Re: [Automake-NG] [PATCH 2/2] [ng] all, check, install: move more processing at make runtime

[Stefano Lattarini]

On 06/15/2012 11:46 AM, Akim Demaille wrote:
> 
> Le 15 juin 2012 à 11:34, Stefano Lattarini a écrit :
> 
>>> Sometimes you using ifdef, and sometimes ifeq "".  Are they equivalent?
>>>
>> AFAIU, these are equivalent:
>>
>>   ifdef VAR
>>
>>   ifneq "$(VAR)" ""
>>
>> But of course, there no way to express something like this with ifdef:
>>
>>  ifneq "$(VAR-1)$(VAR-2)" ""
> 
> Yes, of course.  Maybe note also this warnings from make.info:
> 
>      Often you want to test if a variable has a non-empty value.  When
>      the value results from complex expansions of variables and
>      functions, expansions you would consider empty may actually
>      contain whitespace characters and thus are not seen as empty.
>      However, you can use the `strip' function (*note Text Functions::)
>      to avoid interpreting whitespace as a non-empty value.  For
>      example:
> 
>           ifeq ($(strip $(foo)),)
>           TEXT-IF-EMPTY
>           endif
> 
>      will evaluate TEXT-IF-EMPTY even if the expansion of `$(foo)'
>      contains whitespace characters.
> 
>>> There's no way to avoid forking another Make here?
>>>
>> So far I've found no natural way (note that such a fork is there also in
>> mainline Automake, so this is not a performance regression at least).
> 
> I know.
> 
>>  The
>> problem is that all $(BUILT_SOURCES) must be created before we run any of
>> all-recursive, all-am, or *any of their prerequisites*.  I don't know
>> whether there is any natural way to accomplish that with GNU make.  Any
>> suggestion?
> 
> I don't know.  Maybe using order only dependencies?
> 
> %LOCAL-HEADERS% $(BUILT_SOURCES) :| $(if $(SUBDIRS),all-recursive,all-am)
>
This is backwards: all-am should be run *after* the $(BUILT_SOURCES) have
been built, not before.

> all: %LOCAL-HEADERS% $(BUILT_SOURCES) $(if $(SUBDIRS),all-recursive,all-am)
>
I already tried to use order-only prereqs to tackle this problem.  The failure
was spectacular :-(

The main issue is that order-only dependencies doesn't have actually anything
to do with ordering:

  <http://lists.gnu.org/archive/html/bug-make/2012-06/msg00003.html>

But now that I understand what order-only prereqs are truly about, I think they
can still be used to shave the extra $(MAKE) invocation if it's not truly
necessary: have 'Makefile' depend on $(BUILT_SOURCES) as an order-only prereq,
so that any missing $(BUILT_SOURCES) is built before any other target is
considered (because GNU make is still deciding whether it ought to rebuild
the Makefile).

I will give that a try later today, or tomorrow.

>>>> am__mkdir = test -d $1 || $(MKDIR_P) $1
>>>
>>> Is the "test" really needed?  "mkdir -p /already/exists" should silently
>>> exit 0, shouldn't it?
>>>
>> Yes, but then it would cause an extra fork even if the target directory
>> already exists.  I reckon that would quite displease our Cygwin users ;-)
> 
> Ah, yes, test is expected to be a built-in.  Thanks.
> 
>> Thanks.  I understand this is an ACK then?
> 
> 
> Sure!
>
Thanks, will push shortly.

Stefano