Ignacio Vazquez-Abrams wrote:
On Sat, 2008-11-01 at 10:37 -0700, Toshio Kuratomi wrote:
> Also, do we trust mock with unaudited spec files? I know that we do
> trust it with unaudited tarballs but I don't know if this is a reason to
> open things up further.
Doesn't the chroot mitigate most of the issues there might be in the
source package?
It's supposed to but we have had issues in the past where the build
process modified the host environment. I don't know if we traced that
down to something escaping the chroot or if it was something that mock
did before entering the chroot. In either case, if we go for a web
app-only we need to decide whether we're comfortable building unaudited
spec files from someone who may not have a Fedora Account yet (Note: You
presently only need to have a bugzilla account when you submit your
first package. This could be changed to cla_done for use of the
review-o-matic web app) via a web app hosted in Fedora Infrastructure.
If it was a script run on a reviewer's machine this would be something
each reviewer could decide for themselves, possibly after prereviewing a
certain portion of the package.
A VM can probably mitigate the rest.
As in creating and tearing down a xen guest every time a build is
requested? That might help. review-o-matic would need the ability to
do that, though, and Infrastructure needs to decide that they want to
host a web app that has the ability to kick off creation and destruction
of VMs.
-Toshio