Re: [TCLCORE] TIP #115: Making Tcl Truly 64-Bit Ready

[George H. <ga...@si...>]

In message <3DB...@ma...>, "Donal K. Fellows" writes:

 : Given that the compiler already enforces sensible alignments, all that I'm
 : really having to do is look for opportunities to gather similarly
 : sized valuestogether so that if the compiler can pack sub-word
 : values more tightly, it has an adequate opportunity to do so.

Are you in favor of reordering or not?  I was struck by the fact that
your TIP isn't.  

 : > Also, if you break binary 32-bit compatibility between Tcl 8.x
 : and 9.x, then you risk that only few people will adopt it.
 : 
 : /me boggles!
 : 
 : The vast majority of people are going to accept 9.* sooner or later, largely
 : because of the wealth of cool new features and performance enhancements that
 : it will have, many of which I obviously have no knowledge about at this 
 : time.  That's precisely what's happened in the past, and it will happen
 : again. Diehards who prefer unsupported code might stay with 8.*, but only 
 : where they haven't decided that 7.* more suits their particular needs...

Now maybe I'm inferring something that isn't there.  I assume that 9.0
will contain major incompatibilities.  You suggest that 9.0 will be
reserved with revamping the APIs and other enhancements, but no
language incompatibilities.

Can we all agree that this will be the case?  If I can recompile all
my extensions (even if modifications are required) and my Tcl code
runs without changes, that's a perfectly fine plan depending upon the
time frame.  But if 9.0 makes incompatible changes to the Tcl (i.e. my
Tcl code doesn't work the same way under 9.0), I argue that the
"wealth of cool new features" and "performance enhancements" will be
dwarfed by the Tcl-code applications already out there.

The point is that 64-bit developers only need a version that runs with
64-bit capabilities.  That's something that can be done in the short
term.  We haven't even agree yet as to what the 9.0 features are and
what the time frame they will be available.  I don't like the idea of
linking of 9.0 features with true 64-bit platform support.  It's going
to make testing just that much harder to do.

 : > After all, who's going to port all those extensions and
 :  applications out there to 9.0?

 : I'd like porting to 9.0 to be no harder than a recompile, though
 : that will not be the case with some APIs (out and in/out parameters
 : are the key problem he re. By contrast, input parameters and return
 : values tend to be easier to deal with in binary-compatible
 : fashions.) However, one of the points of having a 9.0 is that it is
 : not binary compatible, and that we will not need to maintain
 : "cruft" compatability.

 : I'm very sorry, George, but I just can't see why you are kicking up all 
 : this fuss.  Without evidence of some compiler/platform combination
 : that has difficulties, I can't see structure alignment concerns as
 : anything other than a problem that has been conveniently solved by
 : other people.

I think you need to explicitly make the case for no language/command
changes for Tcl/Tk 9.0.  

--gah