I'm convinced mod_include is at fault regarding this problem between it
and the new table api.  My reason is that mod_include makes the assumption
that it is the *only* module which is using run_sub_request()... which is
bogus.  If another module were to run_sub_request() on something which
invoked mod_include then it would muck around with the environment of a
request it shouldn't be touching. 

I have an idea how to deal with it.  My fix will allow #included SSI
documents to modify the environment, but won't allow any other type of
include to modify the environment.  To draw a parallel with /bin/sh,
"#include" of an SSI document is similar to "source" of another shell
script.  But you can't "source" a binary program and expect it to modify
the environment of your shell. 

By way of contrast, apache-1.2 and earlier allows a #included object to
muck with the environment.  (And even that can be buggy for other
reasons). 

Comments? 

Dean