Re: build failure

[John W. Eaton]

On 06/24/2016 10:44 PM, Ben Abbott wrote:
> > On Jun 24, 2016, at 8:54 AM, Ben Abbott <address@hidden> wrote:
> >
> > > On Jun 24, 2016, at 08:39, c. <address@hidden> wrote:
> > >
> > > > On 24 Jun 2016, at 03:28, Ben Abbott <address@hidden> wrote:
> > > >
> > > > I tried jwe’s earlier suggestion to me, and it worked.
> > > >
> > > > $ make libinterp/corefcn/oct-tex-parser.cc
> > > > YACC     libinterp/corefcn/oct-tex-parser.cc
> > > > updating libinterp/corefcn/oct-tex-parser.output
> > > > updating libinterp/corefcn/oct-tex-parser.h
> > > >
> > > > Ben
> > >
> > > Thanks, this worked for me too!
> > >
> > > But if I delete the whole build directory then do
> > >
> > > configure && make
> > >
> > > I still get the problem so there might still be something
> > > to fix in the build process?
> > >
> > > Shall I submit a bug?
> > >
> > > c.
> >
> > Yes. There is a problem building from scratch. I assume this is a blocker?
> >
> > I'll browse the makefile when I have time and see if I can determine where the 
> > logic is failing.
> >
> > Ben
>
> Carlo, does the attached work for you?
>
> John, does the attached changeset look correct?

Does it work for you?  That's the same kind of dependency that is listed 
for oct-parse.{h,cc,yy}, but it still doesn't solve all the problems for 
me.  I'm also not sure why the rules for oct-parse and oct-tex-parser 
were different.

In any case, the .h file doesn't really depend on the .cc file, so 
that's not the correct dependency.  It does solve the problem of being 
able to do

  make oct-tex-parser.h

when oct-tex-parser.cc does not exist because the dependency chain will 
then cause the .cc file to be generated and, as a side effect, the .h 
file will also be created.  But this change doesn't solve the problem of 
creating the .h file when it is missing but the .cc file is present.

What we really want is a set of rules that tells make that %.h and %.cc 
are both generated by one invocation of bison so that the rule is 
triggered (but only once!) when either or both are missing.  The suffix 
rules generated by automake rules don't do that, and I don't think there 
is any way to make a suffix rule do that properly.  So I think this is 
really an automake bug, but one that is not likely to be fixed unless it 
becomes possible to generate modern GNU Make style Makefiles.

I checked in the following changeset in an attempt to work around the 
problem:

  http://hg.savannah.gnu.org/hgweb/octave/rev/b5e75ce299d8

jwe