building rc on QNX4

building rc on QNX4

[Sam Roberts]

rc looks really good to me, so I built it over the weekend, and
I'm starting to use it. I think I've seen a few problems, but I'll
play some more and see what I can turn up, in the meantime, here's
what I had to do to get it to build:

- it has setpgid(), but not setpgrp();
- it also has a utoa in <stdlib.h>, but not with the same args;
- and it's va_list isn't an lvalue.

So, are these patches accpetable? QNX's default bastardized ksh
is ok for interactive use, but not great for programming, and
I don't think I have to bash bash on the rc mailing list... 
(sorry, I couldn't resist ;-)

Also, there's a few things I've always figured should be easy,
like piping fds other than stdout->stdin, that are under rc.

Oh, and QNX also lacks a groff tex formatting system... do you
think you could distribute a pre-processed rc.1 as rc.man, so
it could be read by less? And how about postscript as well
(groff -man rc.1>rc.ps)?

I found a Linux box to format rc.1 on, but...

Anyhow, I'm sure I'll have more questions as I play.

So far so good, and thanks a bunch!
Sam


diff -ur rc-1.6b2-orig/config.h.in rc-1.6b2/config.h.in
--- rc-1.6b2-orig/config.h.in	Wed Oct 13 07:55:03 1999
+++ rc-1.6b2/config.h.in	Tue Apr 25 09:31:27 2000
@@ -107,6 +107,9 @@
 /* Define if you have the mkfifo function.  */
 #undef HAVE_MKFIFO
 
+/* Define if you have the setpgid function.  */
+#undef HAVE_SETPGID
+
 /* Define if you have the setpgrp function.  */
 #undef HAVE_SETPGRP
 
diff -ur rc-1.6b2-orig/configure.in rc-1.6b2/configure.in
--- rc-1.6b2-orig/configure.in	Fri Dec 10 06:25:53 1999
+++ rc-1.6b2/configure.in	Tue Apr 25 09:31:20 2000
@@ -38,7 +38,7 @@
 AC_TYPE_UID_T
 AC_CHECK_TYPE(ssize_t, long)
 
-AC_CHECK_FUNCS(getgroups setpgrp setrlimit strerror)
+AC_CHECK_FUNCS(getgroups setpgrp setpgid setrlimit strerror)
 RC_FUNC_GETGROUPS
 
 RC_FUNC_SIGSETJMP
diff -ur rc-1.6b2-orig/print.c rc-1.6b2/print.c
--- rc-1.6b2-orig/print.c	Thu Aug 19 05:22:03 1999
+++ rc-1.6b2/print.c	Tue Apr 25 09:36:17 2000
@@ -70,9 +70,9 @@
 	return FALSE;
 }
 
-static char *utoa(unsigned long u, char *t, unsigned int radix, const char *digit) {
+static char *utostr(unsigned long u, char *t, unsigned int radix, const char *digit) {
 	if (u >= radix) {
-		t = utoa(u / radix, t, radix, digit);
+		t = utostr(u / radix, t, radix, digit);
 		u %= radix;
 	}
 	*t++ = digit[u];
@@ -120,7 +120,7 @@
 		while (*altform != '\0')
 			prefix[pre++] = *altform++;
 
-	len = utoa(u, number, radix, table[upper]) - number;
+	len = utostr(u, number, radix, table[upper]) - number;
 	if ((flags & FMT_f2set) && (size_t) format->f2 > len)
 		zeroes = format->f2 - len;
 	else
diff -ur rc-1.6b2-orig/proto.h rc-1.6b2/proto.h
--- rc-1.6b2-orig/proto.h	Wed Oct 13 10:57:22 1999
+++ rc-1.6b2/proto.h	Tue Apr 25 09:37:34 2000
@@ -27,7 +27,11 @@
 objects of type va_list.  Of course, most places don't have this yet,
 but where it does exist we need to use it. */
 #ifndef __va_copy
+#ifndef __QNX__
 #define __va_copy(x,y) (x)=(y)
+#else
+#define __va_copy(x,y) x[0]=y[0]
+#endif /* __QNX__ */
 #endif
 
