Re: dependencies in other folders

[Eduardo Suárez]

Philip Guenther wrote:
> Right.  Let's look at the error message:
>   make: stat: ../ab/../ab/../ab/../ab/../<elided>/ab/../ab/../ab/foo.c:
> File name too long
>   
...
> So, the problem is that the pattern rules results in a prerequisite
> that matches the same rule again *and* refers to the same file.  The
> other pattern rule doesn't have that problem with 'bar.c' because the
> prerequisite is at a different directory level, so when the rule tries
> to match again it fails because the prereq doesn't exist.
>   
I thought that being in other directory, it wouldn't match the target. 
Sorry for bothering that.
> The first consideration in solving this is whether you can remove
> these rules and instead just use the VPATH variable or the vpath
> directive.  You should read chapter 4.5 "Searching Directories for
> Prerequisites" in the info pages for the details.  That's probably the
> simplest and cleanest way to solve this.
>   
Let's say target and prerequisite have the same extension (e.g. think of 
an 'sed' processing):
################# START HERE
mkdir test_make test_make/a test_make/b
touch test_make/b/foo.c
cat<<EOF | tr "!" "\t" > test_make/a/Makefile
VPATH=../b
%.c: %.c
!cp \$< \$@
EOF
cd test_make/a
make foo.c
################# END HERE

Then I get:
make: Circular foo.c <- foo.c dependency dropped.
make: `../b/foo.c' is up to date.

Is it possible to solve it using VPATH?
> If there's some insurmountable blocker to using VPATH or vpath, then I
> would consider changing the pattern rules to have two colons.  This
> makes them 'terminal' rules, so that make will only apply them if the
> prerequisite file exists and make won't try to build that prerequisite
> file from something else.  If the source files (../ab/foo.c and
> ../../b/bar.c) need to themselves be built as part of the run, then
> this won't work.
>   
With double colon it seems to work pretty fine.

Thanks a lot for your answer,
-Eduardo