By using this site, you agree to our updated Privacy Policy and our Terms of Use. Manage your Cookies Settings.
443,719 Members | 1,875 Online
Bytes IT Community
+ Ask a Question
Need help? Post your question and get tips & solutions from a community of 443,719 IT Pros & Developers. It's quick & easy.

Python/Tkinter/tk crash [long]

P: n/a
Hi all,

I was creating a Tkinter widget in the style of the reversed tabs below Excel
worksheets and I stepped in a serious problem: the code I made makes python
crash with a seg fault, bus error or X11 BadGC error on both Solaris (2.6 and
2.7) and Linux (Mandrake 8.0); it doesn't crash on Windows. I tried to simplify
the script, but I couldn't reproduce the crash with a simpler code. So the code
below is somewhat long; sorry for that.

To make it crash, just run the script and click on a tab. It usually crashes at
the first click, but you may have to play a bit with the tabs. I tried to run
Python through gdb to see where the crash happens, but it seems to be quite
random. Maybe a memory corruption?

There's a simple workaround, but with drawbacks: at the beginning of the
__update method, if I do not destroy and re-create the Canvas, but simply empty
its contents, the script works. But it keeps the commands declared created at
the tcl level for the former bindings, so it silently eats up memory.

My setup is Python 2.1 with tcl/tk 8.3.4. I searched the bug database but this
bug doesn't seem to be known. But maybe it was corrected in a newer Python or tk
version? Can anyone confirm that?

Thanks a lot in advance.

Here is the code:

--TabRow.py----------------------------------
from Tkinter import *
## CLASS GENERIC.CALLBACK:
## =======================
# Instances are generic callbacks for buttons, bindings, etc...

class GenericCallback:

def __init__(self, callback, *commonArgs):
self.callback = callback
self.__commonArgs = commonArgs

def __call__(self, *args):
return apply(self.callback, self.__commonArgs + args)

## CLASS TAB.ROW:
## ==============
# Excel-style reverse tabs in a row

class TabRow(Frame):

## CLASS ATTRIBUTES:
## -----------------

defaultHeight = 24 # Default height for row
verticalMargin = 2 # Margin at the top and the bottom of the tabs
tabBorderWidth = 10 # Width for the descending and ascending lines at the
# border of tabs
## METHOD __INIT__:
## ----------------
# Constructor
# Parameters: the ones for the Frame class
# Recognized options: the ones for the Frame class +
# - tabs: list of tabs
# - currenttab: active element in tabs, or its index
# - tabchangecallback: function called when the user clicks on a tab
# (1 param = text; returns a boolean)
# - font: the font to use to display the texts in the tabs
# - activebackground: the background for the current tab

def __init__(self, *args, **options):

## INSTANCE ATTRIBUTES:
## --------------------

self.__tabs = ()
self.__tabChangeCallback = None
self.__tabsCanvas = None
self.__currentTabIndex = 0
self.__tabPositions = None
self.__canvasHeight = TabRow.defaultHeight

## Super-init
apply(Frame.__init__, (self,) + args)
## Configure/update
self.configure(options)
## METHOD CONFIGURE:
## -----------------
# Changes the options for the tab row

def configure(self, dictOptions={}, **options):
options.update(dictOptions)
## Get specific options
self.__tabs = options.get("tabs", self.__tabs)
self.__tabChangeCallback = options.get('tabchangecallback',
self.__tabChangeCallback)
## Get index for current tab
if options.has_key('currenttab'):
if type(options['currenttab']) == type(0):
self.__currentTabIndex = options['currenttab']
else:
indices = [i for i in range(len(self.__tabs))
if self.__tabs[i] == options['currenttab']]
if indices: self.__currentTabIndex = indices[0]
## Remember forced height for canvas if any
self.__canvasHeight = options.get('height', self.__canvasHeight)
## Remove unwanted options
needUpdate = 0
for o in ('tabs', 'currenttab', 'tabchangecallback',
'font', 'activebackground', 'height'):
if not options.has_key(o): continue
del options[o]
needUpdate = 1
if options.has_key('bg') or options.has_key('background'): needUpdate = 1
## If needed, apply options on the frame
if options:
apply(Frame.configure, (self,), options)
## If needed, update display
if needUpdate: self.__update()
## METHOD __UPDATE:
## ----------------
# Updates the display

