On Sat, 26 Nov 2016 at 12:54:12 +0000, Johannes Schauer wrote:
> Quoting Adam D. Barratt (2016-11-26 09:40:17)
> > On Sat, 2016-11-26 at 03:34 +0000, Johannes Schauer wrote:
> > > + None of the required targets must attempt to write outside of
> > > the
> >
> > the current wording means "None of the
> > required targets is required to attempt".
>
> I do not understand why what I wrote is equal to [that].
> Could you elaborate? I was not able to find the reason myself.
"None of [set X] must [action y]" can be read as "there is no x in X
for which x must y" - a statement about the absence of a requirement.
What you were trying to say was "it must be true that there is no x
in X for which x *does* y" - a requirement.
(In this case X = the required targets, and y = write outside
{ build directory, /tmp })
> Most notably, the
> required targets must not attempt to write into <file>$HOME</file>.
I would prefer this to be phrased more like:
Most notably, the required targets must not attempt to write to the
building user's home directory
because otherwise, it could be interpreted as forbidding packages in which
the build *does* write to $HOME, but to prevent this from causing the
problems you described, the package's debian/rules does something like
"export HOME = $(CURDIR)/debian/tmp-home" so that the real home directory
remains unaffected.
glib2.0 is one such package: it has build-time regression tests that
write to $HOME, but its debian/rules sets $HOME to a harmless value to
avoid failing on buildds or violating the spirit of the proposed policy.
I would like to make sure this also does not violate the letter of the
policy.
S
