source: kBuild/vendor/gnumake/2005-05-16/TODO.private

                                                            -*-Indented-Text-*-
GNU Make TODO List
------------------

This list comes both from the authors and from users of GNU make.

They are listed in no particular order!

Also, I don't gaurantee that all of them will be ultimately deemed "good
ideas" and implemented.  These are just the ones that, at first blush,
seem to have some merit (and that I can remember).

However, if you see something here you really, really want, speak up.
All other things being equal, I will tend to implement things that seem
to maximize user satisfaction.

If you want to implement some of them yourself, barring the ones I've
marked below, have at it!  Please contact me first to let me know you're
working on it, and give me some info about the design--and, critically,
information about any user-visible syntax change, etc.


The Top Item
------------

If you know perl (or want to learn DejaGNU or similar), the number one
priority on my list of things I don't have time to do right now is
fixing up the GNU make test suite.  Most importantly it needs to be made
"parallelizable", so more than one regression can run at the same time
(essentially, make the "work" directory local).  Also, the CWD during
the test should be in the work directory or, better, a test-specific
temporary directory so each test gets a new directory; right now
sometimes tests leak files into the main directory which causes
subsequent tests to fail (some tests may need to be tweaked).  Beyond
that, any cleanup done to make writing, reading, or handling tests
simpler would be great!  Please feel free to make whatever changes you
like to the current tests, given some high-level goals, and that you'll
port the current tests to whatever you do :).


The Rest of the List
--------------------

 1) Allow variables/functions to expand to other make rules which are
    then interpreted, with newlines handled correctly.  This is a
    biggee, and is on my plate.  I already have partially-working code.

 2) Option to check more than timestamps to determine if targets have
    changed.  This is also a very big one.  It's _close_ to my plate :),
    and I have very definite ideas about how I would like it done.
    Please pick something else unless you must have this feature.  If
    you try it, please work _extremely_ closely with me on it.

 2a) Possibly a special case of this is the .KEEP_STATE feature of Sun's
     make.  Some great folks at W U. in Canada did an implementation of
     this for a class project.  Their approach is reasonable and
     workable, but doesn't really fit into my ideas for #2.  Maybe
     that's OK.  I have paperwork for their work so if you want to do
     this one talk to me to get what they've already done.

     [K R Praveen <praveen@cair.res.in>]

 3) Currently you can use "%.foo %.bar : %.baz" to mean that one
    invocation of the rule builds both targets.  GNU make needs a way to
    do that for explicit rules, too.  I heard a rumor that some versions
    of make all you to say "a.foo + a.bar : a.baz" to do this (i.e., a
    "+" means one invocation builds both).  Don't know if this is the
    best syntax or not... what if you say "a.foo + a.bar a.bam : a.baz";
    what does that mean?

 4) Multi-token pattern rule matching (allow %1/%2.c : %1/obj/%2.o,
    etc., or something like that).  I have an implementation of this
    already, it just needs some refinement... maybe.  Additionally I
    think it only works for static pattern rules; it might need to be
    fixed up to work with normal pattern rules, too.

 5) More robust clock skew detection algorithm: less false hits.  I have
    some notes on this from various discussions.

 6) Provide MAKETARGETS and MAKEVARIABLES variables, containing the
    names of the targets and variables defined in the makefile.

    Actually, I now think a $(targets ...) function, at least, would be
    better than a MAKETARGETS variable.  The argument would be types of
    targets to list: "phony" is the most useful one.  I suppose
    "default" might also be useful.  Maybe some others; check the
    bitfields to see what might be handy.  This one is pretty easy.

 7) Some sort of operating-system independent way of handling paths
    would be outstanding, so makefiles can be written for UNIX, VMS,
    DOS, MS-Windows, Amiga, etc. with a minimum of specialization.

    Or, perhaps related/instead of, some sort of meta-quoting syntax so
    make can deal with filenames containing spaces, colons, etc.  I
    dunno, maybe something like $[...]?  This may well not be worth
    doing until #1 is done.

 9) Right now the .PRECIOUS, .INTERMEDIATE, and .SECONDARY
    pseudo-targets have different capabilities.  For example, .PRECIOUS
    can take a "%", the others can't.  Etc.  These should all work the
    same, insofar as that makes sense.

10) A syntax that specifies a build order _without_ implying a
    dependency relationship.  That is, a way to say "A must be built
    before B" that doesn't force B to be built when A changes.  This is
    very important for parallel builds: sometimes you need some stuff
    done first but you don't want to rebuild everything because of it.

11) Improved debugging/logging/etc. capabilities.  Part of this is done:
    I introduced a number of debugging enhancements.  Tim Magill is (I
    think) looking into options to control output more selectively.
    One thing I want to do in debugging is add a flag to allow debugging
    of variables as they're expanded (!).  This would be incredibly
    verbose, but could be invaluable when nothing else seems to work and
    you just can't figure it out.  The way variables are expanded now
    means this isn't 100% trivial, but it probably won't be hard.

12) Integration of Guile as an embedded scripting language.  This means:
    allowing Guile functions to be declared in makefiles somehow, then
    providing a syntax for invoking them.  At least one formulation of
    that would have the function resolve to a string which would be
    substituted in the makefile, kind of like $(shell ...) does now, but
    using the embedded interpreter so there's no process forked of
    course.  Obviously this is an optional add-on feature.

    It could be more advanced than that, even, who knows?  Maybe make
    could provide Guile functions that allow Guile scripts more direct
    access to internal make structures, somehow.  This kind of thing
    needs a lot of thought.

    Also there's always the flip side: in some very fundamental ways
    make isn't the best choice right now for a complex build tool.  It's
    great for simple-to-medium tasks, but there are already other tools
    available for the really tough situations.  Ask yourself,
    realistically, how much work is worthwhile to add to make, given the
    fundamentals you can't really overcome without significantly
    affecting backward compatibility--and then why not use another tool
    in the first place?

    Something to think about.