Memory corruption in mythpreviewgen if mpg file contains errors

[Lawrence Rust <lvr@…>]

I have noticed several segfaults and aborts logged by mythbackend when running mythpreviewgen. It appears that problems occur if the recording contains errors in the transport stream due to reception problems. I can reliably reproduce the problem with this recording ​http://www.softsystem.co.uk/download/mythtv/7941_20111029115050.mpg (30MB) when invoked as: mythpreviewgen -v most --infile 7941_20111029115050.mpg

Backtrace attached.

NB there's a bug in mythpreviewgen/main.cpp that needs this patch to specify a filename:

--- a/mythtv/programs/mythpreviewgen/main.cpp
+++ b/mythtv/programs/mythpreviewgen/main.cpp
@@ -98,7 +98,7 @@ int preview_helper(const QString &_chanid, const QString &star
 
     uint chanid = _chanid.toUInt();
     QDateTime recstartts = myth_dt_from_string(starttime);
-    if (!chanid || !recstartts.isValid())
+    if (!QFileInfo(infile).isReadable() && (!chanid || !recstartts.isValid()))
         ProgramInfo::ExtractKeyFromPathname(infile, chanid, recstartts);
 
     ProgramInfo *pginfo = NULL;

[danielk]

First thanks for the report.

However if this patch fixes anything it appears that may just a side-effect of breaking something else. As I read it the logic is saying if the chanid or recording start time are not valid and the file is not readable then we should extract the chanid and recording start time from the filename. But it would seem to me that it is immaterial at this point whether the file is readable and moreover if it is you would want to extract those keys so you could generate a preview.

I can't download the sample so I can't reproduce the underlying issue and the backtrace doesn't pinpoint the problem issue well enough in this case. If you still have the sample please make it available again and I'll reopen the ticket and look at the underlying issue here.