Roland Mainz wrote:
> Alan Coopersmith wrote:
>
>> Note that this will move the default location used for files
>> created with tempnam() etal. from the on-disk /var/tmp to the
>> tmpfs /tmp. Does anything depend on being able to access
>> such files across reboots (like editor autosave/recovery) ?
>>
>
> AFAIK no since APIs like |tmpfile()| and |tmpnam()| are for temporary
> data only and they should not be used for permanent storage (e.g. AFAIK
> it is a bug when applications would use these APIs for storing permanent
> data).
>
I'm not sure it is architecture, but another attribute of moving from
/var/tmp to /tmp is that the size of "TMPDIR" is now constrained by the
size of the swap slice, rather than the "/var" slice (and a limited
percentage of swap). Configurations that used to run, may now fail.
Understand, this is a change we could allow, but considering that the
justification/benefit for this appears to be aesthetics, is it worth
it? (BTW: A lot of Solaris installations have rather small swap areas
because the install default is rather dinky)
Of course, pointing TMPDIR at /var/tmp/$USER addresses this particular
concern.
Also, what do the editors use for their temporary file? It would seem
to be a bit of a problem if they used TMPDIR. :-) (If they do, its
just a bug, but bugs happen and there are a couple too many editors in
the world, its not only the ones shipping with Solaris.) Then again, it
would seem that we would lose a fair amount of the benefit of per-user
directories if they didn't use them. (BTW, can I shoot whoever decided
that making me manually remove a vim *.swp file after a recovery was a
good idea? Speaking of things to fix for the benefit of the novice
user....)
Staying silent on the "EOU for system administrators" aspect of per-user
subdirectories, I think its a bad idea to move this from non-volatile
storage to volatile. (For that matter, just due to the issues with
slices/partitioning, I think its would be a bad idea to move it at all.)
>From today's perspective, non-volatile may be suboptimal, but it wasn't
a random choice. See:
1991/024 TMPDIR default Joseph Kowalski closed approved
07/22/92 Jarrett Rosenberg
Yep, this was the 24th PSARC case *ever* submitted (That's me as project
lead for SunOS 5.0, Jarrett is the ARC member ).
These choices were made with an eye to compatibility with SunOS 4.x and
System V.
- jek3
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://mail.opensolaris.org/pipermail/shell-discuss/attachments/20070427/fa85bb3f/attachment.html>
