If anything in start_new_log did logging (for example, by sending a
statustext), we end up infinitely recursing.
With the patch:
diff --git a/libraries/AP_Logger/AP_Logger_File.cpp b/libraries/AP_Logger/AP_Logger_File.cpp
index 69b8ef0431..eb422d10f8 100644
--- a/libraries/AP_Logger/AP_Logger_File.cpp
+++ b/libraries/AP_Logger/AP_Logger_File.cpp
@@ -778,6 +778,7 @@ void AP_Logger_File::PrepForArming()
*/
void AP_Logger_File::start_new_log(void)
{
+ gcs().send_text(MAV_SEVERITY_WARNING, "Starting new log");
stop_logging();
start_new_log_reset_variables();
pbarker@bluebottle:~/rc/ardupilot(master)$
We see:
at ../../libraries/AP_Logger/AP_Logger_File.cpp:781
this=0x555555ad9d30, pBuffer=0x7fffff8209d0, size=75, is_critical=true)
at ../../libraries/AP_Logger/AP_Logger_Backend.cpp:372
this=0x555555ad9d30, pBuffer=0x7fffff8209d0, size=75)
at ../../libraries/AP_Logger/AP_Logger_Backend.h:32
this=0x555555ad9d30, message=0x7fffff820b10 "Starting new log")
at ../../libraries/AP_Logger/LogFile.cpp:466
this=0x555555a6d758 <copter+11384>,
message=0x7fffff820b10 "Starting new log")
at ../../libraries/AP_Logger/AP_Logger.cpp:752
this=0x555555a6e708 <copter+15400>, severity=MAV_SEVERITY_WARNING,
fmt=0x5555557d64d0 "Starting new log", arg_list=0x7fffff820be0,
dest_bitmask=1 '\001') at ../../libraries/GCS_MAVLink/GCS_Common.cpp:1847
this=0x555555a6e708 <copter+15400>, severity=MAV_SEVERITY_WARNING,
fmt=0x5555557d64d0 "Starting new log", arg_list=0x7fffff820be0)
at ../../libraries/GCS_MAVLink/GCS.cpp:53
this=0x555555a6e708 <copter+15400>, severity=MAV_SEVERITY_WARNING,
fmt=0x5555557d64d0 "Starting new log")
at ../../libraries/GCS_MAVLink/GCS.cpp:60
at ../../libraries/AP_Logger/AP_Logger_File.cpp:781
this=0x555555ad9d30, pBuffer=0x7fffff820dc0, size=75, is_critical=true)
at ../../libraries/AP_Logger/AP_Logger_Backend.cpp:372
I'm not aware of any instances in the code where this will actually
happen - but it could easily sneak in.
These appear to have been replaced with instance fields on the UBX1 and
UBX2 messages.
Also mark the Instance field on UBX1, UBX2 as an instance field.
Instead of creating a MessageWriter explicitly to log uploaded missions
we fiddle the state of the one stored in the startup message writer.
This has the advantage of being much more likely to log the entire
mission (as the buffer constraint is removed), and with the addition of
a time-remaining check much less likely to cause a timing glitch.
LogStructure: add MAG log information
LogStructure: add MODE log information
LogStructure: add PID log information
AP_Logger: add ATT information
AP_Logger: add documentation for BAT log message
LogStructure: add RCIN and RCOU messages
AP_Logger: add VIBE message information
These are really "shouldn't happen", and only happen at startup, so
impact of the memory leak is quite limited.
Fixing this requires putting in a virtual destructor on the
MessageWriter class - this and the calls add up to several hundred bytes
of flash.
