[mythtv-commits] Ticket #13647: Backend segfaults (probably in JobQueue)

[MythTV]

#13647: Backend segfaults (probably in JobQueue)
--------------------------------+-------------------------------
 Reporter:  warpme              |          Owner:  David Hampton
     Type:  Bug Report - Crash  |         Status:  assigned
 Priority:  minor               |      Milestone:  31.1
Component:  MythTV - General    |        Version:  Master Head
 Severity:  medium              |     Resolution:
 Keywords:                      |  Ticket locked:  0
--------------------------------+-------------------------------
Changes (by David Hampton):

 * owner:  (none) => David Hampton
 * status:  new => assigned
 * milestone:  needs_triage => 31.1


Comment:

 This looks like it was caused by my change in [3a94e530ec], but its got me
 scratching my head.

 Thread !#1, Frame !#2 is a call to ProgramInfo::!SubstituteMatches with
 the command string for a job. That calls QDateTime::toLocalTime (Frame
 !#1) with this=0x7f8e0c00ad70. Looking at the time_dtr array, that pointer
 is the zeroth entry in the array of pointers to !DateTime objects, so this
 matches with the value of i=0 and looks correct.

 The pointers that array match up with what I would expect to see given the
 !ProgramInfo class definition  A QDateTime is 8 bytes in size; basically
 just a pointer to private data.  The addresses in the time_dtr array go up
 by 8 from 0th to 1st entry, down by 24, and up by 8. This is because the
 order of the date time fields don't match that of the class definition,
 but the four objects are all adjacent in memory.  I.E., they are in order
 3,4,1,2 compared to the !DateTime objects in the class definition.

 Looking at the qdatetime.h include file, the call to toLocalTime does
 nothing other than call  toTimeSpec(Qt::!LocalTime), so that tracks with
 the call from Frame !#1 to Frame !#0.

 I don't understand why this function is failing, unless the pointer to the
 QDateTime object is bad, which makes no sense. The commit changed the
 array "wrapper" from a C array to a C++ array, but the declaration of the
 contents of the array didn't change.

 What's the command line for your user job !#1? That's what's being kicked
 off when the system crashes? What platform is crashing? OSX or something
 ARM based?

 I added some test code to MythTV to exercise the
 ProgramInfo::!SubstituteMatches function where the crash is occurring. It
 works for me on both Linux and OSX. Can you pull the latest code and run a
 'make test' and see if it passes?

-- 
Ticket URL: <https://code.mythtv.org/trac/ticket/13647#comment:1>
MythTV <http://www.mythtv.org>
MythTV Media Center