Re: automake parallel install

[Havoc Pennington]

Tom Tromey <address@hidden> writes:
> >>>>> "Tom" == Tom Tromey <address@hidden> writes:
> 
> Tom> My current thinking is that we would name the installed version
> Tom> and the install directories after the "install version".  For
> Tom> anything in the 1.5 series (1.5.1, 1.5-p1, 1.5c, whatever), this
> Tom> would be "1.5".  Then we would guarantee that for a given
> Tom> "install version" we would only do bug fixes.
> 
> I decided not to do this.  It is simpler and easier to understand if
> we simply always add the version number.

You mean install /usr/bin/automake-1.5.1p5? 

I don't understand quite how it works then.  How do I use a specific
version of automake in my script or spec file?

People can manage a new ABI from time to time (automake-1.4,
automake-1.5) but thinking about every possible automake bugfix
release is kind of overboard.  Are you really going to break
compatibility with every bugfix release?  If not, why shouldn't all
compatible releases be interchangeable? Why would I want to install
10 bugfix-only releases in parallel?

That is, it makes sense to put "automake-1.5" in a script or spec
file, but not to put "automake-1.5.1p5" in there because it'll change
with annoying frequency and the micro version is not relevant, only
the "ABI version." Adding ".1p5" is like bumping the soname without
changing the ABI, resulting in zillions of pointless copies of the
interface. Installing 1.4, 1.5, and 1.6 is pretty sane, installing all
bugfix releases is nuts. ;-)

The idea is that whenever users of a tool expect a specific interface,
they should invoke a command name that represents exactly the set of
tool versions that support that interface, neither more nor less.

BTW I wrote up a general discussion of the parallel install problem:
 http://pobox.com/~hp/parallel.html
This is mostly about libraries though.

Havoc