[9fans] Common code - isatty

[9fans] Common code - isatty

[Einar Karttunen]

Hello

While reading source code I noticed several different isatty
implementations. They are slightly different and I would be interested
in 1) which is the correct implementation and 2) whether a patch
to fix the others would be welcome and whether a common location 
in a library or cut-paste is preferred.

The locations include at least:
ape/lib/ap/isatty.c      - check whether fd2path ends in "/dev/cons"
cmd/ip/telnet.c          - dirstat -> "cons"
cmd/rc/plan9.c           - dirstat -> "ptty" or same dev,type,qid as "/dev/cons"
cmd/ssh/ssh.c            - fd2path ends in "/dev/cons"
cmd/upas/common/libsys.c - dirstat -> "cons"
libstdio/setvbuf.c       - dirstat -> same dev,type and qid as "/dev/cons"

- Einar Karttunen

Re: [9fans] Common code - isatty

[Russ Cox]

the correct approach is to check whether the end of
fd2path is /dev/cons.  the qid check is an older approach
that was used before fd2path.

i'm hesitant to put it in a library because most programs
(e.g., ls) should not behave differently when running on
a console.  if it's there, it may be overused like in unix.

but i'd gladly apply a patch fixing telnet, rc, upas,
and libstdio.

russ

Re: [9fans] Common code - isatty

[Ronald G. Minnich]

On Wed, 13 Apr 2005, Russ Cox wrote:

> i'm hesitant to put it in a library because most programs (e.g., ls)
> should not behave differently when running on a console.  if it's there,
> it may be overused like in unix.

good call. don't do it. We're fighting a lot of isatty() stupidity on our 
clusters, and it's a nightmare. I agree that it would get overused ...

ron

Re: [9fans] Common code - isatty

[Dan Cross]

"Ronald G. Minnich" <rminnich@lanl.gov> writes:
> On Wed, 13 Apr 2005, Russ Cox wrote:
> > i'm hesitant to put it in a library because most programs (e.g., ls)
> > should not behave differently when running on a console.  if it's there,
> > it may be overused like in unix.
> 
> good call. don't do it. We're fighting a lot of isatty() stupidity on our 
> clusters, and it's a nightmare. I agree that it would get overused ...

Or put it into libidiot.h just so there's only one copy of it.

Oh, you think I'm joking?

	- Dan C.

Re: [9fans] Common code - isatty

[boyd, rounin]

>> good call. don't do it. We're fighting a lot of isatty() stupidity on our 
>> clusters, and it's a nightmare. I agree that it would get overused ...

i agree, but for some unix things you need isatty().  i remember
asking this question some years ago during a Python port.

not sure i'd use fd2path() tho, but that's just me.
--
MGRS 31U DQ 52572 12604

Re: [9fans] Common code - isatty

[Russ Cox]

> i agree, but for some unix things you need isatty().  i remember
> asking this question some years ago during a Python port.

for unix things you should use ape.

> not sure i'd use fd2path() tho, but that's just me.

fd2path is the right answer.  
would you add a system call instead?

russ

Re: [9fans] Common code - isatty

[Bruce Ellis]

Dan!  Stop being so blunt and correct!

20 years ago it became clear that there was no bloody way
to determine if a program was being run interactively.

Given that ... the various hacks are convenient.
e.g. "rc" rather than "rc -i" but still ... It is a cesspool.

brucee

On 4/14/05, Dan Cross <cross@math.psu.edu> wrote:
> "Ronald G. Minnich" <rminnich@lanl.gov> writes:
> > On Wed, 13 Apr 2005, Russ Cox wrote:
> > > i'm hesitant to put it in a library because most programs (e.g., ls)
> > > should not behave differently when running on a console.  if it's there,
> > > it may be overused like in unix.
> >
> > good call. don't do it. We're fighting a lot of isatty() stupidity on our
> > clusters, and it's a nightmare. I agree that it would get overused ...
> 
> Or put it into libidiot.h just so there's only one copy of it.
> 
> Oh, you think I'm joking?
> 
>        - Dan C.
> 
>

Re: [9fans] Common code - isatty

[boyd, rounin]

> fd2path is the right answer. 

sure looks nice, 'cept for the strings.

> would you add a system call instead?

nope, call me old fashioned -- stat(2).
--
MGRS 31U DQ 52572 12604

Re: [9fans] Common code - isatty

[Russ Cox]

> sure looks nice, 'cept for the strings.

then plan 9 is not for you.

Re: [9fans] Common code - isatty

[Bruce Ellis]

sure, strcmp() is great for implementing isatty().
let's have a "best hack" bake-off.

brucee

On 4/14/05, Russ Cox <russcox@gmail.com> wrote:
> > sure looks nice, 'cept for the strings.
> 
> then plan 9 is not for you.

Re: [9fans] Common code - isatty

[Dan Cross]

Bruce Ellis <bruce.ellis@gmail.com> writes:
> sure, strcmp() is great for implementing isatty().
> let's have a "best hack" bake-off.

Ever use a spent 7.62mm round to short out a blown resister?

Where's Boyd.... 

	- Dan ``Reaches...'' C.

Re: [9fans] Common code - isatty

[Dan Cross]

Bruce Ellis <bruce.ellis@gmail.com> writes:
> Dan!  Stop being so blunt and correct!

Sorry, I just got out of the Marine Corps (hooray!  And there was
much rejoicing...).  It'll take me a few months before I can do
the former.  The latter is a new phenomenon.

	- Dan C.

Re: [9fans] Common code - isatty

[jmk]

make it consistent and rename it:

int ismaybesomethinglikeaterminal(int);

Re: [9fans] Common code - isatty

[boyd, rounin]

> Ever use a spent 7.62mm round to short out a blown resister?

no, i like the 9x19mm 124gr FMJ, but you can't argue with the 7.62 ...
--
MGRS 31U DQ 52572 12604

>

Re: [9fans] Common code - isatty

[Ronald G. Minnich]

On Thu, 14 Apr 2005, Bruce Ellis wrote:

> 20 years ago it became clear that there was no bloody way to determine
>    if a program was being run interactively.

yeah, it's why so many of our fortran runtimes guess wrong ...  they do
something stupid like "oh, this is !isatty(), I'll buffer stdout".  then
they core dump. Then users get no error message. Then users yell at me.

I really do hate isatty()

ron

Re: [9fans] Common code - isatty

[Rob Pike]

following jim's lead, you could make it
  whocaresifitsatty()
or maybe the best thing is to put into libc
  int isatty(int) {return 0;}

-rob

Re: [9fans] Common code - isatty

[Ronald G. Minnich]

On Thu, 14 Apr 2005, Rob Pike wrote:

>   int isatty(int) {return 0;}

excellent. 

You know, we did do this on linux via LD_PRELOAD hackery, and of course it
didn't work with glibc, since library functions in glibc don't bother
calling that function -- inlines, macros, all that great stuff you know --
it's really important to optimize isatty() calls, as they are so frequent.

ron

Re: [9fans] Common code - isatty

[Bruce Ellis]

i kinda like that approach.  a lot of POSIX garabage
coud be replaced with stubs.  then python written in perl
written in bash would never make it to plan9.

brucee

On 4/15/05, Rob Pike <robpike@gmail.com> wrote:
> following jim's lead, you could make it
>  whocaresifitsatty()
> or maybe the best thing is to put into libc
>  int isatty(int) {return 0;}
> 
> -rob