Ian Lance Taylor wrote: > Sebastian Biallas <sb@xxxxxxxxxxx> writes: > >> My question is: Which command line switch of gcc is preferred for adding >> include directories? > > -I Yes, I though so, too. But there is a problem with it, see below. > >> And: What's the reason that -I modifies system >> headers includes? > > It doesn't. It just changes the search path used to find header > files. Yes, but it modifies the search path for "" *and* <> includes. This happens: $ ls process.h test.cc $ cat test.cc #include <unistd.h> $ cat process.h #error Wrong process.h included! $ g++ -c -mno-cygwin -o test.o test.cc $ g++ -c -I. -mno-cygwin -o test.o test.cc In file included from /usr/lib/gcc/i686-pc-mingw32/3.4.4/../../../../i686-pc-mingw32/include/unistd.h:11, from test.cc:1: ./process.h:1:2: #error Wrong process.h included! [The line 11 of unistd.h reads #include <process.h>] So, the -I. option adds '.' to the list of system header directories. That means if I have a file which accidently has some "forbidden" name which is used somewhere deep in the system headers, everything breaks. I'm not sure if this is actually a (cygwin) bug or a feature, but this certainly makes -I unusable for me... > If you want to do something more complex, see the documentation for > -iquote and -isystem. They probably don't do what you want, but they > may be instructive. -iquote is what I want, but: It changed syntax recently (was -I- before). So it's not (easily) possible to write a simple makefile for use with gcc3 and gcc4. And automake adds -I. by default.
