[9fans] varargs question

[9fans] varargs question

[erik quanstrom]

as i was drifting off to sleep the other night,
i was wondering about varargs.  (clearly, it's
time for a vacation.)

and i wondered why kenc doesn't add an argument
count before the first vararg.  (bwc pointed out
that the address following the last vararg would be
more useful.)  va_* could be updated to deal with
the silent extra and abort on access beyond the end
of the actual variable arguments.  a function to check
that list == end could added for the paranoid.

since all code is prototyped these days,
is there any reason why this is a bad idea that i
haven't thought of?

- erik

Re: [9fans] varargs question

[Russ Cox]

On Thu, Aug 20, 2009 at 6:56 AM, erik quanstrom<quanstro@quanstro.net> wrote:
> and i wondered why kenc doesn't add an argument
> count before the first vararg.  (bwc pointed out
> that the address following the last vararg would be
> more useful.)  va_* could be updated to deal with
> the silent extra and abort on access beyond the end
> of the actual variable arguments.  a function to check
> that list == end could added for the paranoid.

what problem are you trying to solve?

the limit would check only that the right number of
argument bytes are consumed, but not that they are
interpreted correctly.  print("%s %d", 1, "hello") would
still crash, as would print("%s", 1, "hello").

#pragma varargck is more precise and can be
done at compile time instead of needing to wait
until the code trips at run time.

russ

Re: [9fans] varargs question

[erik quanstrom]

> what problem are you trying to solve?
>
> the limit would check only that the right number of
> argument bytes are consumed, but not that they are
> interpreted correctly.  print("%s %d", 1, "hello") would
> still crash, as would print("%s", 1, "hello").
>
> #pragma varargck is more precise and can be
> done at compile time instead of needing to wait
> until the code trips at run time.

varargck is very effective.  but it only works if the pragma
is used.  (i count 47 functions with ... in /sys/include and
only 27 #pragma varargck argpos.)   and it only works with
functions taking print verbs.  anyone else, is out-of-luck.
for example, the stdio functions.

here are some functions that don't appear to have varargck
that likely should:

auth_wep
ctlerror
hfail
vtDebug
vtFatal
threadsetname
vtlogprint
vtlog

- erik