In stringobject.c and unicodeobject.c, the formatting codes for 'e', 'E', 'f', 'F', 'g', and 'G' all get routed to formatfloat().  formatfloat() is essentially identical in stringobject.c and unicodeobject.c.

'F' is mapped to 'f'.  There is a test for excess precision for 'f' and 'g', but not one for 'G'.

The "type == 'g'" test should be "(type == 'g' || type == 'G')".

I assume this bug is in 2.5 and 2.6 as well, although I haven't verified that, yet.