On 5 December 2006 at 16:13, Michael Hanke wrote:
| Hi,
|
| On Tue, Dec 05, 2006 at 08:32:54AM -0600, Dirk Eddelbuettel wrote:
| >
| > Moin,
| >
| > On 5 December 2006 at 14:29, Michael Hanke wrote:
| > | Hi Dirk,
| > |
| > | On Tue, Dec 05, 2006 at 06:36:53AM -0600, Dirk Eddelbuettel wrote:
| > | > | And python dies :(
| > | >
| > | > Not for me. Currnetly the R window is 'behind' Emacs and the (KDE)
konsole
| > | > (xterm emulator) and I can happily replot:
| > | Replotting works for me as well, but python dies if the plot window
| > | comes back to foreground again (no need to replot anything). Actually I
| > | do not know what process is dying, but the python interpreter is heavily
| > | affected.
| >
| > Ah. Didn't try that. Now I'll have to wait til I get back home from work :-/
| >
| > | > All on testing too. Not a bug in R as far I can tell (and the other
way to
| > | > tell is to open a file device (png(), pdf(), .... in R, not sure what
the RPy
| > | > way is) and plot there.
| > | Strange.
| > |
| > | This is obviously not an R bug. I can do all this with plain R and
| > | everything is fine. I tested i386 sarge, i386 and amd64 etch.
| >
| > I think we'd need something reproducible before we can and pester Greg [ rpy
| > upstream ] with this. We both agree that R is innocent in this.
| Right. However, it seems to me that whatever happens, happens in some
| place like the eventloop of the plotting window. Running python with strace
| reveals no additional call between 'everything just works' and the crash when
| moving the partially hidden window.
|
| > | But doing it from python with rpy segfaults on both testing machines,
| > | but *not* on sarge.
| > |
| > | > You may need to tweak your window manager, though....
| > | Well, I have no clue what I should tweak. I'm using the package defaults
| > | on all machines and they work well for all other windows.
| > |
| > | BTW: I tried to run it remotely from the sarge machine (connected to
amd64
| > | etch via ssh) therefore using a different xserver and the *same* window
manager
| > | that proved to work when running rpy locally. The bug is still present,
but the
| > | error message is slightly different:
| > |
| > | *** caught segfault ***
| > | address 0x16e8790, cause 'invalid permissions'
| >
| > That's weird one too.
| >
| > Could it be something having to with X11 / modules / permission /
accelerated
| > X11 / ... ?
| That's all standard as well. The sarge Xserver (xfree) is using plain vesa,
| while both etch machine use the closed-source Nvidia module (with xorg). But
the
| window crashes using both xservers, so we cannot easily put the blaim on
| Nvidia. Also, running other (even OpenGL) apps works just fine (also via
| ssh on both xservers).
|
| Just to be sure: I tried to reproduce this bug on another machine
| running etch (i386), xorg (with open-source ati driver). Same bug,
| python dies as soon as I move the half-hidden window.
|
| Here is what I used for strace:
|
| [EMAIL PROTECTED]:~$ cat rpytest.py
| import time
| from rpy import *
|
| x = range(0, 10)
| y = [ 2*i for i in x ]
| r.plot_default(x, y)
|
| time.sleep(30)
|
|
| And here is the last part of the strace output:
|
| [EMAIL PROTECTED]:~$ strace python rpytest.py
|
| [very long part stripped]
|
| rt_sigaction(SIGINT, {0x80e4ca0, [], 0}, {0xb70fb620, [INT],
SA_RESTART}, 8) = 0
| futex(0x859d638, FUTEX_WAKE, 1) = 0
| futex(0x81b98c8, FUTEX_WAKE, 1) = 1
| select(0, NULL, NULL, NULL, {30, 0}Error during wrapup: C stack usage
is too close to the limit
|
| *** caught segfault ***
| address (nil), cause 'memory not mapped'
|
|
| Please notice the select call which stems from the sleep function.
| Between this call and the crash were several seconds. It crashed exacly
| when I started to move the window.
Excellent work. Let me just re-confirm this once I get home this evening. It
does look like something is wrong when it comes redrawing the windows, so it
is probably the Python-to-R-and-back interface as we both agree that R itself
ot probably not to blame. I'll CC Greg now as an early heads-up
Greg: the whole story is at http://bugs.debian.org/401670 -- Michael found
that if a (partially or fully) hidden plot window owned by R and spawned from
RPy comes to the fore, R takes a hard landing and crashes. Let us know if
you have questions, but the mail thread between Michael and myself is fully
in the archive.
Dirk
--
Hell, there are no rules here - we're trying to accomplish something.
-- Thomas A. Edison
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
