Re: [TCLCORE] TIP #291: Add the 'platform' Package to Tcl

[Damon C. <da...@tc...>]

    Wait a minute.  How is it that the 'platform' package is not able to 
identify a new platform exactly?  As I understand it, it probably works 
pretty similar to the package I wrote for my own stuff.  We look at 
tcl_platform to pull some information and then twiddle it to make it 
pretty.  IE:

tcl_platform(byteOrder) = bigEndian
tcl_platform(machine) = sun4u
tcl_platform(os) = SunOS
tcl_platform(osVersion) = 5.6
tcl_platform(platform) = unix
tcl_platform(user) = lwv26

    Becomes Solaris-sparc.  Why?  Because SunOS-sun4u looks stupid 
(thanks, Sun).  So, what would platform::identifier do if it ran into 
Solaris and we'd never seen it before?  You'd get SunOS-sun4u since the 
default is most likely to just return 
$tcl_platform(os)-$tcl_platform(machine).

    What does a repository do when it doesn't recognize that OS?  Well, 
one of three things:

1.  Let's say the repository really DOES have packages for Solaris, but 
it thinks the platform is supposed to be Solaris-sparc since that's what 
it is in the newest 'platform' package, but the user doesn't have that 
yet.  The repository code could just as easily do a quick map since it 
knows SunOS-sun4u is really Solaris-sparc and install the correct 
binaries.  Might even update the platform package while we're there.

2.  Let's say the repository doesn't know what the hell you're talking 
about.  The package installer could try and grab the 'platform' package 
first before grabbing other packages to make sure it had the latest.  
This would all be done on the package install side, not the repository.  
It's pure-Tcl, and it just uses tcl_platform, so there's no real risk of 
it being incompatible with the version they're running.

3.  Screw it.  I really don't have binaries for that platform, so you 
asking for them does no good anyway.

    This isn't that tough.  Every platform, regardless of whether we've 
heard of it or not, is going to give us SOMETHING in tcl_platform.  And 
if the 'platform' package has never heard of it, we're going to get a 
default identifier.  Maybe it's the right one, maybe it's not.  Either 
way, if we've never heard of it, chances are good it's not in our 
repository anyway.

Damon


> Hmm, there's a counterargument to that, though.  As Joe points out,
> the Core platform package is always likely to be obsolete on a
> given installation.  As such, it's likely to be the first thing
> downloaded from a package repository.  And if it *is* written in
> Tcl, then it oughtn't to need anything extra to run from the
> repository.
>
> Nevertheless, it feels strange to have something in the repository
> that is a dependency of virtually every package there . The natural
> question would arise, "why isn't this in the Core?" 
>
> In short, there are forces pulling 'platform' into the Core,
> and forces keeping it out. Perhaps the only real resolution 
> of this approach-avoidance conflict would be to get off our arses 
> and sort out "bundled packages."  That way, the Core installation could 
> come with at least *some* platform package, and those that need to (which 
> might indeed be most of us) could load an updated version from the \
> repository.
>
>