 #if STDC_HEADERS
@@ -84,9 +88,11 @@
 
 #else /* HAVE_SETPGRP */
 
+#ifndef HAVE_SETPGID
 /* Nothing doing. */
 #define setpgid()
 #define tcsetpgrp()
+#endif /* HAVE_SETPGID */
 
 #endif /*HAVE_SETPGRP */
 

-- 
Sam Roberts, sroberts at uniserve dot com, www.emyr.net/Sam

Re: building rc on QNX4

[Scott Schwartz]

| Oh, and QNX also lacks a groff tex formatting system... do you
| think you could distribute a pre-processed rc.1 as rc.man, so
| it could be read by less? And how about postscript as well
| (groff -man rc.1>rc.ps)?

Years ago, Henry Spencer wrote a man-page-only-roff in awk for cnews,
for just such a contingency.

Re: building rc on QNX4

[Carlo Strozzi]

On Wed, Apr 26, 2000 at 11:02:31AM -0400, Sam Roberts wrote:
 >
 >rc looks really good to me, so I built it over the weekend, and

I agree, it is a neat shell, much better than Bourne and derivatives
for scripting. I am awful at C programming, but I think adding a few
addons to rc should be easy. The only thing that I really miss is a
'read' builtin. Using line(1) did not work, at least on Linux, apart
from the fact that being line(1) an external process it is not really
efficient, as it is often called from inside a loop.

Also, I think that the name 'rc' is sometimes a too short one. On AIX,
for instance, it conflicts with the system startup script /etc/rc.
I use Linux, so that's not a problem for me, nevertheless I would
really like that the default 'make install' stuff set up a symlink
to rc with a longer and more meaningful name, like 'rcsh'.

cheers  -carlo
-- 
Attachments only in HTML/RTF/Postscript please !                    
Per favore solo attachments in formato HTML/RTF/Postscript !

Re: building rc on QNX4

[Sam Roberts]

Said Scott Schwartz <schwartz@bio.cse.psu.edu>:
> Said Sam <sroberts@uniserve.com>:
> | Oh, and QNX also lacks a groff tex formatting system... do you
> | think you could distribute a pre-processed rc.1 as rc.man, so
> | it could be read by less? And how about postscript as well
> | (groff -man rc.1>rc.ps)?
> 
> Years ago, Henry Spencer wrote a man-page-only-roff in awk for cnews,
> for just such a contingency.

That he's such a gifted awk hacker is pretty cool, but
I think it's easier to add a rule to the rc build to pack
the .man in with the distribution than to find that code!

Vrl, for example, has a vrl.man that comes with it.

--

Speaking of which, is this an appropriate place to talk about
vrl? I posted some diffs to the author, no response yet (only
been a day or so -- I'm just mentioning, not griping).

I had some trubles with it's termcap usage. It seems to
think that @7 is the keypad END key/capability. This wasn't
defined on my system, and while I found it defined in a bunch
of Linux termcap entries (mostly in what looked like old PC
consoles), it wasn't in the ansi entries.

Poking around it seemed the right thing was to use kh for HOME
and kH for END (vrl was using kh for HOME and @7 for END).

Any comments?

--

Also, I can't call 'login' from an rc shell, strangely, it
complains about not being session leader. I'm quite baffled.

I have to look into exactly what it takes to be a session
leader, calling setsid() at the beginning of main didn't
cut it, and the ps output for a ksh and rc don't seem to
show a difference in terms of how the sid, pgrp, and controlling
terminal are set up.

--

Also, I just noticed ^L doesn't clear the screen. I'll look
into that sometime.

--

Ciao!
Sam

Re: vrl (was: Re: building rc on QNX4)

[Gert-Jan Vons]

Sam Roberts wrote:

>Speaking of which, is this an appropriate place to talk about
>vrl? I posted some diffs to the author, no response yet (only
>been a day or so -- I'm just mentioning, not griping).

That would be me, since I'm the author :-) I assume you sent it to my 
(old) usa.net address, I'll check my inbox there.

