On May 3, 2012, at 11:27 AM, Andreas Dilger wrote: > On 2012-05-03, at 6:10 AM, Pavel Raiskup wrote: >>> Something like that looks reasonable. >>> One major comment, though, is that the output of >>> tar should be parsable, i.e., it should be possible >>> to look at the output, figure out what format it is, >>> and parse it reliably. The current format doesn't have >>> that property, but these extended formats ought to, no? >> >> I'm not sure if I follow you correctly -- you mean some 'universal' parser >> that gets tar's -t output and parses it alway correctly and gets as much >> info as possible? >> >> IIUC -- In previous proposal it is not so easy to achieve this goal in >> case that there is the 'vv' (not 'vvv') output requested. There may occur >> some ambiguity between --xattrs lines and --acls "added" lines, yes? >> >> .. other possibility is to add two bytes -- 'CHARACTER COLON' before each >> additional line. Extended attributes e.g. "x:" for --xattrs, "a:" for >> --acls and "s:" for security context. Again -- the 'vvv' is not needed >> now and even its header can go out. Or .. >> >> .. do you (or other listener of this channel) see even other better >> solution? > > If you are looking for an output format that is relatively easy for humans to > read, but is still well formatted and machine parseable, please consider > using YAML. This is a colon-delimited text format, similar to what you > propose above, but has a few more features. There are YAML parsers for > virtually every language, so should be considered as the first choice of > machine-readable textual output, rather than inventing some new format.
YAML would be reasonable, or JSON or XML. I'm not very particular. I would like to see a written proposal for such a format. If it looks reasonable, we could easily add it to bsdtar as well. Cheers, Tim
