Python crashes after attaching pty to a konsole kpart

[Martin Böhm]

Binary package hint: python-kde3

Python crashes when a pty is attached to a konsole KPart.

The code snippet that causes the crash:

        (self.master, self.slave) = pty.openpty()
        self.konsole.setPtyFd(self.master) # this line triggers it

The result:

kdecore (KProcess): WARNING: setPty()
kdecore (KProcess): WARNING: _attachPty() 10
In file tools/qgvector.cpp, line 176: Out of memory
KCrash: Application 'gd-test.py' crashing...

Affected applications:
* kubuntu dist-upgrade tool (update-manager)
* kubuntu gdebi tool (gdebi) -- to be released for Gutsy

[Martin Böhm (martin.bohm)]

On my older machine ( Pentium III, 320MB RAM ), it does crash every time for:
 * Kubuntu 7.04 with KDE 3.5.6
 * Kubuntu 7.04 with KDE 3.5.7

and it won't crash for:
 * Kubuntu 6.10 with pre-update repositories

[Scott Kitterman (kitterman)]

I took a bit of a look at this and have some ideas.

Would you please attach your gd-test.py program to the bug so I can test without having to recreate it.

[Martin Böhm (martin.bohm)]

Uploading a file that should be enough to test the error ( I'm not on an affected machine, so someone has to test it).

[Scott Kitterman (kitterman)]

Thanks for the test case. That does it.

[Martin Böhm (martin.bohm)]

I fixed a forgotten "import" in the test case that made it dependent on gdebi-kde.

[Scott Kitterman (kitterman)]

I'm pretty sure this is in Konsole somewhere and I'm equally sure I won't be able to fix it.

[Frode M. Døving (frode)]

from gutsy qt, tools/qgvector.cpp:

QGVector::QGVector( uint size )·· · // create vectors with nullptrs
{
    len = size;
    numItems = 0;
    if ( len == 0 ) {· · · · // zero length
· vec = 0;
· return;
    }
    vec = NEW(Item,len);
    Q_CHECK_PTR( vec ); // this is line 176
    memset( (void*)vec, 0, len*sizeof(Item) );· // fill with nulls
}

[Martin Böhm (martin.bohm)]

Explanation for the "Out of memory" message, quoting Qt Docs:

void Q_CHECK_PTR ( void * p )
  If p is 0, prints a warning message containing the source code file name and line number, saying that the program ran out of memory.

[Scott Kitterman (kitterman)]

That would match my theory that the oom message is a symptom not where the
root cause is.

[Scott Kitterman (kitterman)]

Just to keep things interesting...

The edgy-updates SRU had a kdelibs part too. Here is the diff from the pty related patch from edgy-updates to gutsy. Someone who actually understands the code is going to have to peck through it and see if there are any gems.

[Jonathan Riddell (jr)]

Postponing, this is hardly a new bug, it doesn't affect everyone and we'll look at it after feature freeze.

[Jonathan Riddell (jr)]

C++ test app
http://kubuntu.org/~jriddell/konsole/kdebase_pty_test.cc

Python test app
http://kubuntu.org/~jriddell/konsole/konsole.py
http://kubuntu.org/~jriddell/konsole/qt-inst.py

[Martin Böhm (martin.bohm)]

After extensive testing and patching, I have discovered that the bug is located in the kdebase/konsole/konsole/TEHistory.cpp file. Reverting it to pre-feisty state helps.

[Henrik Nilsen Omma (henrik)]

Moving milestone to beta.

[Martin Böhm (martin.bohm)]

This diff of TEHistory.cpp and TEHistory.h (between KDE 3.5.7 a 3.5.5) contains the problematic code.

[Jonathan Riddell (jr)]

I've uploaded to gutsy and there havn't been any problems.

Since this is needed for the dist upgrade tool I've prepared a patch for feisty-proposed, attached.

[Scott Kitterman (kitterman)]

I ran Martin's test script that had (before) reproduced the problem on my Gutsy machine and it now runs correctly. If someone can provide a Feisty Konsole package to test, I can test on the same machine there.

[Martin Pitt (pitti)]

Changelog: Can you please use the standard "LP: #1234" syntax and describe the patch and its rationale a bit? (This will be read by users upgrading their stable releases).

The diff is quite intrusive, and I am not able to verify it with eyeballing. However, this more or less seems to be a revert to a previous state of the code, so if it has been tested in gutsy, I am fine with it.

[Scott Kitterman (kitterman)]

The proposed change solves the problem on my KDE 3.5.6 Feisty machine. Ran the mhb's script before the update and got the OOM crash. Ran it after with no problem.

[Scott Kitterman (kitterman)]

Marking High for Feisty because dist-upgrades to Gutsy on low memory machines (RAM + swap < ~ 1.5 GB) will fail.

[Jonathan Riddell (jr)]

uploaded to feisty-proposed

[James D. Freels (freelsjd)]

I now have this problem in feisty due to an upgrade to kdebase et al on 09/26 (3.5.6-0ubuntu20.4). I would rather not upgrade to gutsy at the moment and my konsole is NOT usable. Crashes with the following error message:

konsole
kdecore (KProcess): WARNING: _attachPty() 11
KCrash: Application 'konsole' crashing...

when opened from gnome-terminal !

[Scott Kitterman (kitterman)]

That's a different bug than this one.

[Thanos (r-lemme)]

>>>> Marking High for Feisty because dist-upgrades to Gutsy on low memory machines (RAM + swap < ~ 1.5 GB) will fail.

I have the same error with 3Gb ram + 2Gb swap

[Brian Murray (brian-murray)]

I verified that the test cases provided no longer crash with konsole version 3.5.6-0ubuntu20.3. I used both Martin's gd-test.py and Riddell's konsole.py test applications. They both caused crashes with konsole version 3.5.6-0ubuntu20.4.

[Jonathan Riddell (jr)]

package moved to -updates

[Matevž Jekovec (matevz-jekovec)]

I'm still having out-of-RAM crashes in kubuntu gdebi. Using the latest updated gutsy.

[Matevž Jekovec (matevz-jekovec)]

I have 512MB of physical RAM and 700MB swap.
gdebi-kde returns
kdecore (KProcess): WARNING: setPty()

kdecore (KProcess): WARNING: _attachPty() 27

before crashing.

[Scott Kitterman (kitterman)]

Probably a different bug. Please file it against thar package.