Downsides of scanning all .o files for typedefs

I'd been getting weird results for the last couple of days while
pgindent-ing various patches. I eventually realized that the cause
was that the current typedefs list marks "c", "string", and a few
other common words as typedefs. This seems pretty uncool. Further
investigation shows that the reason is that these names are used as
typedefs in a couple of the ecpg regression tests; which the old
find_typedefs code never picked up on, but the OS X implementation
does.

Now, it's actually rather pointless to collect typedef names from
the ecpg tests, since pgindent won't process files with .pgc
extensions anyway (and I doubt it would work well to try).

So we could either revise these test cases to use less-generic
typedef names, or we could just skip ecpg/test/ in find_typedefs.
For the moment I've got dromedary using the attached quick-hack patch
to do the latter. Any thoughts on the best long-term answer?

regards, tom lane