Re: [PATCH 3/4] Use consistent extensions for generated header files.

[Paolo Bonzini]

Joel E. Denny wrote:
> On Sun, 2 Nov 2008, Akim Demaille wrote:
> 
>> I'd be happy that no change be made to lalr1.cc before I have pushed the 150+
>> patches waiting in my branch.  We are waiting for 2.4 to be released, as far
>> as I am concerned, this is a feature-freeze period for bison.
> 
> My file naming and %language changes turned out to be much hairier than I 
> thought.  After seeing Paolo's comments and patches, it's clear to me 
> we're not going to reach consensus about what should be done with this any 
> time soon.
> 
> I'm going to mark %language as experimental for 2.4.  The Java skeletons 
> are already experimental anyway.  Then I'd like to release 2.4 
> immediately.  Today if possible.  Is everyone fine with that?

Before that, what about adding some tests for output file names (at
least those that are undeniably correct or undeniably wrong) and mark
the wrong ones as XFAIL?  This would give a ground for further discussion.

I copy here the changed tests between my patch and yours:

>  AT_CHECK_OUTPUT([[foo.bison]],, [[-LC++ -dv -g -x -y]],
> -                [[foo.cc foo.hh foo.output foo.dot foo.xml ]cpp_aux])
> +                [[y.tab.cc y.tab.hh y.output y.dot y.xml ]cpp_aux])
>  AT_CHECK_OUTPUT([[foo.bison]],, [[-LJava -v -g -x -y]],
> -                [[foo.java foo.output foo.dot foo.xml]])
> +                [[y.tab.java y.output y.dot y.xml]])
>  AT_CHECK_OUTPUT([[foo.bison]],, [[-LC -dv -g -x -obar-c]],
> -                [[bar-c foo.h foo.output foo.dot foo.xml]])
> +                [[bar-c bar-c.h bar-c.output bar-c.dot bar-c.xml]])
>  AT_CHECK_OUTPUT([[foo.bison]],, [[-LC -dv -g -x -obar-c -y]],
> -                [[bar-c y.tab.h y.output y.dot y.xml]])
> +                [[bar-c bar-c.h bar-c.output bar-c.dot bar-c.xml]])
> +AT_CHECK_OUTPUT([[foo.bison]],, [[-LC++ -dv -g -x -obar.C]],
> +                [[bar.C bar.H bar.output bar.dot bar.xml ]cpp_aux])
>  AT_CHECK_OUTPUT([[foo.bison]],, [[-LC++ -dv -g -x -obar.cpp]],
> -                [[bar.cpp foo.hh foo.output foo.dot foo.xml ]cpp_aux])
> +                [[bar.cpp bar.hpp bar.output bar.dot bar.xml ]cpp_aux])
>  AT_CHECK_OUTPUT([[foo.bison]],, [[-LJava -v -g -x -odir/bar.Java]],
> -                [[dir/bar.Java foo.output foo.dot foo.xml]])
> +                [[dir/bar.Java dir/bar.output dir/bar.dot dir/bar.xml]])

It seems to me that we are safe saying that for now -L/%language should
be used with -b/%file-prefix, and that we should not include:

1) tests that use -L without -b

2) tests that use -b and -o together

Everything else can be included in 2.4, possibly XFAILed.  XFAILs could
also help seeing *what* is actually experimental.  Unfortunately, here
is already evening and I won't have time to do that.

It's a bit sad that this realization came so late in the release cycle,
but so goes life.

Paolo