On Nov 9, 2011, at 12:34 , Bill Ross wrote:
> [..] When I started working on Amber in the Kollman lab at UCSF
> (1990), distributions were made by copying a shared tree from DEC's VMS OS,
> that in principle could have had anyone's experimental changes in it. For
> example, earlier while at Berkeley, as a user I had to fix Amber code that
> had an incomplete edit done to it.
>
> Linux had not been heard of. Cray had its own weird OS, and no one could have
> predicted that *nix would 'win' over VMS. In the absence of the standardized
> build setups that we know today, I designed a 2-step install process for most
> unixes and fortrans of the day that called for choosing a 'machine file'
> and typing 'make install'. This was the ancestor of the current setup.
Thanks for the interesting perspective. I for one entered the field in the early 90's when DEC was going strong and duking it out with the other Unix vendors. I do consider myself lucky to have escaped VMS. <ducking>
Back to business - I realize I placed too much emphasis on the autoconf tools. However, the details of the build process itself matter less. Actually, the current Makefile-based approach is transparent and quite idiomatic.
My main confusion arose because the build system implicitly assumes that $AMBERHOME is both the src+build directory and the installation destination, i.e., the place to receive bin/, lib/, include/ and so on. The issue is exacerbated by the fact that the source tree is decidedly not just the result of a simple untar, but rather a combination of licensed downloads, patches, and careful instructions. It is good practice to separate the sources+build trees from the installation destination, among other reasons to provide stability across recompiles. I got that now by managing several copies of the tree. The key bit from the canonical tool chain would be that the installation destination can be chosen at the configure stage with a --prefix=<dir> option.
In short, if there ever is an effort to review the make system, a --prefix=<dir> option would be most welcome.
With best regards,
Michael
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Nov 09 2011 - 13:00:03 PST
