On 30 May 2005 at 15:02, Chris Lawrence wrote:
| On 5/30/05, Adrian Bunk <[EMAIL PROTECTED]> wrote:
| > reopen 307683
| > tags 307683 -sarge
| > severity 307683 serious
| > thanks
| >
| > On i386, both the package in unstable and the package in unstable fail
| > if mawk is not installed.
| >
| > If you use mawk, you have to add a build dependency.
|
| It seems to me that the proper place to fix this is (still) in
| r-base-dev, since it's what's choosing mawk or gawk (and I'd have to
Don't think so. Mawk is priority required. It can be assumed to be present.
| guess which mawk|gawk was being used on which architecture to fix it
| in this particular package - not to mention I suspect r-base-mapdata
AFAICT you have a particular package with a particular problem due to
specific needs for an AWK interpreter than are satisfied by only one of mawk
or gawk.
The solution, as far as I can tell, is to depend on the variant you require,
and then to overrride the all-encompassing default set in R itself --- R
doesn't know some packages need a special one, so it defaults to what it
finds at built-time: mawk:
[EMAIL PROTECTED]:~> grep -i awk /etc/R/Renviron
AWK=${AWK-'mawk'}
If that is not good enough for a given package, shouldn't that package be fixed?
| has the same problem). In short, r-base-dev needs to depend on the
| awk used when it was built -- the simplest solution I can see is to
| use mawk on all architectures (overriding configure's detection on
| amd64 and possibly other architectures) and having a fixed dependency
| on mawk.
I could specify something like AWK=mawk at compile-time, build-depend and
depend in r-base-dev. Seems heavy-handed to me, though.
Dirk
--
Statistics: The (futile) attempt to offer certainty about uncertainty.
-- Roger Koenker, 'Dictionary of Received Ideas of Statistics'
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
