GNU make -- params to user defined functions

[David Taylor]

In GNU make, it's unclear from the manual (but very clear from looking
at the code) whether the params to user defined functions are simply
eval'ed or recursively eval'ed.

They are, in fact, simply evaulated.  I believe that having them be
recursively evaluated is more useful.

I want to do iteration within user defined functions.  And the param
to be supplied as an argument has recursively evaluated variables that
will vary as the iteration variable changes.

[I actually currently have the iteration outside of any user defined
functions and it works just fine.  But, I'm making a change where
depending upon a selection early in the makefile, the ``higher level''
variables will be defined differently and the iteration will change
between being a single iteration versus being a double iteration.]

>From a perusal of the mailing list archives, I take it that there was
some concern that changing how the params to user defined functions
are evaluated might not be backwards compatible.

With a one character change to func_call, I can get my stripped down
example to work; without it, I have yet to find the right incantation.
So I'm, of course, in favor of such a change.

How's this for a proposal:

. a new phony target (just like .SECONDEXPANSION), called
  .RECURSIVEFUNCTIONPARAMS or something equally clueful.

  If the phony target is not seen, you get the old behavior.  Should
  therefore be totally backwards compatible.

. a check in read.c (just after the check for .SECONDEXPANSION)
  that sets an appropriate variable.

. a change to func_call to pass 0 or 1 to define_variable depending on
  whether the aforementioned variable is 0 or 1.

. an addition to the features list so that a makefile can check to see
  if the make being used supports the feature and can error out if not.

. a little verbiage for the make.texi file so that it's documented.

If the above stands a chance of being accepted, I might implement it.

What should the phony target name be?  What should the feature name
be?

NOTE: EMC does not currently have a copyright assignment on file for
GNU make; but we do have one on file for GCC, BINUTILS, and GDB.  I
would assume (hopefully not too naively) that adding MAKE to the list
would be straightforward for the lawyers.  If there is interest, I can
start the process.