On 5/1/19 9:23 am, Dave Reisner wrote:
On Fri, Jan 04, 2019 at 11:20:04AM +1000, Allan McRae wrote:
On 28/12/18 9:15 am, Eli Schwartz wrote:
Our sed parser for xdelta3 headers will greedily match on ":" which coincidentally is also the character we use to define a version with an epoch.
While we are at it, simply use sed for the whole pipeline, rather than using both grep and sed.
Fixes FS#61195
Signed-off-by: Eli Schwartz <eschwartz@archlinux.org>
OK.
As an aside, the fact we are rewriting package version parsing code in bash highlights again that repo-add really should be converted to a libalpm program.
Not sure I understand this -- isn't the problem here the crappy output format of xdelta3? We're not really parsing versions, just contending with the human readable, colon-delimited table that we're forced to use. Rewriting against libalpm won't fix anything here, though if we were to rewrite in C, we could link to the "hidden" xdelta3 library or implement our own crappy vcdiff parser (eww, please no).
Yes - my initial through was this was fixing parsing of versions from nicer output, but after I sent that emails I saw that was confounded with xdelta output "goodness"...
I don't want to open a can of worms (maybe I do), but I think delta handling is the least of the problems with repo-add. I'm aware of many users (myself previously included) that would either replace repo-add entirely, or wrap it substantially in order to make it truly useful. One such example is respose [0]. If we want to consider a rewrite, maybe we should rethink our approach to repo management and make a better tool, not just an equivalent one.
I agree. I was going to come back to the design of the new replacement after the writer had been added to libalpm. I like repose, but I found it does not really work if you wanted more than one repo in a package directory. It is very focused to a single workflow. Saying that, I think we can do better than the current repo-add which is far from great... Allan