>I had some trubles with it's termcap usage. It seems to
>think that @7 is the keypad END key/capability. This wasn't
>defined on my system, and while I found it defined in a bunch
>of Linux termcap entries (mostly in what looked like old PC
>consoles), it wasn't in the ansi entries.
>
>Poking around it seemed the right thing was to use kh for HOME
>and kH for END (vrl was using kh for HOME and @7 for END).

kH is the "last-line" key according to FreeBSD's termcap(5). What is the 
OS you are using, what is the tty type, and on what hardware? (please 
mail me the termcap entry for your tty)

"last-line" for going to the end of a line seems a bit strange, but if 
that is what your tty sends when you hit the End key... I could fall back 
to kH if there is no @7 defined.

>--
>Also, I can't call 'login' from an rc shell, strangely, it
>complains about not being session leader. I'm quite baffled.

What kind of OS are you using? I can use "login" under solaris and 
freebsd without any problems.

>Also, I just noticed ^L doesn't clear the screen. I'll look
>into that sometime.

Nope. When you use vrl, ^L redisplays the line, useful in case it got 
garbled by the output of some background process. Note also that vrl is 
about lines, it doesn't (want/need to) know about the screen. The unix 
'clear' command handles that very well.



         Gert-Jan
-----
"If you are good, you will be assigned all the work. If you
  are really good, you will get out of it."
     - Dilbert

Re: vrl (was: Re: building rc on QNX4)

[Sam Roberts]

> 
> >Speaking of which, is this an appropriate place to talk about
> >vrl? I posted some diffs to the author, no response yet (only
> >been a day or so -- I'm just mentioning, not griping).
> 
> That would be me, since I'm the author :-) I assume you sent it to my 
> (old) usa.net address, I'll check my inbox there.

Yep, that's what's in 1.3's README! ;-)
 
> >I had some trubles with it's termcap usage. It seems to
> >think that @7 is the keypad END key/capability. This wasn't
> >defined on my system, and while I found it defined in a bunch
> >of Linux termcap entries (mostly in what looked like old PC
> >consoles), it wasn't in the ansi entries.
> >
> >Poking around it seemed the right thing was to use kh for HOME
> >and kH for END (vrl was using kh for HOME and @7 for END).
> 
> kH is the "last-line" key according to FreeBSD's termcap(5). What is the 
> OS you are using, what is the tty type, and on what hardware? (please 
> mail me the termcap entry for your tty)

QNX4, using their pterm (the Photon Terminal, Photon being the GUI
system), with ANSI emulation. The esc sequence for END (by observation)
is \E[Y, but is not present in their /etc/termcap (I guess most programs
are useing terminfo now). Below is the termcap:

qansi|qansi-m|qansi-8859m|QNX ANSI:am:G0:co#80:it#8:li#25:ti=\E[?7h:\
    :AL=\E[%dL:DC=\E[%dP:DL=\E[%dM:DO=\E[%dB:IC=\E[%d@:LE=\E[%dD:\
    :RA=\E[?7l:RI=\E[%dC:SA=\E[?7h:SF=\E[%dS:SR=\E[%dT:UP=\E[%dA:\
    :ae=^O:al=\E[L:as=^N:bt=\E[Z:cb=\E[K\E[X:cd=\E[J:ce=\E[K:ch=\E[%i%dG:\
    :cl=\E[2J\E[H:cm=\E[%i%d;%dH:cs=\E[%i%d;%dr:ct=\E[2g:dc=\E[P:dl=\E[M:do=\E[B:\
    :ec=\E[%dX:ho=\E[H:ic=\E[@:is=\E[?7h\E[0;10;39;49m:\
    :k1=\EOP:k2=\EOQ:k3=\EOR:k4=\EOS:k5=\EOT:k6=\EOU:k7=\EOV:k8=\EOW:k9=\EOX:\
    :k;=\EOY:kB=\E[Z:kC=\ENa:kD=\E[P:kF=\E[a:kI=\E[@:kN=\E[U:kP=\E[V:kR=\E[b:kT=\ENb:\
    :ka=\ENd:kb=\177:kd=\E[B:kh=\E[H:kl=\E[D:kr=\E[C:ku=\E[A:\
    :le=\E[D:ll=\E[99H:mb=\E[5m:md=\E[1m:me=\E[m^O:mh=\E[2m:mk=\E[9m:mr=\E[7m:\
    :nd=\E[C:nw=\EE:op=\E[39;49m:rp=%.\E[%p2%{1}%-%db:\
    :se=\E[27m:sf=\E[S:so=\E[7m:sr=\E[T:st=\EH:ue=\E[24m:up=\E[A:us=\E[4m:\
    :ac=``aaffggjjkkllmmnnooppqqrrssttuuvvwwxxyyzz{{||}}~~:\
    :vb=\E[?5h\E[?5l:ve=\E[?25h\E[?12l:vi=\E[?25l:\
    :ws=\E[5m:we=\E[m:bo=\E[1m:be=\E[m:

