They have no compiler, only an accretor that gloms additional code onto the existing binary. I use the word binary loosely; it is not uncommon for them to "improve" fundamentally flawed data structures by moving to a larger base notation - to trinary, then quadrary, etc. At this point, some of their products are in base 17.
They never go back to fix bugs where they occur. They write new code to workaround the earlier failure case. I asked why they don't go back and just fix the bug where it happens. I was told "We can't go back and change it. That code's already done!" Their solution for insuring that failing code will be able to get to its workaround is the GOTO statement. GOTO is sprinkled liberally around other code, pointing to functions and routines that do not exist yet. If, down the road, it is discovered that the old code has a bug, they find out which GOTOs exist in that code that do not point to anything yet, pick one, and write the workaround there.
I could go on, but I am being told that we need to celebrate the successful compilation (by which I mean accretion) of a particularly complex workaround for a bug that has been known about for two years.
http://crisper.livejournal.com/316634.html#cutid1