Can't start CVS gnus.

Can't start CVS gnus.

[Matthias Andree]

Hi,

I have some issues with a recent CVS gnus (emacs 21.2 as shipped for
SuSE Linux 8.1):

1. it cannot find mm-guess-mime-charset, claims the function's
definition was void.

I was able to fix this with (load-library "mm-util")

Then, 2. occurred:

2. it cannot figure gnus-byte-compile, claims the function's definition
was void.

I was able to fix this with (load-library "gnus-util")

Then, it cannot find nnheader-parse-naked-head and barfs again.

What's going on here?

Re: Can't start CVS gnus.

[Matt Armstrong]

Matthias Andree <matthias.andree@gmx.de> writes:

> Hi,
>
> I have some issues with a recent CVS gnus (emacs 21.2 as shipped for
> SuSE Linux 8.1):
>
> 1. it cannot find mm-guess-mime-charset, claims the function's
> definition was void.
>
> I was able to fix this with (load-library "mm-util")
>
> Then, 2. occurred:
>
> 2. it cannot figure gnus-byte-compile, claims the function's definition
> was void.
>
> I was able to fix this with (load-library "gnus-util")
>
> Then, it cannot find nnheader-parse-naked-head and barfs again.
>
> What's going on here?

For whatever reason, your lisp/gnus-load.el is probably not generated
correctly.  Is that file empty?  Mine is about 1400 lines long.

Re: Can't start CVS gnus.

[Matthias Andree]

Matt Armstrong <matt@lickey.com> writes:

> For whatever reason, your lisp/gnus-load.el is probably not generated
> correctly.  Is that file empty?  Mine is about 1400 lines long.

Not quite, removing SuSE's emacs-url package fixed that for now. I
believe I smell some breakage there...

(I still have an issue with their emacs-auctex package: if installed,
list-load-path-shadows doesn't work).

-- 
Matthias Andree

Re: Can't start CVS gnus.

[Kai Großjohann]

Matthias Andree <ma@dt.e-technik.uni-dortmund.de> writes:

> Matt Armstrong <matt@lickey.com> writes:
>
>> For whatever reason, your lisp/gnus-load.el is probably not generated
>> correctly.  Is that file empty?  Mine is about 1400 lines long.
>
> Not quite, removing SuSE's emacs-url package fixed that for now. I
> believe I smell some breakage there...

url is part of W3, and W3 has its own version of the mm-foo
packages.  The Gnus version is more recent and you (or the SuSE
packagers) need to make sure that the Gnus version is used instead of
the W3 version.

C-h v load-path RET might tell you if the right version is found
first.

kai
-- 
~/.signature is: umop ap!sdn    (Frank Nobis)

Re: Can't start CVS gnus.

[Matthias Andree]

kai.grossjohann@uni-duisburg.de (Kai Großjohann) writes:

> url is part of W3, and W3 has its own version of the mm-foo
> packages.  The Gnus version is more recent and you (or the SuSE
> packagers) need to make sure that the Gnus version is used instead of
> the W3 version.
>
> C-h v load-path RET might tell you if the right version is found
> first.

I do have /home/emma/.lisp as the --prefix for Gnus and first up in the
path, so that's not it. There must be some other obscure breakage, maybe
introduced by SuSE's site scripts or whatever else might be executed
before my .gnu-emacs-custom. I haven't done research on that. Is there
something like "strace" or "truss"? I'd like to see which files emacs is
loading. Sure, I can use the real syscall tracer, but that's good for a
lot of screen clutter.

My intent was to document the band-aid fix among the Gnus experts so
they can ask people "recent SuSE? try rpm -e emacs-w3 emacs-url and see
if it helps".

-- 
Matthias Andree

Re: Can't start CVS gnus.

[Raymond Scholz]

* Matthias Andree <ma@dt.e-technik.uni-dortmund.de> wrote:

> I do have /home/emma/.lisp as the --prefix for Gnus and first up in the
> path, so that's not it. 

Hm, just to be sure, try M-x list-load-path-shadows

Cheers, Ray
-- 
Look Ma, this man can twist his fingers as if they were made of rubber,
isn't that amazing? -- Not really, he's been using Emacs for years...!