def __update(self):
## (Re)create canvas for tabs
if self.__tabsCanvas is not None:
self.__tabsCanvas.grid_forget()
self.__tabsCanvas.destroy()
self.__tabsCanvas = Canvas(self, bg=self.cget('background'),
height=self.__canvasHeight)
self.__tabsCanvas.grid(row=0, column=0, sticky='nswe')
## Build tabs
tabIndex, pos = 0, 0
self.__tabPositions = []
activeTabRight = 0
for text in self.__tabs:
## Standard tag + specific tag if tab is the current one
tabTag = 'TAB_%s' % tabIndex
tags = [tabTag]
if tabIndex == self.__currentTabIndex: tags.append("CURRENT_TAB")
tags = tuple(tags)
## Remember tab position
self.__tabPositions.append(pos)
## Draw text
textId = self.__tabsCanvas.create_text(pos + TabRow.tabBorderWidth,
self.__canvasHeight / 2,
text=text, anchor=W, tags=tags,
font=('helvetica', 10, 'bold'))
## Polygon for tab, including line from left side if current tab
textBBox = self.__tabsCanvas.bbox(textId)
x = textBBox[2]
coords = [
pos, TabRow.verticalMargin,
pos + TabRow.tabBorderWidth,
self.__canvasHeight - TabRow.verticalMargin,
x, self.__canvasHeight - TabRow.verticalMargin,
x + TabRow.tabBorderWidth, TabRow.verticalMargin ]
if tabIndex == self.__currentTabIndex:
coords = [0, TabRow.verticalMargin] + coords
activeTabRight = x + TabRow.tabBorderWidth
## Get polygon background
polygOpt = {'fill' : self.__tabsCanvas.cget('background'),
'outline':'', 'tags':tags}
if tabIndex == self.__currentTabIndex: polygOpt['fill'] = 'white'
## Draw polygon
polygId = apply(self.__tabsCanvas.create_polygon, coords, polygOpt)
lineId = apply(self.__tabsCanvas.create_line, coords,
{'fill':'black', 'tags':tags})
## Put it under text
self.__tabsCanvas.lower(lineId)
self.__tabsCanvas.lower(polygId)
## Binding for tab change
self.__tabsCanvas.tag_bind(tabTag, '<ButtonRelease-1>',
GenericCallback(self.__changeTab, tabIndex, text))
## Update position and tab index
pos = x + TabRow.tabBorderWidth / 2
tabIndex += 1
## End of display: draw line from active tab to right border
## and put active tab on top
self.__tabsCanvas.create_line(activeTabRight, TabRow.verticalMargin,
pos + TabRow.tabBorderWidth / 2,
TabRow.verticalMargin)
self.__tabsCanvas.tag_raise("CURRENT_TAB")
## METHOD __CHANGE.TAB:
## --------------------
# Called when the user clicks on a tab

def __changeTab(self, tabIndex, text, event=None):
if self.__tabChangeCallback is None: return
if not self.__tabChangeCallback(text): return
self.__currentTabIndex = tabIndex
self.__update()

if __name__ == '__main__':
root = Tk()

def ct(t):
print t
return 1

r = TabRow(root, tabs=('foo', 'bar', 'spam'), tabchangecallback=ct)
r.pack()
root.mainloop()
---------------------------------------------
--
- Eric Brunel <er*********@pragmadev.com> -
PragmaDev : Real Time Software Development Tools - http://www.pragmadev.com

Jul 18 '05 #1
Share this Question
Share on Google+
8 Replies


P: n/a
On Thursday 14 August 2003 10:10 am, Eric Brunel wrote:
Hi all,

I was creating a Tkinter widget in the style of the reversed tabs below
Excel worksheets and I stepped in a serious problem: the code I made makes
python crash with a seg fault, bus error or X11 BadGC error on both Solaris
(2.6 and 2.7) and Linux (Mandrake 8.0); it doesn't crash on Windows. I
tried to simplify the script, but I couldn't reproduce the crash with a
simpler code. So the code below is somewhat long; sorry for that.

Eric,

Have you looked at Pmw (Python Mega Widgets) they have a NoteBook (tabbed
pane) widget that looks like what you want...
Regards
Martin

Jul 18 '05 #2

P: n/a
I can duplicate the signal 11 on RedHat Linux 9 and the following
relevant packages installed:
python-2.2.2-26
tcl-8.3.5-88
tk-8.3.5-88

