Re: GNUstep installation, was Compiling from scratch.

[Richard Frith-Macdonald]

On Saturday, October 25, 2003, at 02:10 PM, Ian Jones wrote:

> Well, I sent one patch yesterday for -base configure.ac which 
> hopefully now will try to use ffi if it can't find ffcall. I haven't 
> looked at compile-all yet but I'm willing to do that if other people 
> are busy elsewhere or see other things being more important.

I already applied a very similar fix to the one Pascal produced to fix 
the problem with demanding that you be root to run the script,
but I haven't addressed the issue of stopping (and printing a useful 
message) when configuring/building/installing part of the system fails 
... I was planning to do that today, but I'd much rather leave it to an 
expert perl/shell programmer!  Please go ahead.

> I would very much like to see gnustep easy to install and when 
> installed working reasonably well, which may mean exiting the build 
> from compile-all as Chris Vetter has previously stated if it can't 
> find critical depandancies, such as ffcall or ffi not being installed, 
> with clear and understandable explanation as to what has gone wrong. I 
> would also like to make available a script that would run prior to 
> compile-all which would basically do a CVS checkout of Alex Malmbergs 
> latest_semi_stable tag; this would provide a gnustep installation 
> which would be kept reasonably up to date and tested probably by 
> myself and or whoever else would like to get involved to make sure it 
> is kept to a reasonable quality. Hopefully this will provide 
> developers and users who wish to give GNUstep a go a much better 
> experience and impression of the project as a whole and through this 
> we might gain a little extra man power.

I like the idea of sticking to a semi-stable tag ... I don't think we 
should be encouraging naive users to use the latest (unstable) cvs 
code, but we do need to ensure that they are kept fairly up to date.  
Alex manages a very good compromise position.

> Some things that have been brought up by other and ideas of my own for 
> compile-all:
> 1) Ability to choose between a user installation and developer 
> installation, difference being development tools

The compile-all is for the core ... perhaps another one which would 
include PC and Gorm wqould be good.

> 2) Ability to choose which backend you want to use x11/art (obviously 
> providing a default for people who don't know what's what)
> 3) exiting with good explanation after compilation of -base if 
> critical depandancys aren't met that would make gnustep install but 
> work with very limited functionality if at all.

I think clearer (to a naive user) messages from the base configure 
scripts would be good.  Then the compile-all could refer people to the 
end of the configuration log.  That way the work done in coming up with 
clear error messages would be of use whether people were building using 
the compile-all script or using the recommended procedures documented 
by Dennis and others.