Re: depcomp deficiency

[Alexandre Duret-Lutz]

>>> "RW" == Ralf Wildenhues <address@hidden> writes:

[...]

 RW> There are two possible ways to go: try to move the -M<X> flags to the
 RW> end at the compile stage (to fool the c89 wrapper and still use fastdep
 RW> mode), or do the test with the flags added earlier on the command line.

 RW> Notes:
 RW> - For aesthetics, it is desirable to have the source file name appear at
 RW> the end of the command lines at compile time
 RW> (see the comment in automake/lib/am/depend2.am for a rationale).

Plus some tools (like libtool) assume that the last argument is
the source file.

[...]

 RW> - The gcc3 depmode, when issued with help of the `depcomp' script, will
 RW> be slowed down somewhat now.  This isn't really an issue for makefiles
 RW> generated by Automake, as they bypass the script at compile time with
 RW> the %FASTDEP% switch. 

Agreed.  Could you add a short comment in depcomp's gcc3 mode to
explain that while you are at it?

[...]

 RW> Attached are two alternative patches, both of which should fix the
 RW> issue.  OK to install the second, and drop the first?

Sure!  Thanks.

[...]

 RW> * lib/depcomp (gcc3): Put dependency extraction flags before the
 RW> `-c' flag, so they appear at the same position as in %FASTDEP%
 RW> mode in depend2.am.  Fixes build failure for FreeBSD's c89,
 RW> which ignores unknown options only after the first non-option.
 RW> Bug report against M4 by Nelson H. F. Beebe.

[...]

-- 
Alexandre Duret-Lutz

Shared books are happy books.     http://www.bookcrossing.com/friend/gadl