issues

[Peter Stephenson <pws@xxxxxxxxxxxxxxxxx>]

These are mainly some open replies to Bart and Sven on things while I've
been away, and memos to myself.  Sorry to lump them together, but the
most likely alternative was not to do it at all.


64-bit stuff in 3.0.6: Well, I described the final result as experimental,
and that's certainly true when you're using zsh memory management because
of the new alignment constraints (on the architectures in question,
sizeof(long long) == sizeof(double) so the system's malloc is quite used to
using 8 byte alignment).  But as it has to be explicitly turned on there
would be no harm done to ingenu installers.  It's probably going to take
some work and there's a conference coming up, so it'll be some way down the
list, and if 3.0.6 is ready to go it's not worth waiting for possibly
suboptimal 64-bit integer support.  (I still have to check the compilation
log with --enable-lfs that Andrej sent me in more detail, there are one or
two things that look like they need tweaking, but nothing disastrous.)

$_ in preexec: this gets set later than you might think, after preexec is
executed, so that you get the previous command line, and doesn't track all
the syntax to find the last word.  I always found it of rather limited use
anyway; I imagine Paul stuck it in because it was in the ksh manual and
didn't worry too much how it worked --- it's probably one of those things
like $LINENO that works better in ksh.  Conceivably one could do more
sophisticated things as a module with special parameters.

quoting in completion:  if you do
  % echo "`<RET>
you get
  dquote bquote> 
from the cmdpush() stuff added by Sven to lex.c.  Could that be of any use
in completion?  Could we make a stack of contexts available and choose
quote behaviour based on that?  It would mean removing the delete-quote
code, so it doesn't sound too simple.  As for the test I modified for
detecting quotes (long ago, and I didn't make it very sophisticated), it
was simply because you didn't have any choice about how the shell behaved,
so it seemed important to have a good guess.

The cp -f thing in configure.in was just me being dozy.  However, I
deliberately didn't use mv because I want config.status to exist with the
right permissions, otherwise the edited version needs chmod +x; this was
just a short cut so that ... >config.status worked.  Using mv and chmod is
another possibility.

${..."$(...)"...}: I haven't looked yet, but I don't see why the change
shouldn't work in 3.0.6; the test will simply have to be in a different
place as the nested expansion (I was going to say subduced representation,
I'm that confused at the moment) mechanism is less sophisticated.  It's
simply a case of ignoring null tokens and relying on quoted dollars and I'm
pretty sure the basic algorithms for these go way back.  By the way, it's
possible there are other places where ignoring null tokens in this way
would be useful, but around a nested expansion is the most obvious.

the minor parameter fixes:  I didn't check if parameters with PM_UNSET were
allowed in the parameter table in 3.0, if they aren't then the problem goes
away.  I'll try and remember to fix the other thing to make it consistent.

NO_RCS/NO_GLOBAL_RCS: I didn't get replies apart from Bart's, i.e. no
protests from people who use it to turn off .zlogout, so I suppose it's OK
to change the behaviour so you can unset either at any time (the first
alternative) and mention it in the appropriate section of the FAQ.

the su thing: looks like I'd better alter this again.  I don't see any harm
in working around su problems if shell users are going to be stuck with
them --- and to claim it's a an su bug when zsh emulates sh just because it
saw the s of su seems to me to be a bit over the top.  Shame it's too
difficult to test for explicitly, but we could restrict it to, say, SunOS 4
(that's the only one I know about).  maybe I'll just go back to the first
patch that came.

history changes: I think there were two major patches of Wayne's; the only
things I did were turn on share_history for ksh emulation, which removes a
notable incompatibility, and shorten the name of incappendhistory.  It's
definitely a functional change, it doesn't really need to be in 3.0.6, I
would have said.

it would be nice to have real argument passing to zle functions, but I
would guess it's a lot more complicated than what Wayne added for numeric
arguments (though that's extremely useful), since it will depend on the
command being called how the argument is treated, and passing around
strings of characters as arguments means working out how the interface with
the standard character input mechanism should work.

Probably the stuff on pipelines and terminals is mostly or all applicable;
I don't remember this changing much in 3.1.

typeahead problem (pasing multiple lines when zle is already active,
again): just occurred to me that you can't even tell whether the typeahead
is for the shell anyway. consider 'cat<RET>echo<RET>^D' or
'echo<RET>echo<RET>' --- only when the first command has finished do you
know what input is for zsh.  sorry.


I've probably forgotten lots of things already, but maybe you'll remind
me.

-- 
Peter Stephenson <pws@xxxxxxxxxxxxxxxxx>       Tel: +39 050 844536
WWW:  http://www.ifh.de/~pws/
Dipartimento di Fisica, Via Buonarroti 2, 56127 Pisa, Italy