My traceback looks like the following. I believe that the lines listed
as "Tk_GetItemTypes" actually correspond to static functions defined in
tkCanvas.c.
#0 0x4011b1a7 in Tk_GetItemTypes () from /usr/lib/libtk8.3.so
#1 0x4011aec1 in Tk_GetItemTypes () from /usr/lib/libtk8.3.so
#2 0x4011ac25 in Tk_GetItemTypes () from /usr/lib/libtk8.3.so
#3 0x400d1a7c in Tk_HandleEvent () from /usr/lib/libtk8.3.so
#4 0x400d1e7c in TkQueueEventForAllChildren () from /usr/lib/libtk8.3.so
#5 0x401c300d in Tcl_ServiceEvent () from /usr/lib/libtcl8.3.so
#6 0x401c326d in Tcl_DoOneEvent () from /usr/lib/libtcl8.3.so
#7 0x40062115 in Tkapp_MainLoop (self=0x815dfb0, args=0x8269358) at Modules/_tkinter.c:1696
#8 0x080d0df4 in PyCFunction_Call ()
#9 0x0807a65e in PyEval_EvalCode ()
#10 0x0807b0ce in PyEval_EvalCodeEx ()
#11 0x0807c62b in PyEval_GetFuncDesc ()
#12 0x0807a5a3 in PyEval_EvalCode ()
#13 0x0807b0ce in PyEval_EvalCodeEx ()
#14 0x08077fc5 in PyEval_EvalCode ()
#15 0x08097e29 in PyRun_FileExFlags ()
#16 0x08096d90 in PyRun_SimpleFileExFlags ()
#17 0x080966da in PyRun_AnyFileExFlags ()
#18 0x08053a19 in Py_Main ()
#19 0x08053469 in main ()
#20 0x42015574 in __libc_start_main () from /lib/tls/libc.so.6

This leads me to two things: first, reproduce this using a libtk
compiled with debugging information. second, this is likely to be a Tk
bug and not a Python bug given where it happens. It should be possible
to write a test-case that is purely tcl code..

In fact, the following tcl code crashed on me once (not in gdb, so I
can't compare stack traces), and when I run it under valgrind the *first*
click I make sprays "invalid read" and "invalid write" errors such as
==23796== Invalid write of size 4
==23796== at 0x4026ACEB: (within /usr/lib/libtk8.3.so)
==23796== by 0x4026AC24: (within /usr/lib/libtk8.3.so)
==23796== by 0x40221A7B: Tk_HandleEvent (in /usr/lib/libtk8.3.so)
==23796== by 0x40221E7B: (within /usr/lib/libtk8.3.so)
==23796== Address 0x41493D98 is 188 bytes inside a block of size 460 free'd
==23796== at 0x40161048: free (in /usr/lib/valgrind/valgrind.so)
==23796== by 0x402D8368: TclpFree (in /usr/lib/libtcl8.3.so)
==23796== by 0x402DD4F4: Tcl_Free (in /usr/lib/libtcl8.3.so)
==23796== by 0x402685C5: (within /usr/lib/libtk8.3.so)

Now, in the python and wish cases, the shared libraries are loaded at
different addresses. but note that the number of bytes between
TK_HandleEvent and the next stack frame inwards (called Tk_GetItemTypes
by gdb) is the same distance on my system.

Jeff

# --- tk program kinda like original python program
proc p {} {
puts p
destroy .c
canvas .c
grid .c

set tabIndex 0
set pos 0
foreach p { a b c } {
set tags[list TAB_$tabIndex]
if {$tabIndex == 1} {
lappend tags CURRENT_TAB
}
set textId [.c create text $pos 10 -text $p -anchor w -tags $tags]
.c bind TAB_$tabIndex <ButtonRelease> p
incr tabIndex 1
incr pos 20
}
}
p
# ---

Jul 18 '05 #3

P: n/a
Martin Franklin wrote:
On Thursday 14 August 2003 10:10 am, Eric Brunel wrote:
Hi all,

I was creating a Tkinter widget in the style of the reversed tabs below
Excel worksheets and I stepped in a serious problem: the code I made makes
python crash with a seg fault, bus error or X11 BadGC error on both Solaris
(2.6 and 2.7) and Linux (Mandrake 8.0); it doesn't crash on Windows. I
tried to simplify the script, but I couldn't reproduce the crash with a
simpler code. So the code below is somewhat long; sorry for that.


Eric,

Have you looked at Pmw (Python Mega Widgets) they have a NoteBook (tabbed
pane) widget that looks like what you want...


I do know and use extensively this package. The behaviour I'd like to implement
is somewhat different than the one in Pmw.NoteBook. Please also note that
despite its length, the code I posted is only an extract of the actual code.