So, I can just add ":@7=\E[Y:", but I looked at /usr/lib/terminfo/terminfo.src,
and checked out the other emulation pterm can do, the native QNX terminal
type:

term   terminfo              termcap
----   --------              -------
qnx    kend=\377\250         kH=\0377\0250
ansi   kend=\E[Y             Nothing at all...

Thus I made the obvious assumption that vrl should use 'kH', and I
should add 'kH' to the termcap entry for ansi. Then rc/vrl would
work fine with both emulations.

It doesn't surprise me that this is wrong, QNX is relatively new,
and some of their support for legacy Unix stuff is shaky.

So, what's the "right" way? A note in the README for QNX4 users saying
to add :@7\E[Y: to "qansi" and :@7=\0377\0250: to "qnx" in their
termcaps so that all works well?

> "last-line" for going to the end of a line seems a bit strange, but if 
> that is what your tty sends when you hit the End key... I could fall back 
> to kH if there is no @7 defined.
> 

> >Also, I just noticed ^L doesn't clear the screen. I'll look
> >into that sometime.
> 
> Nope. When you use vrl, ^L redisplays the line, useful in case it got 
> garbled by the output of some background process. Note also that vrl is 
> about lines, it doesn't (want/need to) know about the screen. The unix 
> 'clear' command handles that very well.

And fn clear { /bin/echo \f } is working well for me, thanks for the clarification.

Sam

Re: rc not session leader?

[Sam Roberts]

> >Also, I can't call 'login' from an rc shell, strangely, it
> >complains about not being session leader. I'm quite baffled.
> 
> What kind of OS are you using? I can use "login" under solaris and 
> freebsd without any problems.

Russ Cox also tells me that login works fine under his system. I
kind of expected that rc was working fine everywhere else.

.............

And in the process of gathering the ps outputs to post here, I
just figured it out. Duh.

login has to be *execed*, not forked/and execed. So login is
a builtin shell alias for "exec login" in QNX's korn shell, so
login gets execed by the shell, the session leader and is happy.
I was trying to run login, so login was being execed by a forked
child of rc, NOT the session leader, and it was failing.

fn login { exec builtin login $* } seems appropriate here...

Thanks!
Sam

Re: vrl (was: Re: building rc on QNX4)

[Gert-Jan Vons]

Sam Roberts wrote:

>Yep, that's what's in 1.3's README! ;-)

FYI, the latest version of vrl (1.3.2, which includes 2 bug fixes) can be 
found at http://vons.free.fr/tools/index.html

>QNX4, using their pterm (the Photon Terminal, Photon being the GUI
>system), with ANSI emulation. The esc sequence for END (by observation)
>is \E[Y, but is not present in their /etc/termcap (I guess most programs
>are using terminfo now). Below is the termcap:
>....
>So, what's the "right" way? A note in the README for QNX4 users saying
>to add :@7=\E[Y: to "qansi" and :@7=\0377\0250: to "qnx" in their
>termcaps so that all works well?

That solution fixes the problem not only for "vrl" but for other 
termcap-aware programs as well.

I registered for a free copy of QNX, so I should be able to have a look 
at this myself somewhere during the summer.



         Gert-Jan
-----
How come wrong numbers are never busy?

Re: building rc on QNX4

[Tim Goodwin]

> Also, I think that the name 'rc' is sometimes a too short one. On AIX,
> for instance, it conflicts with the system startup script /etc/rc.

Why is this any more a "conflict" than that between /usr/bin/passwd and
/etc/passwd?  OK, so /etc/rc is executable, which /etc/passwd isn't, but
surely you don't ever want to run /etc/rc by hand?  And even if you do,
who puts /etc in their path?  (I don't use AIX, by the way.)

