Re: distribute xz-compressed tarballs, too?

[Ryan Schmidt]

On Jun 4, 2010, at 10:38, Bruno Haible wrote:

>> Have you considered distributing xz-compressed tarballs for gettext?
>> They would be about half the size of the gzip-compressed ones:
> 
> True, but it would need 50% more disk space on the ftp server and each
> of its mirrors. 

I would have thought the extra disk space needed on the mirrors would be 
inconsequential, whereas the bandwidth savings of serving smaller files to the 
tens of thousands of downloads you must get would be substantial.


> Then, it requires educating the users about what kind of file this is
> and with which tools they should unpack it:
> 
>  - The 'file' command from not too long ago does not know about this format:
>    $ file gettext-0.18.1.*
>    gettext-0.18.1.tar.gz: gzip compressed data, from Unix, last modified: Fri 
> Jun  4 15:02:42 2010, max compression
>    gettext-0.18.1.tar.xz: data
> 
>  - The 'tar' program knows about this format only since version 1.22.

Perhaps there is a small support burden there. But certainly other projects are 
providing their source as xz files today, presumably with some success.


> The benefit would be marginal:
>  - You get 500 GB disks for about $ 100 or less nowadays.
>  - Most users have a bandwidth of 100 KB/sec or more. It does not really
>    matter whether a file arrives in 150 sec or 75 sec.
>  - Disk and network hardware is now frequently used for movies, which
>    has file sizes starting at 10 MB and going up to 5 GB.
>  - Distributors are free to repack the tarballs (in fact, Mandriva used
>    .bz2 for everything some time ago).

The primary objection that occurs to me is that the software used to decompress 
xz files itself uses gettext. At least it does in MacPorts for example: 
xz-devel is the software we use to extract xz files, and it depends on gettext, 
therefore gettext cannot have a dependency on xz-devel. Therefore a package 
manager must decide between offering a smaller download of gettext, or 
providing an unlocalized xz.