Thanks anyway.
--
- Eric Brunel <er*********@pragmadev.com> -
PragmaDev : Real Time Software Development Tools - http://www.pragmadev.com

Jul 18 '05 #4

P: n/a
Jeff Epler wrote:
I can duplicate the signal 11 on RedHat Linux 9 and the following
relevant packages installed:
python-2.2.2-26
tcl-8.3.5-88
tk-8.3.5-88

My traceback looks like the following. I believe that the lines listed
as "Tk_GetItemTypes" actually correspond to static functions defined in
tkCanvas.c.
#0 0x4011b1a7 in Tk_GetItemTypes () from /usr/lib/libtk8.3.so
#1 0x4011aec1 in Tk_GetItemTypes () from /usr/lib/libtk8.3.so
#2 0x4011ac25 in Tk_GetItemTypes () from /usr/lib/libtk8.3.so
#3 0x400d1a7c in Tk_HandleEvent () from /usr/lib/libtk8.3.so
#4 0x400d1e7c in TkQueueEventForAllChildren () from /usr/lib/libtk8.3.so
#5 0x401c300d in Tcl_ServiceEvent () from /usr/lib/libtcl8.3.so
#6 0x401c326d in Tcl_DoOneEvent () from /usr/lib/libtcl8.3.so
#7 0x40062115 in Tkapp_MainLoop (self=0x815dfb0, args=0x8269358) at Modules/_tkinter.c:1696
#8 0x080d0df4 in PyCFunction_Call ()
#9 0x0807a65e in PyEval_EvalCode ()
#10 0x0807b0ce in PyEval_EvalCodeEx ()
#11 0x0807c62b in PyEval_GetFuncDesc ()
#12 0x0807a5a3 in PyEval_EvalCode ()
#13 0x0807b0ce in PyEval_EvalCodeEx ()
#14 0x08077fc5 in PyEval_EvalCode ()
#15 0x08097e29 in PyRun_FileExFlags ()
#16 0x08096d90 in PyRun_SimpleFileExFlags ()
#17 0x080966da in PyRun_AnyFileExFlags ()
#18 0x08053a19 in Py_Main ()
#19 0x08053469 in main ()
#20 0x42015574 in __libc_start_main () from /lib/tls/libc.so.6

This leads me to two things: first, reproduce this using a libtk
compiled with debugging information. second, this is likely to be a Tk
bug and not a Python bug given where it happens. It should be possible
to write a test-case that is purely tcl code..


Thanks a lot: you're absolutely right. The tcl code you wrote behaves exactly
the same than the Python script I sent: when I run it, the first click on any
text ends in a seg fault or a bus error. I should have tried to reproduce the
bug in pure tcl, but I'm not really fluent in it...

I'll try to figure out exactly what happens and submit the problem to the
c.l.tcl newsgroup.

Thanks!
--
- Eric Brunel <er*********@pragmadev.com> -
PragmaDev : Real Time Software Development Tools - http://www.pragmadev.com

Jul 18 '05 #5

P: n/a
Eric Brunel <er*********@pragmadev.com> wrote in message news:<bh**********@news-reader5.wanadoo.fr>...
Hi all,

I was creating a Tkinter widget in the style of the reversed tabs below Excel
worksheets and I stepped in a serious problem: the code I made makes python
crash with a seg fault, bus error or X11 BadGC error on both Solaris (2.6 and
2.7) and Linux (Mandrake 8.0); it doesn't crash on Windows.


FWIW: it crashes under Red Hat 7.2+Python 2.2 and under Red Hat 7.3+Python 2.3
too.
Jul 18 '05 #6

P: n/a
Martin Franklin wrote:
On Thursday 14 August 2003 10:10 am, Eric Brunel wrote:
There's a simple workaround, but with drawbacks: at the beginning of the
__update method, if I do not destroy and re-create the Canvas, but simply
empty its contents, the script works. But it keeps the commands declared
created at the tcl level for the former bindings, so it silently eats up
memory.


does it still eat up memory if you, for example, change the top of the
__update method to:

def __update(self):
if not self.__tabsCanvas:
self.__tabsCanvas = Canvas(self, bg=self.cget('background'),
height=self.__canvasHeight)
self.__tabsCanvas.grid(row=0, column=0, sticky='nswe')
self.__tabsCanvas.delete("all")


Yes it does: the call to delete('all') on the canvas doesn't delete the
callbacks registered at the tcl level for all the bindings done in the
canvas. These commands are only deleted when the canvas itself is deleted.