> I use Linux, so that's not a problem for me, nevertheless I would
> really like that the default 'make install' stuff set up a symlink
> to rc with a longer and more meaningful name, like 'rcsh'.

Sounds a bit close to `tcsh' to me :-).  But if that's what you want,
just say

    sh configure '--program-suffix=sh'

You can also use `--program-prefix' to tack things onto the beginning of
the name, or `--program-transform-name' to do arbitrarily weird things
to it.

Tim.

Re: building rc on QNX4

[Carlo Strozzi]

On Tue, May 02, 2000 at 10:41:13AM -0400, Tim Goodwin wrote:
 >> Also, I think that the name 'rc' is sometimes a too short one. On AIX,
 >> for instance, it conflicts with the system startup script /etc/rc.
 (...)
 >Sounds a bit close to `tcsh' to me :-).  But if that's what you want,
 >just say
 >
 >    sh configure '--program-suffix=sh'

Hmm ... that's not the point. What I mean is that the new additional
name should be set by the mainstream distribution of Rc, just to acquire
such name. This is especially true for Debian GNU/Linux, where whoever
acquires a name first "wins". Anyway, this is not at all a big issue,
and I am not going to insist on it :-).  Mine was just a general
concern about using too short names for programs less ubiquitous
than /bin/sh.

A more important issue IMHO is whether Rc should provide a built-in
read function, similar to the one offered by most Bourne shells; another
one is whether it will ever make it possible not to export everything to
the environment by default.

Any thoughts about this ?

bye  -carlo
-- 
I can read MIME or uuencoded e-mail attachments in PDF, Postscript, HTML,
RTF or text formats. Please do not send Word or Excel files. 

Re: building rc on QNX4

[Tim Goodwin]

> A more important issue IMHO is whether Rc should provide a built-in
> read function, similar to the one offered by most Bourne shells;

There is no *need* to make `read' a builtin: see the EXAMPLES file in
the distribution for an alternative.  One of rc's design goals is to
avoid unnecessary builtins.  So, no, I don't think this is likely to
happen.

(I accept that there is some work to be done, first in terms of
documentation: the EXAMPLES file is not as visible as it should be.
Also, I have some vague ideas for the future about a package of small
utilities to work with rc (and other shells); one of these would be
something like `line'---unfortunately not available everywhere---and
make a building block for an efficient `read' function.)

>                                                                  another
> one is whether it will ever make it possible not to export everything to
> the environment by default.

I brought this up a few months ago, and the consensus seemed to be that
the current scheme works well enough in practice.  Have you encountered
a problem?

Tim.

Re: building rc on QNX4

[Chris Siebenmann]

| There is no *need* to make `read' a builtin: see the EXAMPLES file in
| the distribution for an alternative.

 Unfortunately the example doesn't work in several very useful
situations: it will badly flub the common scripting idiom of
'generate stuff | {while (read foo) {....}}', never processing
most of the actual output.

 In order to make this work, you need to find a utility that is
guaranteed to read no more than the first line of standard input.
Most utilities (eg awk) will happily eat an entire large buffer
worth of standard input and then give you the first line back.
This is unfortunate if you wanted to read the rest of the buffer's
worth of input later.

 read is the one extra command that I really think should be a
builtin. I believe there are good versions that are simple to
add (through the existing hooks) in the mailing list archives[*].

	- cks
[*: ftp://ftp.sys.utoronto.ca/pub/rc/ for new people.]

Re: building rc on QNX4

[Tim Goodwin]

>  In order to make this work, you need to find a utility that is
> guaranteed to read no more than the first line of standard input.

Yup.  What if such a utility were distributed with rc, or as part of an
rc-utils package?

>  read is the one extra command that I really think should be a
> builtin.

OK.  Any other views?

Tim.

Re: building rc on QNX4

[David Luyer]

Tim Goodwin wrote:
> >                                                                  another
> > one is whether it will ever make it possible not to export everything to
> > the environment by default.
> 
> I brought this up a few months ago, and the consensus seemed to be that
> the current scheme works well enough in practice.  Have you encountered
> a problem?

