Re: .PHONY and .SECONDARY interaction in make 3.81: bug/feature?

[Paul D. Smith]

%% Alexey Neyman <address@hidden> writes:

  an> revision 1.69 made .SECONDARY target not only set the 'all_secondary' 
  an> variable to 1, but also sets ->intermediate flag on all files. 
  an> Therefore, the check_dep() function in remake.c doesn't notice that 
  an> version.c needs to be rebuilt unconditionally.

  an> The following patch solves the problem. No regressions introduced on 
  an> i686-pc-linux-gnu (Fedora Core 2).

Hm.  This solves the problem for .PHONY files, but what if you use the
old-style FORCE target method to force rebuilds:

    FORCE: ;
    version.c: FORCE

This doesn't work with .SECONDARY set globally: it thinks FORCE doesn't
need to be remade (even though it doesn't exist), so version.c doesn't
need to be remade, etc.

Unfortunately it's not exactly clear from the documentation (which is
all I have to go on) what it means to set .SECONDARY globally.  If you
take the documentation at face value, the current (3.81) behavior is
correct and the bug you're reporting is actually an enhancement request,
that marking a target as .PHONY take precedence over .SECONDARY and
.INTERMEDIATE, whereas right now it's the other way around.  Which is
fine with me as a change, BTW.


However, I'm not sure that's really what the original intent of the
global .SECONDARY target was... or else I'm not sure all the
ramifications were completely thought out.  Note you cannot define a
global .INTERMEDIATE target!  Why, then, can you define .SECONDARY
globally, that imputes all the behavior of .INTERMEDIATE to all targets
in the system--except for automatic deletion?  It seems strange and
unusual to me.

I have to wonder if the original point of the global .SECONDARY was
really just to turn off all automatic deletions of any intermediate
files, without really forcing every target to be considered
intermediate.  To me that would make more sense.


On the other hand, if you read the information provided with bug # 2515,
which originally fostered this change, Bill Curry (the author of the
bug) clearly feels that the former ("face value" reading) of the docs is
correct and is what RMS intended when he introduced the feature,
apparently at the request of the project that Bill is working on.  So, I
guess maybe that is correct.  I personally would probably have
implemented this differently, because it seems that we have a matrix of
two different capabilities (delete/don't delete intermediates, and
build/don't build non-existent files) and the current set of targets to
set the capabilities doesn't cover the entire matrix: there are some
things you can't do.  I would have preferred to define the targets in a
more orthogonal manner to cover the entire matrix.  Then you could have
used them in combination to get the behavior you wanted more precisely.

-- 
-------------------------------------------------------------------------------
 Paul D. Smith <address@hidden>          Find some GNU make tips at:
 http://www.gnu.org                      http://make.paulandlesley.org
 "Please remain calm...I may be mad, but I am a professional." --Mad Scientist