Re: [PATCH] fix:awk: CONVFMT segfaults with %n, wrong output with %p

On Tue, 2026-05-05 at 12:54 +0200, Roberto A. Foglietta wrote:
> On Tue, 5 May 2026 at 11:15, Hans Ulli Kroll via busybox
> <[email protected]> wrote:
> 
> > diff --git a/editors/awk.c b/editors/awk.c
> > index dd8f4ac420..b26793e792 100644
> > --- a/editors/awk.c
> > +++ b/editors/awk.c
> > @@ -969,7 +969,14 @@ static const char *fmt_num(const char *format, double n)
> >                 const char *s = format;
> >                 char c;
> > 
> > -               do { c = *s; } while (c && *++s);
> > +               do {
> > +                       c = *s;
> > +                       if (c == 'n' || c == 'p') {
> > +                               puts(format);
> > +                               exit(0);
> 
> 
> ./busybox awk 'BEGIN { CONVFMT="%.2f n" ; x=3.14; print x "" }'
> Your patch prints:
> %.2f n
> Instead of
> 3.14 n

so the whole CONVFMT is a mess.

There is *curently* no proper way in busybox to support floating point
specifiers

> 
> gawk compliance isn't possible to achieve within busybox constraints
> and also useless:
> 
> 1 - as long as awk syntax acceptance is valid also for gawk (awk subset of gawk)
> 2 - every unacceptable / rejected syntax provide exit 1 to warning users
> 3 - no silent failures or crashed are generated

point taken

> 
> Finally, in my V2 patch "p" isn't a numeric identifier but a pointer
> (indexing s = format content)
> 
> > +                       }
> > +               } while (c && *++s);
> > +
> >                 if (strchr("diouxX", c)) {
> >                         snprintf(g_buf, MAXVARFMT, format, (int)n);
> >                 } else if (strchr("eEfFgGaA", c)) {
> > --
> > 2.54.0
> 
> Best regards, R-

patch ?

Ulli