on one machine is the same implementation on another
> (there are implementations that use more extensive search, for example,
> 7-zip and zopfli), it is just a coincidence that BusyBox gzip uses an
> implementation that's compatible with the most popular (zlib, I think), but
> it shouldn't be "guaranteed", let alone becoming a "standard" on
> implementaion or on the default behaviors.
> I am not suggesting to change the patch now. I'm just mentioning that
> relying on default behavior of any program, for any script, is a bad idea
> in general. (E.g. Do specify -6 if your script expects it, even though it
> seems redundant, because user setting can always override the program
> defaults.)

Daniel Edgecumbe | esotericnonsense
Kalix NO, Sverige | +358 46 584 2810
email at |

More information about the busybox mailing list