Re: Can't start CVS gnus.

[Matthias Andree]

Raymond Scholz <ray-2002@zonix.de> writes:

> * Matthias Andree <ma@dt.e-technik.uni-dortmund.de> wrote:
>
>> I do have /home/emma/.lisp as the --prefix for Gnus and first up in the
>> path, so that's not it. 
>
> Hm, just to be sure, try M-x list-load-path-shadows

Same result, with another obstacle: I cannot use list-load-path-shadows
while emacs-auctex is installed: (from messages buffer):

| Loading shadow (source)...
| eval-buffer: Symbol's function definition is void: TeX-add-style-hook

Don't ask me from under what stone /this/ zombie rose again. I'm running
out of torches.

-- 
Matthias Andree

Re: Can't start CVS gnus.

[Kai Großjohann]

Matthias Andree <ma@dt.e-technik.uni-dortmund.de> writes:

> I do have /home/emma/.lisp as the --prefix for Gnus and first up in the
> path, so that's not it. There must be some other obscure breakage, maybe
> introduced by SuSE's site scripts or whatever else might be executed
> before my .gnu-emacs-custom. I haven't done research on that. Is there
> something like "strace" or "truss"? I'd like to see which files emacs is
> loading. Sure, I can use the real syscall tracer, but that's good for a
> lot of screen clutter.

Maybe it works to defadvise `load' to print a message.

Also, it might help to look in the Emacs init files that the
emacs-url package is installing.

kai
-- 
~/.signature is: umop ap!sdn    (Frank Nobis)

Re: Can't start CVS gnus.

[Reiner Steib]

On Mon, Nov 25 2002, Matthias Andree wrote:

> kai.grossjohann@uni-duisburg.de (Kai Großjohann) writes:
[...]
>> C-h v load-path RET might tell you if the right version is found
>> first.
>
> I do have /home/emma/.lisp as the --prefix for Gnus and first up in the
> path, so that's not it. 

Are you sure? If some library is loaded in the site file(s) and you
change load-path (later) in your ~/.emacs, you will not get what you
expect.

> There must be some other obscure breakage, maybe introduced by
> SuSE's site scripts or whatever else might be executed before my
> .gnu-emacs-custom. I haven't done research on that.

Calling Emacs with `--no-site-file' prevents loading any (SuSE) site
files (site-start.el, ...). I assume you already checked your ~/.emacs
file, so that `-q' should not be necessary.

Bye, Reiner.
-- 
       ,,,
      (o o)
---ooO-(_)-Ooo--- PGP key available via WWW   http://rsteib.home.pages.de/

Re: Can't start CVS gnus.

[Matthias Andree]

Reiner Steib <4uce.02.r.steib@gmx.net> writes:

> Are you sure? If some library is loaded in the site file(s) and you
> change load-path (later) in your ~/.emacs, you will not get what you
> expect.

I believe typing load-path C-j in *scratch* does not lie to me.

> Calling Emacs with `--no-site-file' prevents loading any (SuSE) site
> files (site-start.el, ...). I assume you already checked your ~/.emacs
> file, so that `-q' should not be necessary.

I'll try once I feel like debugging SuSE again...

-- 
Matthias Andree

Re: Can't start CVS gnus.

[Kai Großjohann]

Matthias Andree <ma@dt.e-technik.uni-dortmund.de> writes:

> Reiner Steib <4uce.02.r.steib@gmx.net> writes:
>
>> Are you sure? If some library is loaded in the site file(s) and you
>> change load-path (later) in your ~/.emacs, you will not get what you
>> expect.
>
> I believe typing load-path C-j in *scratch* does not lie to me.

It gives you the _current_ value of load-path.

But the value used when loading Gnus might have been different.

kai
-- 
~/.signature is: umop ap!sdn    (Frank Nobis)

Re: Can't start CVS gnus.

[Reiner Steib]

On Wed, Nov 27 2002, Matthias Andree wrote:

> Reiner Steib <4uce.02.r.steib@gmx.net> writes:
>
>> If some library is loaded in the site file(s) and you change
>> load-path (later) in your ~/.emacs, you will not get what you
>> expect.
>
> I believe typing load-path C-j in *scratch* does not lie to me.

