🚀 go-pugleaf

I've been installing various GNUish programs at a site where the
"installation" target directory is normally mounted read-only (to
enable replication and to avoid some types of accidental lossage.)
This target directory ("/import/gnu/") is copied from a writable
master directory hierarchy ("/import-writable/gnu/") on a regular
basis.

This causes problems with many of the "configure"-generated makefiles
because it is assumed that the target directory for installing
programs ("/import-writable/gnu/") is the same as the string which is
built into some programs as a default directory for finding
executables or data files.

For example, GCC 2.2.2 has the problem that "libdir" and "libsubdir"
are used both to specify where to install things (install-command) and
as part of a string built into some executables (eg cccp.o, gcc.o)
Note that it doesn't completely work to specify libdir as a relative
directory ("../../") because that means that a relative directory name
will get built into the exectutables.

Many other GNU installations break similarly in this situation.

It would be nice if people would make the effort to provide two
separate makefile variables for the two different cases of specifying
* where to install executables and data and where programs should
  write shared data (eg find database)
* where to expect to find installed executables and read-only data
  (eg gcc include and compiler phases.)