Consider the following script:

--eatmem.py---------------------------
from Tkinter import *

root = Tk()

cnv = Canvas(root)
cnv.pack()

def spam(*w): print spam

def eat():
cnv.delete(ALL)
l = cnv.create_line(10, 10, 10, 30)
cnv.tag_bind(l, '<1>', spam)
root.after(100, eat)

eat()

root.mainloop()
--------------------------------------

If you run it and monitor the size of the running process either with the
task manager on Windows or a "ps -A -o vsz -o args | grep eatmem' on Linux,
you'll notice the memory occupied by the program slowly but steadily grows
(you'll have to wait a little before it starts growing). Remove the bind, and
it doesn't happen anymore.

The same happens for commands in menus: they're only deleted when the whole
menu goes away, not when the entry having the command is deleted.

If you know something that can be done about that, I'm interested!
--
- Eric Brunel <er*********@pragmadev.com> -
PragmaDev : Real Time Software Development Tools - http://www.pragmadev.com

Jul 18 '05 #7

P: n/a
On Thu, 14 Aug 2003 09:10:22 +0000, Eric Brunel wrote:
Hi all,

I was creating a Tkinter widget in the style of the reversed tabs below Excel
worksheets and I stepped in a serious problem: the code I made makes python
crash with a seg fault, bus error or X11 BadGC error on both Solaris (2.6 and
2.7) and Linux (Mandrake 8.0); it doesn't crash on Windows. I tried to simplify
the script, but I couldn't reproduce the crash with a simpler code. So the code
below is somewhat long; sorry for that.

To make it crash, just run the script and click on a tab. It usually crashes at
the first click, but you may have to play a bit with the tabs. I tried to run
Python through gdb to see where the crash happens, but it seems to be quite
random. Maybe a memory corruption?

There's a simple workaround, but with drawbacks: at the beginning of the
__update method, if I do not destroy and re-create the Canvas, but simply empty
its contents, the script works. But it keeps the commands declared created at
the tcl level for the former bindings, so it silently eats up memory.


As far as I understand, the __changeTab method is called when an event
occurs on the __tabsCanvas. But one of the first actions of the __update
method (which is called by __changeTab) is to destroy the __tabsCanvas
object.

So DURING the callback call you are destroying the object to which the
callback is bound. I wonder if this is the cause of the blow up at the
return of the callback (it may depend on how the memory is reallocated
on different systems).

Just my 0.02 euro guess

--
Pedro
Jul 18 '05 #8

P: n/a
Pedro Rodriguez wrote:
On Thu, 14 Aug 2003 09:10:22 +0000, Eric Brunel wrote:

Hi all,

I was creating a Tkinter widget in the style of the reversed tabs below Excel
worksheets and I stepped in a serious problem: the code I made makes python
crash with a seg fault, bus error or X11 BadGC error on both Solaris (2.6 and
2.7) and Linux (Mandrake 8.0); it doesn't crash on Windows. I tried to simplify
the script, but I couldn't reproduce the crash with a simpler code. So the code
below is somewhat long; sorry for that.

To make it crash, just run the script and click on a tab. It usually crashes at
the first click, but you may have to play a bit with the tabs. I tried to run
Python through gdb to see where the crash happens, but it seems to be quite
random. Maybe a memory corruption?

There's a simple workaround, but with drawbacks: at the beginning of the
__update method, if I do not destroy and re-create the Canvas, but simply empty
its contents, the script works. But it keeps the commands declared created at
the tcl level for the former bindings, so it silently eats up memory.

As far as I understand, the __changeTab method is called when an event
occurs on the __tabsCanvas. But one of the first actions of the __update
method (which is called by __changeTab) is to destroy the __tabsCanvas
object.

So DURING the callback call you are destroying the object to which the
callback is bound. I wonder if this is the cause of the blow up at the
return of the callback (it may depend on how the memory is reallocated
on different systems).


In fact, you're absolutely right, but it still shouldn't result in a crash. I
also discovered that the problem only happens with ButtonRelease events, and not
ButtonPress ones. Apparently, the tcl interpreter doesn't properly remembers
that the pointer on the canvas shouldn't be freed until the binding is
completely executed.

I've reported the problem to the comp.lang.tcl newsgroup. I hope they'll have a
solution for it, because I can't figure out by myself what I should do to
prevent the crash (I'm not that familiar with the code for tcl/tk). For the
moment, I'll use bindings on ButtonPress's rather than ButtonRelease's.

Thanks a lot to all who answered anyway!
--
- Eric Brunel <er*********@pragmadev.com> -
PragmaDev : Real Time Software Development Tools - http://www.pragmadev.com

Jul 18 '05 #9

This discussion thread is closed

Replies have been disabled for this discussion.