It tells you the _current_ value of load-path which is not always the
same as when the site files were processed.

>> Calling Emacs with `--no-site-file' prevents loading any (SuSE) site
>> files (site-start.el, ...). I assume you already checked your ~/.emacs
>> file, so that `-q' should not be necessary.
>
> I'll try once I feel like debugging SuSE again...

JFTR: Some people reported such problems because of Debian's site
files. ;-)

Bye, Reiner.
-- 
       ,,,
      (o o)
---ooO-(_)-Ooo--- PGP key available via WWW   http://rsteib.home.pages.de/

Re: Can't start CVS gnus.

[Matthias Andree]

kai.grossjohann@uni-duisburg.de (Kai Großjohann) writes:

> It gives you the _current_ value of load-path.
>
> But the value used when loading Gnus might have been different.

Indeed. SuSE's script adjusted the load path, and (require 'd) (funny
how lisp fits in) mm-util right away, so I'd have to force a reload.

Killing these packages again, they do no good anyways. I'm not using URL
or W3.

-- 
Matthias Andree

Re: Can't start CVS gnus.

[Matthias Andree]

Reiner Steib <4uce.02.r.steib@gmx.net> writes:

> JFTR: Some people reported such problems because of Debian's site
> files. ;-)

So Emacs needs versioned libraries...

-- 
Matthias Andree

Re: Can't start CVS gnus.

[Kai Großjohann]

Matthias Andree <ma@dt.e-technik.uni-dortmund.de> writes:

> So Emacs needs versioned libraries...

Version numbers are usually the wrong way to approach it.  Instead,
you design a test for the feature you need, then use that test.

In the case of Gnus, it's just that the mm-foo stuff is considered to
be part of Gnus, so there are no extra checks.  (Just like gnus.el
doesn't check that gnus-sum.el has the right version.)

-- 
~/.signature is: umop ap!sdn    (Frank Nobis)

Re: Can't start CVS gnus.

[Matthias Andree]

kai.grossjohann@uni-duisburg.de (Kai Großjohann) writes:

> Matthias Andree <ma@dt.e-technik.uni-dortmund.de> writes:
>
>> So Emacs needs versioned libraries...
>
> Version numbers are usually the wrong way to approach it.  Instead,
> you design a test for the feature you need, then use that test.

How will you resolve version conflicts without version numbers? Testing
for every possible feature of the API is going to be costly.

-- 
Matthias Andree

Re: Can't start CVS gnus.

[Kai Großjohann]

Matthias Andree <ma@dt.e-technik.uni-dortmund.de> writes:

> How will you resolve version conflicts without version numbers? Testing
> for every possible feature of the API is going to be costly.

Sometimes, version numbers are necessary.  But one should try to do
without, first.

Do you think more than a handfull of tests are necessary?
-- 
~/.signature is: umop ap!sdn    (Frank Nobis)

Re: Can't start CVS gnus.

[Matthias Andree]

kai.grossjohann@uni-duisburg.de (Kai Großjohann) writes:

> Sometimes, version numbers are necessary.  But one should try to do
> without, first.

Well, yes.

> Do you think more than a handfull of tests are necessary?

I cannot judge that. My Lisp experience is minimal. I see it uses a
prefix-notation, and that's about all I know.

Nonetheless, it'd be useful if Gnus figured when mm-util for one was
outdated and forced a reload of the version that shipped with it.

-- 
Matthias Andree

Re: Can't start CVS gnus.

[Kai Großjohann]

Matthias Andree <ma@dt.e-technik.uni-dortmund.de> writes:

> Nonetheless, it'd be useful if Gnus figured when mm-util for one was
> outdated and forced a reload of the version that shipped with it.

I feel uneasy about that suggestion, but Gnus could print a warning
for sure.

How does one find out whether one is using the `wrong' mm-foo?
-- 
~/.signature is: umop ap!sdn    (Frank Nobis)

Re: Can't start CVS gnus.

