Mosteo's point:
the much touted Ariane 5 blow-up was not really a software programming problem (and that's not civil aviation anyway), but a software management issue.
The low level cause of the Ariane 5 failure was indeed a SW bug:
A float to integer conversion caused an unprotected exception that in turn caused the navigation computers to report an error code rather than current position. Both failed within one or 2 master clock ticks (1 ms as I recall).
The conversion was unprotected since it was in a calibration routine that was originally not meant to be running after lift-off. This was changed to allow fast recycling in case of a last second abort.
Extensive analysis showed that the registers would not overflow during the first (45 seconds?) following lift-off that the routine continued to run.
The (mis)management issues are much more interesting though:
The system was designed with the assumption that the only hardware was subject to failure, had the SW been designed to continue to provide "I may have a problem but here is my best guess boss data" the flight would have been a success.
The real cause of the disaster though was cutting of testing due to time and budget constraints, since the code had flown previously on prior rockets it was decided that testing could be reduced.
The Arian 5 flight profile was steeper, resulting in the numerical overflow.