vc-mtn bug?

vc-mtn bug?

[Sebastian Rose]

Hi,

I wrote a mail like this before but to no avail. There seems to be a bug
in vc-mtn.el that leeds to dataloss and therefore is very serious
IMHO. I suspected org.el in the first place, but it turned out, that the
bogous behaviour happens only, when starting emacs in a directory under
monotone version controll.

I'll try to describe some of the strange events here and add more
information in subsequent mails when available.



* case 1

  I start emacs in an monotone workspace directory with an existent
  filename as argument:

  sh$ emacs existent.php

  emacs starts as expected. Now I try to open a non-existent file in the
  same directory, using

  M-x non-existent.php

  The debug-message:

  Debugger entered--Lisp error: (wrong-type-argument arrayp nil)
  file-truename(nil)
  find-file-noselect-1(#<buffer non-existent.php> "~/develop/htdocs/shop/php5/web/non-existent.php" nil nil "~/develop/htdocs/shop/php5/web/non-existent.php" nil)
  find-file-noselect("~/develop/htdocs/shop/php5/web/non-existent.php" nil nil t)
  find-file("~/develop/htdocs/shop/php5/web/non-existent.php" t)
  call-interactively(find-file nil nil)

  A second attempt succeeds. This case is reproducable here.




* case 2:

  Edit an *.org file in a monotone workspace directory. The org file is
  under monotone control. Then I try to export the org file as HTML by

  C-c C-e h

  The HTML file does not exist yet, and is not registered in
  monotone. This is the one that can lead to data loss, since an
  arbitrary file is overridden. Preferably one of the files recently
  edited in emacs.

  This one is especially anoying. One time it turned my .emacs into a
  human readable file --- but an XHTML file :(




System is Debian testing and:

   - GNU emacs 23.0.60.2 (CVS version of today)
     (setq debug-on-error t)

   - sh$ mtn --version
     monotone 0.36 (base revision: e4bc808d89e029ce623f9e8f2b10c84006b83fb5)



Regards,

     Sebastian

Re: vc-mtn bug?

[Stefan Monnier]

> I wrote a mail like this before but to no avail. There seems to be a bug
> in vc-mtn.el that leeds to dataloss and therefore is very serious
> IMHO. I suspected org.el in the first place, but it turned out, that the
> bogous behaviour happens only, when starting emacs in a directory under
> monotone version controll.

Thanks.  I believe I've fixed it with a seemingly unrelated commit in
vc.el.  Apparently the real bug is somewhere in kill-buffer (it may
change current-buffer in some unexpected circumstances), but that one
will have to wait a little.


        Stefan

Re: vc-mtn bug?

[Sebastian Rose]

Ahh, OK, thanks a thousand times.

So it will be in CVS head the next weeks or something? Or is it already?

Regards,

        Sebastian