Port syslog module to use module state

[corona10]

While I check the @ericsnowcurrently 's checklist: https://github.com/ericsnowcurrently/multi-core-python/wiki/0-The-Plan
I found that the syslog module still uses the global variable for the following variables.

• S_ident_o
• S_log_open

Both variables are declared as global variables since the original author assumes that in only one instance, only one syslog will exist.(one process one interpreter)

So I think that it's okay to move them into the module state if we consider the per-interpreter model.

• PR: gh-99127: Allow some features of syslog to the main interpreter only #99128

[corona10]

Hmm, I think that I miss understood about syslog, I close ths issue as invalid.

[ronaldoussoren]

I've added some review comments to the PR, it seems to be on right path with some minor issues because the syslog extension changes process global state in the syslog library. Those issues IMHO can be easily fixed.

[corona10]

I've added some review comments to the PR, it seems to be on right path with some minor issues because the syslog extension changes process global state in the syslog library. Those issues IMHO can be easily fixed.

Thanks I will take a look.

[ericsnowcurrently]

Note that currently the syslog module already has the following characteristics:

• an app that embeds CPython may call openlog() independently of any use of the syslog module, and the two may interfere with each other
• two interpreters using the syslog module may interfere with each other in the same way

We can't do much about the embedding case because the posix syslog API doesn't provide a way to get the current configuration (if any) such that you could restore it.

For multiple interpreters, either we leave the current behavior or we try to make each interpreter independent. To do that, we have two options:

• swap in each interpreter's configuration around each syslog.syslog() call
  • call openlog() before and closelog() after
  • call setlogmask() before and after
  • syslog.openlog() and syslog.setlogmask() would only store the values on the module state
• do similar, but be smarter about the single-interpreter case and cases where swapping is not necessary (i.e. same syslog config)

The "smarter" approach would require that the syslog module keep process-global state, along with a lock to protect that state. That would require something similar to what is discussed in https://discuss.python.org/t/20668 (and https://discuss.python.org/t/20663).

---

For reference:

• https://pubs.opengroup.org/onlinepubs/009695399/functions/openlog.html
• https://www.gnu.org/software/libc/manual/html_node/openlog.html
• https://stackoverflow.com/questions/12987180
• https://sourceware.org/git/?p=glibc.git;a=blob;f=misc/syslog.c;h=f67d4b58a448cabdf63f629f72a9df6e009286cf;hb=refs/heads/master
• https://ruby-doc.org/stdlib-2.7.1/libdoc/syslog/rdoc/Syslog.html
• https://docs.python.org/3/library/syslog.html

[ericsnowcurrently]

FWIW, for an implementation of mutli-phase init for a module, I would expect it to either deal with the sort of weird behavior syslog would have, or disallow use in multiple interpreters (at least in problematic cases).