On Mon, 24 Nov 2008, Martin Pitt wrote:
Am I allowed to eddit the environment file in the postinst of
postgresql-plr or what is the general policy for this case.
In principle yes. The creation of that file was pretty much motivated
by things like this in the first place, although primarily intended
for user configuration.
Ah, OK. This sounds reasonable.
However, there are some things to consider:
1. There might be an arbitrary number (including 0) of
/etc/postgresql/MAJOR_VERSION/CLUSTER_NAME/ directories. So if the
semantics of -plr should be to allow using R in all already
existing clusters (which is not unreasonable IMHO), you should add
it to all of them.
In this case MAJOR_VERSION is always 8.3 (or any specific later number)
if I understood upstream correctly. I have probably to loop through
CLUSTER_NAMEs.
2. This change won't have an effect for clusters which are created
after installation of -plr.
These cases should be covered by the included README.Debian.
I see no other chance to copy with this.
3. If you do this, please ensure not to add the environment variable
again if it is already present in the file (including whitespace
fuzzing).
Sure. I would definitely grep for the variable before adding it.
Of those, 2. is probably the trickiest problem.
However, let's step back a bit. To be honest, the mere existance of
this variable seems wrong to me, since it is pretty much an "unbreak
my system" setting and violates Debian Policy 9.9.
Can't -plr itself be patched to default to /usr/lib/R if R_HOME isn't
set? With that, people can still set it to /usr/local/test/R or
whatever, but it would work OOTB.
It seems as if there should be not that much instances of the R_HOME
variable to be changed:
6$ grep -R R_HOME * | grep -v -e ":[[:space:]]*/*\*" -e echo -e printf -e
":[[:space:]]*err" -e README -e '#' -e '^doc'
debian/rules:DEB_MAKE_ENVVARS:=USE_PGXS=1 R_HOME=/usr/lib/R
debian/rules:DEB_MAKE_INSTALL_TARGET:=USE_PGXS=1 R_HOME=/usr/lib/R install
DESTDIR=debian/tmp/
Makefile:ifdef R_HOME
Makefile:r_libdir1x = ${R_HOME}/bin
Makefile:r_libdir2x = ${R_HOME}/lib
Makefile:r_includespec = ${R_HOME}/include
Makefile:R_HOME := $(shell pkg-config --variable=rhome libR)
Makefile:ifneq (,${R_HOME})
Makefile:override CPPFLAGS += -DR_HOME_DEFAULT=\"$(rhomedef)\"
pg_userfuncs.c: unsetenv("R_HOME");
plr.c: r_home = getenv("R_HOME");
plr.c: size_t rh_len = strlen(R_HOME_DEFAULT);
If that isn't possible, I have a thought on how to solve 2. more
elegantly, but let's discuss the general approach first.
As I said - I would feel a proper documentation enough for PL/R - well
release teem does not even consider this package worth to enable our
users a smooth upgrade from Etch to Lenny :-(( so we can expect the few
users to read the docs. But there might be other cases where s solution
for 2. is needed.
Thanks, and "guten Abend",
Kind regards and "gleichfalls"
Andreas.
--
http://fam-tille.de
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
