On Mon, 19 Sep 2005, Norman Gray wrote:
>
> Quite separately, I've drafted a paper for Software Practice and Experience
> on the build system. It's a late-ish draft, but I've decided to leave it for
> a few days before returning for a proper read through. It's currently at
> <http://www.astro.gla.ac.uk/users/norman/temp/star-cvs-1.6.pdf>: I'm not
> looking for anything detailed, but if anyone would like to glance through it
> and make observations of the `too long', `too boring', `I want more detail'
> class, I'd be most appreciative.
Looks excellent on a first pass.
* Is this the intended author list?
* The OSX Tiger problems are I believe down to the -nostdlib in
conjunction with g95 (and the original lack of -single_module).
* Mention of preprocessing probably should mention that lacking native fpp
support in conjunction with a case insensitive file system causes problems
since we use a .F extension.
* We don't actually use ESO catlib so it should really be removed from
thirdparty. We should probably add ncurses to thirdparty with an existence
check (as I found when building 64bit on Mac)
* I have encountered problems in SURF with fortran preprocessing since I
was using a variable named PACKAGE. I had to undefine the FPP PACKAGE
symbol for it to build properly.
* The restFP situation is much better on Tiger with g95 *but* the fact
that the g95 runtime library contains MAIN_ causes problems when linking
C programs with fortran libraries (hence all the dummy MAIN_ and fortran
runtime initialisation shenanigans).
* We have had problems with non-ANSI C (NBS and IDI) with their lack of
void* and prototypes.
* 'phillipics'? Not on dictionary.com
* 'acuquianting'
* Did you want to mention the "tiger team"? I think that week was
essential in giving the work a boost. It really broke the back of the port
and was
--
Tim Jenness
JAC software
http://www.jach.hawaii.edu/~timj
|