On Sun, Nov 28, 2010 at 09:29:28PM +0000, Glynn Clements wrote:
Hamish wrote:
> > > If the only available packages (RPM, .deb, etc) insist
> > > upon installing GUI libraries, complain to the people who
> > > make the packages.
>
> Paolo:
> > OK, so, now I'm complaining
: packagers, please have
> > your saying here.
>
> disk space is very cheap. package maintaining time is not. IMHO
> unused libraries & a couple extra packages do no real harm
It isn't "a couple". Once you link against wx, you're looking at ~60
extra libraries.
I'd suggest that the core GRASS package shouldn't list X, wxWidgets or
(in 6.x) Python as dependencies.
It shouldn't be necessary to split the actual GRASS distribution into
GUI and non-GUI components. Put everything in the non-GUI package, and
have a separate GUI package which is empty (or contains a single dummy
file if the packaging system doesn't allow empty packages) and exists
solely to hold the GUI-specific dependencies.
As said in another mail of mine, it is not a problem brute-splitting GUI
dependencies (e.g. current squeeze release for 6.4 only *suggests*
wxpython dependencies). Moving GUI-related binaries in different
packages is a pain to maintain, but theoretically it can also be done.
It remains a brute hack anyway, which is a symptom of
a fundamental design problem: the whole system is theoretically
layered and modular, but in fact it cannot be really componentized
in a *clean* way. Something I find basically disturbing and not
elegant, sorry my purism.
BTW, I tend to disagree about considering GUI maintainance not influent
in the releasing cycle. It *is* influent and caused many of the
past/present windows/macos delays. Having a sort of grass-toolbox
and grass-gui sub-projects could help a lot (like CNES does
with OTB and Monteverdi).
And like it or not, there are people that do not use the
default GUI or a GUI at all. Splitting would gain consensus
about the core, which is IMVHO the true value of GRASS.
Of course, my 2 cents, as always.
--
Francesco P. Lovergine