* Ralf Corsepius wrote on Fri, Sep 21, 2007 at 06:57:04PM CEST: > On Fri, 2007-09-21 at 16:10 +0200, Ralf Wildenhues wrote: > > > > * Ralf Corsepius wrote on Thu, Sep 20, 2007 at 09:46:12AM CEST: [...] > > > > | Makefile.am:1: `libdir' is not a legitimate directory for `DATA' [...] > > I agree that there should be a straightforward way to put files there. > > But it is right about complaining about the above line, > I disagree. *_DATA should take arbitrary data, no matter what it is and > let the user do with it whatever he wants.
But it's exactly the use case that the original pointer desired, that can break in unexpected ways: if you install a library with wrong permissions, it's not usable on some systems until the permissions are finally fixed. But between the time the library is copied and install-data-hook is executed, some program may need this new library, for example one of the programs that 'make' executes as part of the installation process. A similar issue has been observed before, IIRC with the gettext package: `libtool --mode=install' used `rm -f && ln -s' rather than first trying `ln -sf' for creation of library symlinks; but the `ln' binary happened to depend on the library that was being installed, so installation failed. > IMO, automake is outsmarting itself, here. It's preventing users from unobvious bugs. > > Ultimately, I guess another primary, or a prefix to _LIBRARIES or so, > > should be created to allow this sort of thing. > I don't think this would be wise. Do you have a reason for this? Cheers, Ralf
