btrfs machine readable output [was Re: btrfs patches]

[Goffredo Baroncelli <kreijack@inwind.it>]

Hi Axel,

I add CC:linux-btrfs because I think that what you are saying is important and it was discussed in ML.

[we are talking about the Axel patches to "btrfs sub list" command]
On 2015-10-05 19:27, Axel Burri wrote:
> Hi Goffredo
> 
> Don't get me wrong, I don't want to blame you (or anyone), actually I'm
> only arguing what I think is best for btrfs. I started to write those
> patches because btrfs-scriptability is pretty bad (there is also "btrfs
> sub show" which has pretty un-parseable output"), and I think this is
> very important for the success of btrfs, and no-one seemed to address this.

In the past, when I developed the "btrfs fi usage/btrfs dev usage" I put a lot of effort to have an output easily machine readable: one data per line, key without space; I remember the battle to have in the key "_" instead of space :-)
However other people changed that to a more (human) readable output.

May be that I was wrong; I don't know. 

However for sure, if I look to other projects:
- systemd: it uses some btrfs capability, but it uses the ioctl directly.
- snapper: it uses ioctl directly

So I have to conclude that it is not common to use btrfs (as progs) directly.
Few times I noticed somebodies are working on a libbtrfs. May be this is a better approach.

> 
> I know that nowadays people expect to get shiny output from command-line
> tools, but IMHO it is much more important to have consistent
> machine-readable output before even thinking of human readability (since
> there are so many unix commands to translate it, e.g. "columns"). Now
> this is sadly not the case with btrfs-progs, and it's hard to change it
> when people are used to it. Note that when writing the patches I was
> also thinking of introducing some "--machine-readable" option, but this
> seemed so wrong to me...

I think that "--machine-readable" (or similar) is a better approach: the output of the btrfs command in the past is changed even drastically (give a look to my work on the mkfs.btrfs); but we could force us to have an output "backward compatible" and machine (programmer) friendly if a switch like "--machine-readable" is used.

[...]
> 
> Regards,
> 
> - Axel

BR
Goffredo


-- 
gpg @keyserver.linux.it: Goffredo Baroncelli <kreijackATinwind.it>
Key fingerprint BBF5 1610 0B64 DAC6 5F7D  17B2 0EDA 9B37 8B82 E0B5