I encountered a problem with this years ago but just started using
"su -" when I was trying to get programs to run that it broke.

Basically some OS's (Digital Unix for one I think) have C libraries which
choke horribly if the size of the command line + environment array is greater 
than a certain value, and if you have a number of complex functions in your 
shell this value (64k?) can be a pain.

Although it is really useful to have the shell functions inherited by a shell 
which was spawned from some unrelated program running under the initial shell,
maybe we need some way to prevent this in specific cases?

Maybe some syntax along the lines of:

(+ HOME PATH path USER LOGNAME "fn "*) /usr/sbin/cron &

to pass HOME, PATH, path, USER, LOGNAME and fn_* only

(- "fn "*) /usr/sbin/cron &

to pass everything but fn_*

But then - as with most things this could be done in a function

eg: run sh with HOME USER LOGNAME
    run sh without "fn "*

which could be implemented relatively easily if there was an option to whatis 
to just list variables instead of show what they are.

(eg: "whatis -fq" - list functions, "whatis -vq" - list variables)

Anyway - basically - I think it might be more useful to extend whatis such
that it can be used to implement easily a facility to run a command without 
any environment variables, rather than to actually implement that facility 
internally.

David.
-- 
----------------------------------------------
David Luyer
Senior Network Engineer
Pacific Internet (Aust) Pty Ltd
Phone:  +61 3 9674 7525
Fax:    +61 3 9699 8693
Mobile: +61 4 1064 2258, +61 4 1114 2258
http://www.pacific.net.au        NASDAQ: PCNTF
<< fast 'n easy >>
----------------------------------------------

Re: building rc on QNX4

[Carlo Strozzi]

[-- Attachment #1: Type: text/plain, Size: 2312 bytes --]

On Mon, May 08, 2000 at 04:29:54AM -0400, Tim Goodwin wrote:
 >> A more important issue IMHO is whether Rc should provide a built-in
 >> read function, similar to the one offered by most Bourne shells;
 >
 >There is no *need* to make `read' a builtin: see the EXAMPLES file in
 >the distribution for an alternative.  One of rc's design goals is to
 >avoid unnecessary builtins.  So, no, I don't think this is likely to
 >happen.
 >

Unfortunately the example provided does not work the way 'read' does,
and is an extra-process that needs to be spawned for each input line.
                                                                     
 >> one is whether it will ever make it possible not to export everything to
 >> the environment by default.
 >
 >I brought this up a few months ago, and the consensus seemed to be that
 >the current scheme works well enough in practice.  Have you encountered
 >a problem?

hmm ... not really. Mine is just a general concern. Currently I
initialize every variable that I'm going to use in a script, right
at the beginning of the program, somewhat like what you do with C
whan you declare things first.  That is because unless I maintain a
global name-space I never know what a variable is going to contain 
when the program is called. Having a clean environment would
be desirable in general.

Anyway, the above is another minor issue, it may as well be that the
current scheme is ok. What I _really_ miss is a 'read', or 'line',  
function, for record-oriented input, and I think this feeling is
shared also by others. While things like 'test' are rarely called   
inside a loop, reading lines from a file is the opposite, and spawning
an external process on each line sucks, especially when one uses
the shell for writing Web CGI scripts, like I do.

Beside all that, I am a long-time (Bourne) shell user, and I whish I
had uncovered rc before :-)

Anyway, just to give my two-cent and not just ask for things :-), here
are a couple of functions to emulate often-used external processes like
basename(1) and dirname(1). I've tried to mimic the respective utilities
as closely as I could. Bug-fixes are welcome :-)

bye  -carlo
-- 
I can read MIME or uuencoded e-mail attachments in PDF, Postscript, HTML,
RTF or text formats. Please do not send Word or Excel files. 

