bug#62694: 30.0.50; eglot-tests fails with recent pylsp

[Eli Zaretskii <eliz@gnu.org>]

> From: João Távora <joaotavora@gmail.com>
> Cc: Michael Albinus <michael.albinus@gmx.de>,  62694@debbugs.gnu.org
> Date: Fri, 07 Apr 2023 13:40:42 +0100
> 
> Eli Zaretskii <eliz@gnu.org> writes:
> 
> > Given João's reluctance to help you to find a better solution, 
> 
> FTR I've given a solution about 20 times now that was ignored
> repeteadly.  There is no general bullet-proof solution for the problem
> of broken or misbehaving installations of external tools.  So these
> tests cannot ever be "stable".  You can mark them _all_ unstable.

Great, then we agree.

Michael, would you please mark the relevant tests unstable?

> FTR I've given a solution known to be working in an Ubuntu-based CI
> system, very similar to Debian, for almost 5 years now.  Noone seems to
> be heeding it, so what can I do?
> 
> FTR I've explained at length why the "better solution" that you and
> Michael are conjecturing to be very easy is beyond me.  In my analysis,
> there is no simple Elisp code that can, in this pylsp case, easily
> discern between a functioning installation and a malfunctioning one.
> I've asked for your suggestions on how this can be done and I've not
> received any concrete ideas.
> 
> FTR, earlier this year, I took Michael's idea of adding a version check
> to clangd for eglot-tests.el because it was relatively easy and cheap
> and shown to be working.  _Not_ because there were any reports of people
> with old clangd running make check (absolutely 0 of those too), but
> because I found it easy to do so (and why not appease good old
> Michael?).  There the argument was that in that old Debian Stable debian
> system of EMBA it was not easy to install a newer clangd.  OK.  But, for
> pylsp that is _not_ the case at all, it's a simple one liner.
> 
> So if anyone is being stubborn here, it's _not_ me.

I didn't stay you were stubborn.