[Slackbuilds-users] txz?

[Yth O’Gtha]

So, for the sake of gathering more data, I repackaged all my current
slackbuilds, added times, and got global size.
This is for 293 packages, and "txz0" stands for `xz -0`

`memory` column is a bit devious: it is the sum of resident memory usage for all
runs. Correlated with the maximum usage `maxmem` it can give a global feeling
about whether a compressor requires a lot of memory, or not.
`mean` is just `memory/293` so an average per build, for what its worth.

And I added a line for tar.bz3, doing `tar | bzip3 -j 8`, for 8 cores
It isn't handled by pkgtools, hence it is only for getting some insight into
this new compressor, which proves quite interesting.
Documentation doesn't show a way to « just use as many threads as available
cores », and there is a block parameter which adds more memory, more time, and
reduces size, and alas more decompression memory required, so it may not be
interesting to try to maximize that. Results are shown on the second tbz3 line,
using maximum of `-b 511`, and I'm not convinced, contrary to eg `xv -9`.

|      | real  | user  | sys  | size  | memory | maxmem | mean |
| tgz  | 1316s | 1265s |  72s | 8620M |   6G   |  37M   |  22M |
| tbz  |  239s | 1202s |  89s | 8007M |  12G   | 148M   |  43M |
| txz  | 2843s | 6456s | 139s | 6580M | 104G   |  12G   | 364M |
| txz0 |  278s | 1654s |  83s | 7886M |  10G   |  91M   |  34M |
| tbz3 |  310s | 1510s |  30s | 7135M |  29G   | 881M   | 103M |
| tbz3 |  904s | 1307s |  53s | 7008M |  79G   |  19G   | 273M |

I am overall pretty impressed with lbzip2, though `xz -0` looks slightly
better, a tad bit less memory, a tad bit more cpu and time, and a tad bit less
space.
And there is no question that standard xz is best for large-scale distribution.

-- 
Arnaud <yth at ythogtha.org>