Longstanding issues: structs & standalone Erlang

Douglas Philips dgou@REDACTED
Thu Feb 16 16:17:23 CET 2006


On 2006 Feb 16, at 3:04 AM, Fredrik Thulin indited:
> On Thursday 16 February 2006 01:12, Douglas Philips wrote:
>> On 2006 Feb 15, at 3:12 AM, Fredrik Thulin indited:
>> I don't see how you can avoid that, since you have to run the
>> packaging install commands on each machine, right?
>
> Well, no. It is all about _not_ having to log in to each machine.  
> Either
> you can have something like cron-jobs that fetches all packages you
> want installed from somewhere, and installs them, or you effectively
> execute them from the same set of binaries made available via some  
> kind
> of networked file system. This is what we do - we use AFS.

Yes, I know. AFS was created at CMU. Heh.
And I never said you had to log in to each machine, just that the  
packaging install commands had to run on each machine.
As long as the installation process does not require interaction with  
a human, it can be automated and handled with a cron job.
That doesn't require (though it would be nice if it used) the native  
OS packaging.


> In both these cases, the common part of the problem is the packaging.
> You need to distribute binaries, or instructions about how to generate
> binaries in some way. It is impractical to have to have one way to do
> this for regular programs (like KDE, the Erlang/OTP installation, your
> other favourite applications) and for programs that install "inside"
> any of the others (like CPAN for Perl, or some Erlang way to install
> Erlang applications).

You almost had me there. There is no reason why the "inside Erlang  
way" couldn't in fact call out to the underlying OSes facility.
But it is worse than that, because usually installing new software  
requires special priveleges (root, Administrator, etc.) and that is  
not likely to be available when some user wants to have a new Erlang  
module/application installed RIGHT NOW.

Part of the problem is that there is no concept widely promulgated  
for seperating the OS's packages from the users.


> I am guessing now, but I think lots of people in the Erlang developer
> community more or less only care about the Erlang part of this  
> problem,
> whilst I want to be able to install Erlang applications in the same  
> way
> that I am already installing all the _other_ applications that I
> (really 'we' at Stockholm university) care about. I'm sure this is  
> also
> important to all the UNIX distributions, which package perhaps 4000
> UNIX applications (amongst which Erlang/OTP is one) in a shell/ 
> makefile
> based way already.

Again with the ignoring Windows. Look, I am not unsympathetic to  
ignoring windows, but I think that would disparage a significant  
number of actual Erlang developers.


> So if we want our Erlang applications to be available to Erlang-users,
> then an Erlang-way to install those applications is the right way. If
> we want our Erlang applications to be available to everyone, then a
> more standard way of installing things is the way to go. I firmly
> beleive that this standard way for UNIX systems is "./configure &&  
> make
> && make install", and for Windows it's a single file executable
> installer.

I don't see why we have to choose between the two. Nor do I see why  
we have to do that makes the same solution impractical on Windows.


> The anonymous subversion repo is at svn://anonsvn.it.su.se/buildit/.

Thanks!

		--D'gou




More information about the erlang-questions mailing list