Anyone ever used gnus-change-server?

Anyone ever used gnus-change-server?

[Steinar Bang]

I tried using gnus-change-server yesterday to change the native server.

I've tried that before when changing news servers, but never really had
any success.  The function has never run to completion, as far as I've
been able to tell.

The documentation says that the methods for changing server "take a
looong time to run".  What I've done is just live with the article
number changes, and lose my read marks.

However, this time I just had 5 groups to switch, and a reasonably fast
machine to run Gnus on (a P4), so I thought it may be worth it.

But this morning the P4 machine had been using 100% CPU on this
overnight, and still hadn't completed.

Has anyone ever had any success with gnus-change-server?

Re: Anyone ever used gnus-change-server?

[Steinar Bang]

>>>>> Steinar Bang <sb@dod.no>:

> However, this time I just had 5 groups to switch, and a reasonably
> fast machine to run Gnus on (a P4), so I thought it may be worth it.

> But this morning the P4 machine had been using 100% CPU on this
> overnight, and still hadn't completed.

Still running at 100% CPU.  I think I'll C-g gnus-change-server and
reclaim the process.

Re: Anyone ever used gnus-change-server?

[Steinar Bang]

>>>>> Steinar Bang <sb@dod.no>:

> Still running at 100% CPU.  I think I'll C-g gnus-change-server and
> reclaim the process.

I C-g'd gnus-change-server, exited Gnus and stopped the emacs process,
and started with a fresh emacs process.

I first started wireshark and then let it listen to port 119.

Then, in the freshly started emacs process, I ran
	M-x gnus-change-server RET

Then I let it run for approximately 21h 30min, with a CPU usage of 100%,
or near it, on the emacs process, and then I C-g'd the
gnus-change-server function, stopped the wireshark capture, and looked
at what I had.

And what I had was gnus-change-server contacting the old server and
saying this:

LIST ACTIVE no.teknologi.bil
215 Newsgroups in form "group high low flags".
no.teknologi.bil 0000065091 0000056560 y
.
LIST ACTIVE no.teknologi.bil
215 Newsgroups in form "group high low flags".
no.teknologi.bil 0000065091 0000056560 y
.
LIST ACTIVE no.alt.motorsykler
LIST ACTIVE no.sport.motorsport
LIST ACTIVE no.test
215 Newsgroups in form "group high low flags".
no.alt.motorsykler 0000033522 0000000001 y
.
215 Newsgroups in form "group high low flags".
no.sport.motorsport 0000000621 0000000523 y
.
215 Newsgroups in form "group high low flags".
no.test 0000003801 0000003258 y
.
NEWGROUPS 100131 100308 GMT
231 New newsgroups follow.
.
GROUP no.test
211 510 3258 3801 no.test
GROUP no.sport.motorsport
211 83 523 621 no.sport.motorsport
GROUP no.alt.motorsykler
211 17150 1 33522 no.alt.motorsykler
XOVER 1-4269
224 1-4269 fields follow

and then follows 16645 lines of NOV information, followed eventually
with a . on a single line.

And then nothing, except for a large CPU usage, without any perceivable
result. 

The group in question has 66 ticked articles.  And Gnus already has all
the NOV information for the ticked articles.

Why is it interesting to talk to the original news server at all?
Wouldn't it be enough to contact the new server, get some NOV
information from it, and match the message ids there to the ticked
articles in the original server's group?

On second thought: I need both the ticked articles and the read articles
from the old group.  Or do I?  I only need enough of the newest read
articles of the old group to "sync up" to the same articles on the new
group.

Re: Anyone ever used gnus-change-server?

[Steinar Bang]

>>>>> Steinar Bang <sb@dod.no>:

[snip! wireshark capture of port 119]
> And what I had was gnus-change-server contacting the old server and
> saying this:
[snip!]

What's strange, looking at the code of gnus-move-group-to-server, is
that that method contacts the to-server first, and then the from-server.

And I don't have a single packet captured for the to-server... no
non-standard port number, no SSL, 

Perhaps wireshark has fooled me?  Or perhaps there's something I don't
see? (and that's the reason nothing completes...?)

Re: Anyone ever used gnus-change-server?

[Steinar Bang]

I've done a little edebug'ing and this is what I found out:

 - I did edebug-defun on gnus-move-group-to-server
 - I ran `M-x gnus-change-server RET'
 - On each stop in gnus-move-group-to-server I did `c', until it
   eventually got stuck in moving no.alt.motorsykler with 33523 articles
   seen.  Here are the first lines of the edebug-trace (it's too big to
   include all of it):
    { gnus-move-group-to-server args: ((dummy.group 0 nil) (nntp news) (nn
    } gnus-move-group-to-server result: Translating dummy.group...done    
    { gnus-move-group-to-server args: ((no.test 6 ((1 . 952) (3258 . 3283)
    } gnus-move-group-to-server result: Translating no.test...done        
    { gnus-move-group-to-server args: ((no.sport.motorsport 3 ((1 . 44) 52
    } gnus-move-group-to-server result: Translating no.sport.motorsport...
    { gnus-move-group-to-server args: ((no.alt.motorsykler 3 ((1 . 33523))
 - I left it running for a day or so, and then C-g'd it
 - Then I stepped through some iterations with `n'.  It was spending its
   time in this loop:
	  (while (looking-at
		  "^[0-9]+\t[^\t]*\t[^\t]*\t[^\t]*\t\\([^\t]*\\)\t")
	    (when (setq to-article
			(gnus-gethash
			 (buffer-substring (match-beginning 1) (match-end 1))
			 hashtb))
	      ;; Add this article to the list of read articles.
	      (push to-article to-reads)
	      ;; See if there are any marks and then add them.
	      (when (setq mark (assq (read (current-buffer)) marks))
		(setq marks (delq mark marks))
		(setcar mark to-article)
		(push mark to-marks))
	      (forward-line 1)))
   The test in the while returned t for all of the steppings I did, and
   the cond in the when returned nil for all steppings I did.
   Not really a surprising result if we're talking about 60 ticked
   articles of a total of 33000 or thereabouts

I'm guessing this while loop is where it has spent its time.

So I guess the move-server methods really take too much time to be of
use.  At least if the available article count is too large (with a 14
day expiry, they probably would have worked ok).

Re: Anyone ever used gnus-change-server?

[Lars Magne Ingebrigtsen]

Steinar Bang <sb@dod.no> writes:

> So I guess the move-server methods really take too much time to be of
> use. 

Yeah.  Perhaps it should just be deleted?

-- 
(domestic pets only, the antidote for overdose, milk.)
  larsi@gnus.org * Lars Magne Ingebrigtsen