Stuart Henderson <[email protected]> wrote:

> > I don't think we'll be able to make it default, Theo will object.
> > If you want it, you have to convince *him*, not me.

I've encountered a few situations with _mostly offline machines_ where
the rsync package is need, and thus, 1 file can be copied over for
pkg_add.  I don't think I'm the only person to encounter this situation.

My objection was that rsync suddenly needed a pile of dependencies, and
it would take multiple steps to satisfy them all.  A simple procedure
becomes onerous.

> > The problem is not onerousness, but rather Theo being scared of more
> > compression algorithms, which tend to be rife with overflows...

No, my concern was about the flood of hard to discern dependencies, and
a semi-common installation workaround.

> Oh, I guess the commit message was wrong then
> 
> "Disable lz4, zstd and xxhash again
> The dependencies are too big"
> 
> lz4 has a reasonably good track record (not spotless but not rife
> with overflows) and (unlike rsync's internal zlib, which is 1.2.8 plus
> some rsync patch) is now extensively fuzzed 
> https://github.com/lz4/lz4/releases

Maybe capabilities those make sense.  It would be even better if they were
not dependencies, but included as internal.  I don't know how common that
method is.

> I am far more concerned about rsync itself which is already a labyrinth
> of so many options that it's nigh on impossible to get wide test
> coverage, but *shrug* there's no usable alternative for what it does,
> I jail it, what more can I do?

BTW, it would be great if openrsync could be used in such situations
instead.  The first 90% of the work is done in it, but the next 10%
towards making it fully capable turns out to be another 90% task, and I
guess we are waiting for a wave of developers to step in there improve
those issues.

openrsync would satisfy a majority of use cases.  History shows us we
often have bad outcomes when there is only one implementation of a
tool.

Reply via email to