Move SAGE_TMP to python's tempdir

[a-andre]

SAGE_TMP is currently located in DOT_SAGE. This patch moves it to tempfile.tempdir.

This is part of #17185.

CC: @orlitzky

Component: misc

Branch/Commit: u/aapitzsch/ticket/17828 @ 32cf221

Issue created by migration from https://trac.sagemath.org/ticket/17828

[a-andre]

Commit: 32cf221

[a-andre]

Branch: u/aapitzsch/ticket/17828

[jdemeyer]

comment:2

This is an important functional change which absolutely should be discussed on sage-devel first.

[jdemeyer]

comment:3

You need to consider also the security implications of this change.

[vbraun]

comment:4

In principle I prefer the real tmpdir, but in practice there is some sagenb legacy crud to be considered. In particular, sagenb is potentially a very long-running server that'll rely on fixed temporary directory for the duration of the user login. Might get deleted by tmpwatch while you are on holidays. For the security part: make sure that sagenb with server_pool doesn't do anything dangerous, it creates some world-writable directories in there.

[jdemeyer]

comment:5

If you need long lived temporary files, you need /var/tmp, not /tmp on Unix.

[vbraun]

comment:6

Maybe yes, maybe no. There is no guarantee that /var/tmp is not just a symlink to /tmp, though usually it is not. There is definitely no guarantee that old files are not cleaned there, too.

[jdemeyer]

comment:7

Replying to @vbraun:

Maybe yes, maybe no. There is no guarantee that /var/tmp is not just a symlink to /tmp, though usually it is not. There is definitely no guarantee that old files are not cleaned there, too.

Files in /var/tmp are supposed to survive a reboot, so I conclude from this that they would be kept alive during a running Sage session too. It is not a problem if /tmp and /var/tmp point to the same directory, as long as it's not cleaned too often.

[vbraun]

comment:8

The "survive a reboot" part means that /var/tmp is supposed to be on a hdd/ssd, whereas most distributions put /tmp on a ram disk nowadays. You can count on temp files being kept alive for a few days but not forever.

[orlitzky]

comment:9

I missed this ticket, but I've done this as part of #33213 which eliminates SAGE_TMP entirely after adjusting the library to use python's tempfile mechanisms instead.

[mkoeppe]

Changed author from André Apitzsch to none