[TCLCORE] novem-support branch [Was: Removing "thin wrappers" in Tk]

[Jan N. <nij...@us...>]

2013/2/27 Porter, Don <don...@ni...>:
> Any changes required
> to Tk for no other purpose than to make it compile
> against the "novem" branch in Tcl belong on a
> "novem-support" branch in Tk.  They can move to
> Tk trunk when "novem" moves to Tcl trunk and
> not before.

This "novem-support" branch already exists for Tk. At
the moment, when compiled against 8.6 headers, it
runs with novem (9.0a0). But in order to do that, it contains
some 'hacks', which better solutions should be
searched for, And it contains some experiments,
(like removal of Tcl_RegisterObjType), which still
needs to be TIP'ed.

The only source incompatbilities I encountered where
the use of VOID, STRINGIFY, JOIN and EXTERN
macros, and the use of Tcl_VarEval. Tcl_VarEval
is deprecated, so Tk should not use that. Therefore
I plan to eliminate all calls to Tcl_VarEval in Tk,
hopefully that doesn't raise any objections.

Other things Tk's "novem-support" branch contains
is replacing all "package require Tcl 8.6" with
"package require Tcl 8.6-". Of course, that can
only be backported as soon as Tcl 9.0 is released,
and Tk 8.6 is actually known to run with it.

So, how to further benefit from this? One possibility
(my suggestion) is: <http://core.tcl.tk/tcl/info/3b51901dfc>

With this in place, Tk's "novem-support" branch
could be simplified a lot. Explanation:
- Tcl_PkgPresent up to Tcl_UpVar are just thin wrappers
  around Tcl_PkgPresent2 up to Tcl_UpVar2. I see no
  caveats using macro's here at all.
- The Tcl_Eval, Tcl_GlobalEval, Tcl_SaveResult,
  Tcl_RestoreResult and Tcl_DiscardResult do the
  same as their function counterpart, except that
  interp->result is not saved/restored. For extensions
  which don't access interp->result that's OK, but
  who knows there still are... Therefore I would put
  an ifdef TCL_NO_DEPRECATED around it.
- The Tcl_AddErrorInfo/Tcl_AddObjErrorInfo macro's
  are somewhat less efficient as their functions
  counterparts. They are only used in the error case,
  so I would say "who cares!". But still, better be
  careful.

I think this should be TIP'ed. And maybe there are more
functions like this, which stub-table entries can be removed
in "novem" (9.0) without losing anything.

Regards,
      Jan Nijtmans