[Aldor-l] libaxiom.al: dependencies

[Stephen Watt]

Peter is also correct.  

In your scenario, "a" should be independent of the private info of "c",
but the .ao has to be around to compile it.

-- Stephen


On Sat, May 31, 2008 at 02:54:34PM +0200, Ralf Hemmecke wrote:
> On 05/31/2008 01:57 PM, Stephen Watt wrote:
> >It should be enough for "a" to mention just "b".  
> 
> Hmm, after Peter send me:
> 
> > Yes, but c.ao should be available (ie. in the library search path,
> > or in a library).
> > c will be needed if any types from there are implicitly used.
> 
> I modified the a.ao target of my Makefile to
> 
> a.ao: a.as libbbb.al
> 	rm c.ao
> 	cp libbbb.al liblibrary.al
>  	aldor -llibrary $<
> 
> then I get
> 
> >make
> aldor c.as
> ar r libccc.al c.ao
> ar: creating libccc.al
> cp libccc.al liblibrary.al
> aldor -llibrary b.as
> ar r libbbb.al b.ao
> ar: creating libbbb.al
> rm c.ao
> cp libbbb.al liblibrary.al
> aldor -llibrary a.as
> #1 (Fatal Error) Could not open file `c.ao'.
> make: *** [a.ao] Error 1
> 
> Is that OK?
> 
> >If "b" was created using a version of the compiler that circumvents
> >the "inline from" declaration (thereby allowing inlining from everything),
> >then if "c" changes, "a" may need to be recompiled.
> 
> Well, I don't know if any inlining happens here. I am using v1.1.0.
> 
> Ralf
> 
> >On Thu, May 29, 2008 at 01:08:10PM +0200, Ralf Hemmecke wrote:
> >>Dear Peter, and all who know about the Aldor compiler internals...
> >>
> >>I've checked now what dependencies you have calculated in src_aldor2.
> >>That seemingly differs by a few domain.
> >>
> >>Maybe it is important, but maybe not. Therefore my question. Do you know 
> >>enough about the aldor compiler internals to be able to say something 
> >>about the following?
> >>
> >>Suppose I have 3 files a.ap b.ap, c.ap where a depends directly on b but 
> >>not directly on c. And where b depends directly on c.
> >>
> >>Now I want to create the corresponding .ao file.
> >>
> >>Is it enough if while compiling a.ap that only b.ao is mentioned as a 
> >>library?
> >>
> >>The current compiler 1.1.0 compiles the stuff below without problem. If 
> >>it really is enough to list direct dependencies why should one bother to 
> >>figure out dependencies recursively?
> >>
> >>It might be different if I try to compile to .a files, but I am speaking 
> >>here only of the .ao files which are necessary for the libaxiom.al 
> >>construction.
> >>
> >>Ralf
> >>
> >>---BEGIN a.as
> >>#include "aldor"
> >>#library LLL "library"
> >>import from LLL;
> >>AAA: with {
> >>	afoo: % -> %
> >>} == add {
> >>	Rep == BBB;
> >>	import from Rep;
> >>	afoo(x: %): % == per bfoo rep x;
> >>}
> >>---END a.as
> >>
> >>
> >>---BEGIN b.as
> >>#include "aldor"
> >>#library LLL "library"
> >>import from LLL;
> >>BBB: with {
> >>	bfoo: % -> %
> >>} == add {
> >>	Rep == CCC;
> >>	import from Rep;
> >>	bfoo(x: %): % == per cfoo rep x;
> >>}
> >>---END b.as
> >>
> >>---BEGIN c.as
> >>#include "aldor"
> >>CCC: with {
> >>	cfoo: % -> %
> >>} == add {
> >>	cfoo(x: %): % == x;
> >>}
> >>---END c.as
> >>
> >>---BEGIN Makefile
> >>a.ao: a.as libbbb.al
> >>	cp libbbb.al liblibrary.al
> >>	aldor -llibrary $<
> >>
> >>libbbb.al: b.ao
> >>	ar r libbbb.al $<
> >>
> >>b.ao: b.as libccc.al
> >>	cp libccc.al liblibrary.al
> >>	aldor -llibrary b.as
> >>
> >>libccc.al: c.ao
> >>	ar r libccc.al $<
> >>
> >>c.ao: c.as
> >>	aldor c.as
> >>---END Makefile