[M3devel] ROOT

[Tony Hosking]

On 2 Jul 2009, at 15:47, Jay wrote:

> I half agree.

;-)

> It'll be a few hours/days, maybe a week, but I'll take a stab at  
> only putting the files in lib.
> I don't know if it'll be easy or not but I'll try.
> Therefore no hardlinks.

Cool!

> But still $origin.

I think so, but still not sure.

>  Unless maybe there is consensus that install must be /usr/local/ 
> cm3, unless you build/link yourself, then you can chose.

My inclination is that a binary install (no build/link) is OK to be  
hardwired, but if $origin gives flexibility in that then perhaps worth  
supporting.  But surely, for binary installs we will be using package  
managers to install to standard places so no need to adjust the paths?

>  And the distribution building will have to be sure to...um...? Be  
> done as root and impact the running system??? No..
>  Well, distribution building can make it work somehow, by using - 
> rpath /usr/local/cm3/lib, even if ld is pointed at /tmp/cm3/lib/ 
> libfoo.so. I can look into that, if there is actually firm consensus  
> against $origin and for full paths, and for taking away user choice  
> of install location...but...but....what about non-admin installs?  
> They have to rebuild? As a non-admin installer on cm3, I can  
> probably live with that, but not sure about others..seems not great.  
> Another option is to link upon install, or "fixup" the paths on  
> systems that can do that without relinking.
>    Again I come back to favoring origin pretty strongly.

OK, now I think I understand where you are coming from.  You want  
binary installs (no build/link) for non-admin users to non-standard  
places.  Seems like a desirable thing in theory.

>  Hard links seem perfectly migratable..they tar and untar at least..
>  I get it probably though -- what flags to use to cp? There are too  
> many options, it confuses me.
>   I know hard links can't cross file systems, but often people only  
> hardlink "nearby" files that are the same file system anyway.
>   Hardlinks don't survive on Windows FAT filesystem, probably nobody  
> cares. They work fine on NTFS.
>   Some copy utilities might break them though, yeah, that happens.

I generally avoid them because of the swamp you just described.

>  Full paths not using $origin aren't migratable.

Right.

>
>  - Jay
>
>
> From: hosking at cs.purdue.edu
> To: rcolebur at scires.com
> Date: Thu, 2 Jul 2009 15:37:45 -0400
> CC: m3devel at elegosoft.com
> Subject: Re: [M3devel] ROOT
>
> I think best would be for Jay to reprise his thinking on why all of  
> this was needed.  In general, I oppose hard links on the grounds  
> that they are not trivially migrateable.  Better to have relative  
> symbolic links.  If we can avoid the need for $origin by installing  
> libraries in a fixed INSTALL_ROOT/lib directory (rather than the  
> current symbolic links approach) then I prefer that.
>
> Antony Hosking | Associate Professor | Computer Science | Purdue  
> University
> 305 N. University Street | West Lafayette | IN 47907 | USA
> Office +1 765 494 6001 | Mobile +1 765 427 5484
>
>
>
>
> On 2 Jul 2009, at 14:31, Randy Coleburn wrote:
>
> I keep watching the various commit logs et al and I'm concerned too  
> that I don't readily understand what is going on and what the new  
> requirements will be going forward in terms of environment vars,  
> paths, and config file requirements, etc.
>
> As for ROOT, as an environment var, this is too generic.  If it is  
> required, it should be renamed to be specific, e.g. CM3_ROOT.
>
> Would it be possible to have a online conference about all this?
>
> Regards,
> Randy Coleburn
>
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://m3lists.elegosoft.com/pipermail/m3devel/attachments/20090702/da796b61/attachment-0002.html>