Sean Whitton writes ("Bug#935762: dgit: behaviour of --quilt=auto is surprising
when a quilt series exists"):
> On Sun 25 Aug 2019 at 07:52PM -03, David Bremner wrote:
> We talked about this a bit on IRC and it looks like what happened was
> that David used --quilt=auto in repos which were patches-unapplied, so
> dgit cannot reasonably be expected to have done the right thing.
The result of that would be that dgit would make a .dsc with all the
upstream changes reverted. (Probably by making a new patch reverting
them all.)
This is not very good and we should definitely try to avoid it, at
least by changing the docs.
I don't think we can avoid it programmatically because this situation
is not readily distinguishable from
dgit clone foo
git revert <each one of the patches>
dgit push-source --quilt=auto
where what dgit does is what the user needs.
> Maybe --quilt=auto, --quilt=linear etc. should be explicitly labelled as
> for patches-applied trees in the docs?
That might help. David, what did you expect --quilt=auto to do ?
I ask because that may help us understand how to change the docs.
ISTM that the docs for the non-split-view quilt modes all need to be
rethought now that many people are using split view quilt modes. The
non-split-view quilt fixups all predate split view and the docs talk
only about "fixing up source format `3.0 (quilt)' metadata".
What "fixing up source format `3.0 (quilt)' metadata" mostly means is
"fixing up the situation where the user started with dgit clone and
made some commits".
And, TBH I think --quilt=auto is not really useful enough to deserve
that name. Maybe it should be renamed to --quilt=try-linear or
--quilt=linear+smash or something. "auto" sounds like it
automatically selects the right quilt mode which it very much doesn't
(and which isn't possible, sadly).
Ian.
--
Ian Jackson <ijack...@chiark.greenend.org.uk> These opinions are my own.
If I emailed you from an address @fyvzl.net or @evade.org.uk, that is
a private address which bypasses my fierce spamfilter.
