Re: [9fans] u9fs and va_copy()

Re: [9fans] u9fs and va_copy()

[Aharon Robbins]

> I also have implementations of the
> POSIX %4$d stuff and %'d stuff, with plans for %'f.  Joy.

It's a trip, ain't it?  I did the %4$d stuff for gawk, which allows
i18n-ing an awk program.  T'was a pain.

The %'d stuff isn't terrible, and I'm lucky that I can punt to sprintf
for %'f and the other floating point formats.  (Quote, Let someone else
do the hard part, Unquote :-)

I have to wonder if the POSIX people got this stuff right.  The recent
versions of gawk follow the POSIX standard, which says that floating
point numbers *in the text of the awk program* always use `.' for the
radix point, but use the locale's radix point for *input and output data*.

Fine and good, but I have gotten complaints from people in non-C locales
wondering why their data is no longer interpreted by gawk the way it
used to be...

This whole area is one where you're damned if you do, and damned if you don't.

Whatever,

Arnold

Re: [9fans] u9fs and va_copy()

[Rob Pike]

Yes, you should do this. It's a royal pain, though.  The fmt library needs
a hell of a lot of work to be compliant. I've done it but I don't like the code
and plan to redo it sometime soon.  I also have implementations of the
POSIX %4$d stuff and %'d stuff, with plans for %'f.  Joy.

I'll ship it back when I'm done, but it will be a while; I have other fires at
the moment.

-rob

[9fans] u9fs and va_copy()

[jmk]

Eric Van Hensbergen <ericvh@gmail.com> has provided a fix for u9fs on the PowerPC.
The va_copy() macro is needed if the type of va_list is not an array (as it is on all
the current Plan 9 compilers I believe).

There may be other software ported between Plan 9 and other systems which could run
foul of the va_copy() macro. Should we add it to the Plan9 header files and check
all our code conforms to the correct use of variable arguments as specified in C99?

--jim