emacs-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Emacs tramp troubles with old Sun


From: Michael Albinus
Subject: Re: Emacs tramp troubles with old Sun
Date: Mon, 12 Nov 2007 21:23:35 +0100
User-agent: Gnus/5.110007 (No Gnus v0.7) Emacs/23.0.50 (gnu/linux)

I wrote:

> Stefan Monnier <address@hidden> writes:
>
>> Would it be possible for Tramp to detect the "test: argument expected" error
>> message and force a "cleanup-connection" at that point?
>> [ and of course, do similar things wherever applicable: the idea is to try
>>   and detect when the cached properties become invalid, which will
>>   necessarily happen over time as machines get upgraded/replaced/... ]
>
> Exactly the same idea I'm just thinking about. But one must ensure
> there's no infloop - when "test" does not work, it could be another
> reason but invalid cached values. One shall remember the reason for
> calling cleanup-connection, and when the same problem appears again,
> it's time to give up.
>
> Another idea is to keep a counter for cached properties. After a while
> (100 uses, 100 days, whatever) the values shall be recomputed, just in
> case. Think about a changed default $PATH, where it isn't always an
> obvious error with the old value, when new directories appear.
>
> At least when "uname -s -r" returns a changed value, it is time to
> expire the cached properties for that machine.

Finally, I've implemented only the "uname -sr" check. The other checks
for connection properties cleanups I have had in mind are too
dangerous. Tramp sends remote commands on the fly (when it is necessary
only). Therefore, command sequences are not "atomic" such a way one
could just flush the cache, and reapply the command.

Another drawback would be that error analysis might be impossible when
all relevant data are thrown away once an error happened.

The original problem we have spoken about was that Tramp did not
recognize shell echoing on the "old Sun". Wrong cache entries were
likely a result of _this_ error. I expect to reduce the number of errors
in Tramp :-) Such situations should not happen so often in the future.

>> -- Stefan

Best regards, Michael.




reply via email to

[Prev in Thread] Current Thread [Next in Thread]