Chaps,
thanks for all the feedback on this one. I think I have a better idea of
what's going on (except for why missing_dir isn't set, although looking at
a few example configure files it seems everyone defines the value for that
themselves, it may have been part of the AM_MISSING_PROG macro somewhile
back, but has now been removed according to the change log).
The basic point I missed is that running configure on a distribution state
package doesn't require f/lex, as the C source file is distributed, so the
portability of that file is important.
So the advice seems to boil down to only distribute C source files that
have been generated by flex - ones generated by lex may not be portable.
To support this stance the "missing" script just looks for pre-generated
C source based on the name of the lex file and copies this to lex.yy.c.
If no C source can be found it creates a dummy file with an empty main
function in it (which is what I saw on Solaris). It doesn't attempt to run
lex.
So the snippet of code that Tim inserted means we cannot use lex and
asserts that we shouldn't.
Not being a lex expert I've no idea if that is a stance we want to take.
Cheers,
Peter.
|