In D13018#266137, @elvisangelaccio wrote:

Hmm, still seems to fail here:

$ ctest -R kfileplacesmodeltest
Test project /home/elvis/dev/kde/kio/local-build
    Start 50: kiofilewidgets-kfileplacesmodeltest
1/1 Test #50: kiofilewidgets-kfileplacesmodeltest ...***Failed    0.60 sec
********* Start testing of KFilePlacesModelTest *********
Config: Using QtTest library 5.11.0, Qt 5.11.0 (x86_64-little_endian-lp64 shared (dynamic) release build; by GCC 8.1.0)
QDEBUG : KFilePlacesModelTest::initTestCase() void Solid::Backends::Fake::FakeManager::parseMachineFile() Parsing fake computer XML:  "/home/elvis/dev/kde/kio/autotests/fakecomputer.xml" 

PASS   : KFilePlacesModelTest::initTestCase()
FAIL!  : KFilePlacesModelTest::testInitialState() Compared values are not the same
   Actual   (m_places->rowCount()): 6
   Expected (4)                   : 4
   Loc: [../autotests/kfileplacesmodeltest.cpp(200)]
QDEBUG : KFilePlacesModelTest::testInitialList() Expected: ("/home/elvis", "/", "trash:/", "remote:/", "/media/nfs", "/foreign", "/media/floppy0", "/media/XO-Y4", "/media/cdrom")
QDEBUG : KFilePlacesModelTest::testInitialList() Got: ("/home/elvis", "/home/elvis/Desktop", "/home/elvis/Downloads", "/", "trash:/", "remote:/")
FAIL!  : KFilePlacesModelTest::testInitialList() Compared lists have different sizes.
   Actual   (placesUrls()) size: 6
   Expected (urls) size: 9
   Loc: [../autotests/kfileplacesmodeltest.cpp(236)]

Sadly the commit went without the GIT_SILENT, sorry about that. I did:
arc amend
git commit -amend # to add the GIT_SILENT
arc land
I guess that the arc land called another arc amend.

Please let me know if it's okay to land this kind of changes directly.

Btw, Debian users confirmed that this fixes the issue: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=897687

I'm not sure, but I think that pam could be called from a process that has the stdin/stdout/stderr redirected to pipes, so disconnecting those for a long lived process would avoid a possible broken pipe. Now, if that's so, why it's not clossing stdin and stdout, I don't know.

Update comments as suggested.

I can reproduce the issue in openbox and cinnamon. I think this is related to qt5ct which always tries to print things to stderr. I guess having the debug messages enabled would also trigger the issue.

A somewhat simpler version, just delay the close(2) call. Based on @davidedmundson
suggestion.

Mmh, the accept revision doesn't work as a +1, does it?
I was intending to say +1/thumbs up, but I would still prefer somebody else to review this. After all, I sent forwarded the original patch, clearly I want this to land, but it's up the frameworks/ktexteditor developers/maintainers to decide.

In D12513#253537, @fvogt wrote:

There's a typo in the title, it should be "privilege escalation".

Why dropping syncToDisk? Why is that related to the current issue?

You are planning a bit ahead, aren't you. :)

This commit fixed the Debian bug reported in: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=845553