[Matthias Andree]

kai.grossjohann@uni-duisburg.de (Kai Großjohann) writes:

> Matthias Andree <ma@dt.e-technik.uni-dortmund.de> writes:
>
>> Nonetheless, it'd be useful if Gnus figured when mm-util for one was
>> outdated and forced a reload of the version that shipped with it.
>
> I feel uneasy about that suggestion, but Gnus could print a warning
> for sure.
>
> How does one find out whether one is using the `wrong' mm-foo?

Looking at the initial post of this thread:

| 1. it cannot find mm-guess-mime-charset, claims the function's
| definition was void.

-- 
Matthias Andree

Re: Can't start CVS gnus.

[Kai Großjohann]

Matthias Andree <ma@dt.e-technik.uni-dortmund.de> writes:

> kai.grossjohann@uni-duisburg.de (Kai Großjohann) writes:
>
>> Matthias Andree <ma@dt.e-technik.uni-dortmund.de> writes:
>>
>>> Nonetheless, it'd be useful if Gnus figured when mm-util for one was
>>> outdated and forced a reload of the version that shipped with it.
>>
>> I feel uneasy about that suggestion, but Gnus could print a warning
>> for sure.
>>
>> How does one find out whether one is using the `wrong' mm-foo?
>
> Looking at the initial post of this thread:
>
> | 1. it cannot find mm-guess-mime-charset, claims the function's
> | definition was void.

Thanks!  I've concocted the following patch:

--- gnus.el.~6.125.~	2002-12-02 14:02:09.000000000 +0100
+++ gnus.el	2002-12-03 08:21:04.000000000 +0100
@@ -34,6 +34,10 @@
 (require 'wid-edit)
 (require 'mm-util)
 
+;; Make sure it was the right mm-util.
+(unless (fboundp 'mm-guess-mime-charset)
+  (error "Wrong `mm-util' found in `load-path'.  Make sure the Gnus one is found first."))
+
 (defgroup gnus nil
   "The coffee-brewing, all singing, all dancing, kitchen sink newsreader."
   :group 'news

What do you think?
-- 
~/.signature is: umop ap!sdn    (Frank Nobis)

Re: Can't start CVS gnus.

[Matthias Andree]

Kai Großjohann schrieb am Dienstag, den 03. Dezember 2002:

> +;; Make sure it was the right mm-util.
> +(unless (fboundp 'mm-guess-mime-charset)
> +  (error "Wrong `mm-util' found in `load-path'.  Make sure the Gnus one is found first."))
> +

That's near-to-impossible to achieve on e. g. SuSE and their site
scripts. "Make sure the Gnus one is loaded first" might be more to the
point.

> What do you think?

Other than that, I'd have to retest later. I'll send separate mail when
I got around to do that.

Re: Can't start CVS gnus.

[Kai Großjohann]

Matthias Andree <matthias.andree@gmx.de> writes:

> Kai Großjohann schrieb am Dienstag, den 03. Dezember 2002:
>
>> +;; Make sure it was the right mm-util.
>> +(unless (fboundp 'mm-guess-mime-charset)
>> +  (error "Wrong `mm-util' found in `load-path'.  Make sure the Gnus one is found first."))
>> +
>
> That's near-to-impossible to achieve on e. g. SuSE and their site
> scripts. "Make sure the Gnus one is loaded first" might be more to the
> point.

???  Usually, people do (require 'mm-util).  That searches through
load-path.  So the right one should come first.

Or do you mean that the sequence should be:

    Emacs starts
    Gnus gets added to load-path
    mm-util from Gnus gets loaded
    W3 gets added to load-path

This avoids loading the wrong one, of course.  But this is not a
solution, since it would require loading everything.  That would be
dog slow.

-- 
~/.signature is: umop ap!sdn    (Frank Nobis)

Re: Can't start CVS gnus.

[Matthias Andree]

Kai Großjohann schrieb am Dienstag, den 03. Dezember 2002:

> Or do you mean that the sequence should be:
> 
>     Emacs starts
>     Gnus gets added to load-path
>     mm-util from Gnus gets loaded
>     W3 gets added to load-path

I can tell what the sequence IS:

emacs starts
site-file adds path to old mm-util and require's it
my .gnu-emacs-custom and .gnus scripts are read

It'd require some fiddling with SuSE's site-scripts to get the Rightful
mm-util in. How do I reload the proper mm-util again? Is that possible?

what I'd need is:

1a. unload mm-util
1b. fix load path
2. require mm-util

Re: Can't start CVS gnus.

