Re: RFE: allow for computed version number

[Bob Friesenhahn]

On Fri, 5 Jun 2009, Ralf Wildenhues wrote:
> > Yes.  It assumes the model where anyone who receives the package has the
> > ability to build and maintain it similar to the original maintainer.
>
> But whether the file is distributed in a package tarball or not is
> independent of whether the file is put into the VCS or not.

True.

> Well, this line of thought is actually an argument in favor of putting
> the version in a file, not putting that file in VCS, but in a
> distribution tarball.  That way, the user of the tarball does _not_ need
> the VCS, because the file will be up to date, and `cat version-file`
> will work even where `git describe` will not.

Yes.  Hard-coding something like `git describe` into any 
default/standard Makefile targets is the bane of free software.  For 
example, if a project was to hard code it into the 'dist' target then 
only specially privileged maintainers would be allowed to do 'make 
dist', which runs contrary to the spirit of GNU software.  As long as 
there are appropriate Makefile rules, it is easy to generate a version 
file based on some other information (if desired).

> Again, you are mixing up committing the file with distributing it here.
> As long as the distribution tarball is generated by 'make', it is
> trivially possible to ensure the version file is up to date before
> making the tarball.

By having the Makefile actively query the VCS, or expecting the VCS to 
perform automated text substutions on commit?

Requiring a version file (which may be a target of make) works for me.

Bob
--
Bob Friesenhahn
address@hidden, http://www.simplesystems.org/users/bfriesen/
GraphicsMagick Maintainer,    http://www.GraphicsMagick.org/