Upgrade: psutil 5.8.0

[dimpase]

Cygwin support needs a rebase.

CC: @embray @slel @orlitzky @mkoeppe @kiwifb @antonio-rojas @isuruf

Component: packages: standard

Keywords: upgrade, psutil

Work Issues: Cygwin

Branch/Commit: public/ticket/26253 @ 0fc3c30

Reviewer: Michael Orlitzky

Issue created by migration from https://trac.sagemath.org/ticket/26253

[dimpase]

comment:1

The cygwin-specific patch does not apply, needs a rebase. It's removed for the sake of being able to test on other system (needed on FreeBSD, older package does not work there).

[dimpase]

Branch: u/dimpase/psutils547

[dimpase]

Commit: eee1194

[dimpase]

New commits:

eee1194

updated and removed cygwin patch (needs a rebase)

[dimpase]

comment:3

meanwhile 5.4.8 is out.

[dimpase]

Changed branch from u/dimpase/psutils547 to none

[dimpase]

Changed commit from eee1194 to none

[embray]

comment:5

I keep starting and stopping on reworking my Cygwin stuff for psutil. It needs to be refactored pretty badly and every time I start to work on it I quickly get pulled away to other things.

Is there some specific reason to update the psutil version?

[dimpase]

comment:6

One reason is that the current version does not work on FreeBSD 12.0.
How hard is that refactoring? I have a Windows laptop around, which BSODs now and then due to some hardware issue, but probably is good enough for mild Cygwin work.

A related Cygwin issue is on #26052 - any idea about that?

[embray]

comment:7

It's very non-trivial, unfortunately, and requires upstream cooperation (where in this case the psutil maintainer has been helpful and amenable, but that doesn't mean I can just go and do what I want).

[dimpase]

comment:8

meanwhile psutils 5.5.0 is out

[fchapoton]

Changed keywords from none to upgrade

[embray]

comment:10

I'm working on a new (simpler) version of the Cygwin patch for psutil. Hopefully it will be ready soon.

[dimpase]

comment:11

the latest psutil is 5.6.2.

[kiwifb]

comment:12

It would be good to update. Distros likely use a more recent psutil (I do). It causes at least one optional doctest failure (when combined with fricas) due to changes to output

sage -t --long /usr/lib64/python2.7/site-packages/sage/interfaces/fricas.py
**********************************************************************
File "/usr/lib64/python2.7/site-packages/sage/interfaces/fricas.py", line 351, in sage.interfaces.fricas.FriCAS._quit_string
Failed example:
    p = fricas.pid(); pr = psutil.Process(p); pr                  # optional - fricas
Expected:
    <psutil.Process(pid=..., name='AXIOMsys') at ...>
Got:
    psutil.Process(pid=7277, name='AXIOMsys', started='19:52:17')
**********************************************************************

This is with psutil-5.5.0.

[fchapoton]

Branch: public/ticket/26253

[fchapoton]

Commit: 4d7ce51

[fchapoton]

comment:13

Here is a branch, not tested. I have not removed the patches.

---

New commits:

4d7ce51

update psutil to 5.6.2

[slel]

Changed keywords from upgrade to upgrade, psutil

[slel]

comment:42

Currently our spkg is at psutil 5.2.0,
from 2017-03 (with patches).

Python 3 deprecation warnings in psutil
were resolved in psutil itself by

• psutil pull request 1318,
  merged in 2018-08

With a recent psutil, the first start of
Sage on Cygwin could be warnings-free.

Some user reports mentioning the deprecation warnings:

• 2020-08: Ask Sage question 53120
• 2021-08: sage-support

[slel]

Changed work issues from Cygwin!! to Cygwin

[dimpase]

comment:43

psutil 5.8 does not support Cygwin, as far as I know.

[orlitzky]

comment:44

Do we need psutil at all? I think we can delete it. It's used for three things:

1. To guess at the initial GAP memory pool size, passed to gap via -o and -s. I don't think we need these flags, though: https://www.gap-system.org/Manuals/doc/ref/chap3.html
2. To set the maximum stack size during PARI initialization. Again, we can just set it to a big number; it's a limit, and we're already guessing.
3. Two doctests that call get_memory_usage() to check for leaks.

Those two doctests are not worth keeping it around, even as an optional package, considering it won't work on one of our supported platforms. So I say we,

1. Drop the -o and -s flags to GAP. If the absence of -s causes problems in some cases, we can always pass -s <big_number> since that memory should not actually be allocated. And if that assumption is violated on some platform, then we could special-case that platform.
2. Pass a big value to PARI in libs/pari/__init__.py for its stack size limit. 1024 times the initial size, or something like that.
3. Delete those two doctests.
4. Delete psutil, and sleep easier knowing that we won't be responsible for a fork of it for the rest of our lives.

[dimpase]

comment:45

+1 to removing psutils

[mkoeppe]

comment:46

+1 if it can be done without breaking stuff

[slel]

comment:48

Consider setting to invalid/wontfix after removal of psutil in #32656.

[slel]

Changed author from Dima Pasechnik to none

[slel]

comment:50

Ticket #32656 was merged in Sage 9.5.beta5.

[orlitzky]

Reviewer: Michael Orlitzky