[Kai Großjohann]

Matthias Andree <matthias.andree@gmx.de> writes:

> I can tell what the sequence IS:
>
> emacs starts
> site-file adds path to old mm-util and require's it
> my .gnu-emacs-custom and .gnus scripts are read

This appears broken.

> It'd require some fiddling with SuSE's site-scripts to get the Rightful
> mm-util in. How do I reload the proper mm-util again? Is that possible?

I think the path to the old mm-util is the path to W3?  Of course,
that's still needed.  But nix the loading.

> what I'd need is:
>
> 1a. unload mm-util
> 1b. fix load path
> 2. require mm-util

I suggest to just frob the SuSE init files, they appear broken in
this regard anyway.  What do the SuSE experts say?

(Actually, I think my own Emacs setup is also broken, for people who
want to use W3 together with Gnus 5.9 -- then the W3 directory will
come first in load-path.  Hm.  Oh!  The problem is not there because
my W3 doesn't have mm-util.el :-))

-- 
~/.signature is: umop ap!sdn    (Frank Nobis)

Re: Can't start CVS gnus.

[Matthias Andree]

On Tue, 03 Dec 2002, Kai Großjohann wrote:

> > emacs starts
> > site-file adds path to old mm-util and require's it
> > my .gnu-emacs-custom and .gnus scripts are read
> 
> This appears broken.

Not quite, it's actually reasonable. I have Oort Gnus in ~/.lisp and
Gnus 5.9 in the system directories.

> > It'd require some fiddling with SuSE's site-scripts to get the Rightful
> > mm-util in. How do I reload the proper mm-util again? Is that possible?
> 
> I think the path to the old mm-util is the path to W3?  Of course,
> that's still needed.  But nix the loading.
> 
> > what I'd need is:
> >
> > 1a. unload mm-util
> > 1b. fix load path
> > 2. require mm-util
> 
> I suggest to just frob the SuSE init files, they appear broken in
> this regard anyway.  What do the SuSE experts say?

Haven't asked them, and I'm loathe to do that. They don't even have the
time to look into more serious issues, say, /usr/bin/Check touching both
the $RPM_BUILD_ROOT/blah/fasel and the /blah/fasel files at the same
time which is certainly wrong... send a fix, but haven't seen the update
RPMs yet. ntp does not work as broadcastclient, reported against 7.3,
still unfixed, although a one-line patch that works for me was in the
mail.

Re: Can't start CVS gnus.

[Matthias Andree]

Kai Großjohann schrieb am Dienstag, den 03. Dezember 2002:

> Thanks!  I've concocted the following patch:
> 
> --- gnus.el.~6.125.~	2002-12-02 14:02:09.000000000 +0100
> +++ gnus.el	2002-12-03 08:21:04.000000000 +0100
> @@ -34,6 +34,10 @@
>  (require 'wid-edit)
>  (require 'mm-util)
>  
> +;; Make sure it was the right mm-util.
> +(unless (fboundp 'mm-guess-mime-charset)
> +  (error "Wrong `mm-util' found in `load-path'.  Make sure the Gnus one is found first."))
> +
>  (defgroup gnus nil
>    "The coffee-brewing, all singing, all dancing, kitchen sink newsreader."
>    :group 'news
> 
> What do you think?

"Worksforme".

Re: Can't start CVS gnus.

[kgreiner]

Matthias Andree <matthias.andree@gmx.de> writes:

> what I'd need is:
>
> 1a. unload mm-util

Well, I don't know about an unload but you can do a reload.  That may
be sufficient. 

> 1b. fix load path
> 2. require mm-util


;; Make sure that the right mm-util is loaded.
(unless (fboundp 'mm-guess-mime-charset)
  (let ((load-path (list (file-name-directory (locate-library "gnus")))))
  (load-library "mm-util")))

I have no idea whether this is portable to XEmacs.  It does seem to
work fine in emacs 21.2. 

Kevin

Re: Can't start CVS gnus.