[-- Attachment #2: rcstuff.txt --]
[-- Type: text/plain, Size: 612 bytes --]

#
# Emulate dirname(1)
#

fn DirName {
   ~ $1 () && { echo DirName: too few arguments >[1=2] ; return 1 }
   ~ $1 '' .* && { echo . ; return }
   ~ $1 / && { echo / ; return }
   x=() y=() {
      x = ``(/) {echo -n $1}
      ~ $1 /* && ~ $x(2) () && { echo / ; return }
      ~ $x(2) () && { echo . ; return } 
      * = $x
      while (!~ $#* 0 1) { y = $y/$1 ; shift }
      echo $y
   }
}

#
# Emulate basename(1)
#

fn BaseName {
   ~ $1 () && { echo BaseName: too few arguments >[1=2] ; return 1 }
   ~ $1 '' / && return
   x = () {
      x = ``(/) {echo -n $1}
      echo $x($#x)
   }
}

# End of stuff.

Re: building rc on QNX4

[Stephen Tell]

On Mon, 8 May 2000, Tim Goodwin wrote:

> Date: Mon, 8 May 2000 05:15:01 -0400
> From: Tim Goodwin <tjg@star.le.ac.uk>
> To: rc@hawkwind.utcs.toronto.edu
> Subject: Re: building rc on QNX4
> 
> >  In order to make this work, you need to find a utility that is
> > guaranteed to read no more than the first line of standard input.
> 
> Yup.  What if such a utility were distributed with rc, or as part of an
> rc-utils package?

I suppose that would be sufficient.  The workaround I've used
has been this:

 var = `{sh -c 'read t; echo $t'}  # read var

Starting up sh just to avoid the little bit of code that a read
builtin would add to rc itself or an included utility seems a bit
rediculous.


-- 
Steve Tell | tell@cs.unc.edu | http://www.cs.unc.edu/~tell | KF4ZPF
On Leave from UNC working at Chip2Chip, Inc.  tell@chip2chip.com/919-929-0991

builtins

[Scott Schwartz]

My $0.02:  if you have echo as a builtin, read deserves the same.

My other $0.02:  echo needs a sibling builtin, quote, to enquote it's
arguments so that they can be read by the shell.  Lisp got that right ages
ago.  Inferno got it right more recently.

Followups to the es list, I guess. :-)

Re: builtins

[Carlo Strozzi]

On Tue, May 09, 2000 at 08:37:09PM -0400, Scott Schwartz wrote:
| My $0.02:  if you have echo as a builtin, read deserves the same.

I definitely agree. This is the single most important point IMHO.
The prolem isn't just reading one single line fron stdin, as many unix
utilities would allow that easily, but rather to make possible a construct
like:
          { while ( read a ) echo $a } <file

where 'file' is big enough to blow up the environment if swallowed
all in one go.

Just as a hint, bourne shells solve the problem by silently forking a
subshell to handle the above -- and using a builtin read, of course.

| 
| My other $0.02:  echo needs a sibling builtin, quote, to enquote it's
| arguments so that they can be read by the shell.  Lisp got that right ages
| ago.  Inferno got it right more recently.

I'm not quite sure I am understanding this ...

| 
| Followups to the es list, I guess. :-)

Well, I think the 'read' stuff pertains also to Rc. While we are at it,
I am one of those who rised also the 'export or not to export' issue.
Having a shell that exports everything by default does pose some
problems.  I use the shell to build the "outer frame" of relatively
complex Web sites (search engines, on-line auctions, ...), and I came
to pick the shell for that job after having tried every sort of other
languages, from perl to tcl, to C, to ... you name it. As chance
would have it that the Apache Web server passes an almost completely
empty environment to CGI programs, to make it harder to trick them
into doing wrong things by fiddling with env stuff. Having a shell
that allows not to have everything global by default would help a lot
with building large applications, where the possibility of de-coupling
things may be a critical factor.

Maybe an all-or-none approach would solve both the need of not exporting
and the need of not breaking backward compatibility, and it would also be
be more in-line with the clean design of Rc, which is what I like most.
Beside having a @{...} construct to run things in a subshell, we could have
also a %{...} thing (or whatever), that would pass a completely empty
environment to subprocessess. Something like 'env -', but without spawning
an extra- process. In this way it would be applicable also when calling
functions.  Does that make any sense to you ?

Apart from that, I do not think Rc needs any other stuff built-in. It is
one of the best examples of perfectly usable minimalism I've seen so far :-)

        --Carlo
-- 
I can read MIME or uuencoded e-mail attachments in PDF, Postscript, HTML,
RTF or text formats. Please do not send Word or Excel files.