[Kai Großjohann]

Matthias Andree <matthias.andree@gmx.de> writes:

> On Tue, 03 Dec 2002, Kai Großjohann wrote:
>
>> > emacs starts
>> > site-file adds path to old mm-util and require's it
>> > my .gnu-emacs-custom and .gnus scripts are read
>> 
>> This appears broken.
>
> Not quite, it's actually reasonable. I have Oort Gnus in ~/.lisp and
> Gnus 5.9 in the system directories.

??

Why should the old mm-util be loaded?  My site files just frob
load-path and don't load mm-util.

-- 
~/.signature is: umop ap!sdn    (Frank Nobis)

Re: Can't start CVS gnus.

[Matthias Andree]

kai.grossjohann@uni-duisburg.de (Kai Großjohann) writes:

> Why should the old mm-util be loaded?  My site files just frob
> load-path and don't load mm-util.

Because the emacs-url or emacs-w3m RPMs seem to depend on this and the
SuSE site files prefer to (require 'mm-util) right after adding the path
for these packages. Don't ask.

I can do without these well, I'm just telling this list where the
culprit hides...

-- 
Matthias Andree

Re: Can't start CVS gnus.

[Reiner Steib]

On Wed, Dec 04 2002, Kai Großjohann wrote:

> Matthias Andree <matthias.andree@gmx.de> writes:
>
>> On Tue, 03 Dec 2002, Kai Großjohann wrote:
>>
>>> > emacs starts
>>> > site-file adds path to old mm-util

The path to mm-util is /usr/share/emacs/21.?/lisp/gnus/mm-util.el and
it's added to load-path by lisp/subdirs.el (not site-file.el) as in
the vanilla Emacs distribution.

>>> > and require's it

See below.

>>> > my .gnu-emacs-custom and .gnus scripts are read
>>> 
>>> This appears broken.
>>
>> Not quite, it's actually reasonable. I have Oort Gnus in ~/.lisp and
>> Gnus 5.9 in the system directories.
[...]
> Why should the old mm-util be loaded?  My site files just frob
> load-path and don't load mm-util.

The problem (or I would say "bug") is that SuSE[1] in fact require
some packages in site-start.el or files loaded by it:

,----[ site-lisp/site-start.el (wrapped long lines) ]
| ;; Load some package startups
| ;; --------------------------
| (let ((dir (directory-files "/usr/share/emacs/site-lisp/"
| 			    t "^suse-start-.*\\.el$")))
|      (while dir (load (car dir) nil t t) (setq dir (cdr dir))))
`----

The bug is that those require url:

,----[ site-lisp$ grep -e "require.*url" suse-start-*.el ]
| suse-start-url.el:(require 'url)
| suse-start-w3.el:(require 'url)
`----

AFAICS this is completely needlessly, as they only frob load-path to
include site-lisp/url/ or site-lisp/w3/. So, url requires url-var,
which requires mm-util:

,----[ site-lisp$ grep -e "mm-util" *.el */*.el ]
| url/url-vars.el:(require 'mm-util)
`----

A better solution (for SuSE) would be to use `subdirs.el' as the
vanilla Emacs distribution does. Maybe they would need to add some
auto-load statements somewhere.

For the SuSE-users, I would recommend:

1) Get rid of site-start.el and eventually replace it with a new file.

   # cd /usr/share/emacs/site-lisp
   # cp -p /usr/share/emacs/21.?/site-lisp/subdirs.el .
   # mv site-start.el SuSE-site-start.el

   # emacs site-start.el
   ... and add the stuff you really need (e.g. from SuSE-site-start.el).

2) Rename or remove:
   /etc/skel/.emacs
   /etc/skel/.gnu-emacs
   ~/.emacs
   ~/.gnu-emacs

   # emacs ~/.emacs
   ... and add the stuff you really need.

I've done this (or let's say something similar) with many SuSE- and
Emacs-versions up to SuSE 7.1. AFAICS, it still applies to SuSE 8.0.

Bye, Reiner.

[1] Not only SuSE; I've read about similar problems with Debian too.
-- 
       ,,,
      (o o)
---ooO-(_)-Ooo--- PGP key available via WWW   http://rsteib.home.pages.de/