ash builtin function `read' called without arguments...

[Denys Vlasenko]

On Sat, Feb 9, 2019 at 5:10 PM Cristian Ionescu-Idbohrn
<cristian.ionescu-idbohrn at axis.com> wrote:
>
> On Sat, 9 Feb 2019, Denys Vlasenko wrote:
> > On Wed, Jan 30, 2019 at 2:09 PM Cristian Ionescu-Idbohrn <cristian.ionescu-idbohrn at axis.com> wrote:
> > > ...sets implicit variable REPLY, even though ASH_BASH_COMPAT is not
> > > set.
> > >
> > > That was introduced in 2010-01-12 with commit
> > > 045f4ad92c07434625e168bc8c37aa0e89f6e58e.  The comment is still
> > > present in the code:
> > >
> > > +               /* $IFS splitting. NOT done if we run "read"
> > > +                * without variable names (bash compat).
> > > +                * Thus, "read" and "read REPLY" are not the same.
> > > +                */
> > >
> > > Still, this is a bashism and should be tested with IF_ASH_BASH_COMPAT.
> > >
> > > Dash does not support that:
> > >
> > > $ read
> > > dash: 1: read: arg count
> > >
> > > and The Open Group Base Specifications Issue 6 does not document that
> > > behaviour.
> > >
> > > Yes, I suspect the "feature" may be already abused.
> > >
> > > What do you think?
> >
> > Correctly testing for ASH_BASH_COMPAT *and* HUSH_BASH_COMPAT
> > is tricky. I can think on of disabling "bashism" behavior
> > in case both options are off.
> > But is it worth the trouble?
>
> IMO, yes.  There might be other opinions.  RFC.  Let's wait.
>
> In my case (at work), I have to watch and prevent people from doing
> unportable things.  For me, that's a burden.

ash.c readcmd():
    params.argv = argptr;

hush.c builtin_read(c):
    params.argv = argv;

We can add a "params.argv[0] == NULL" test after these lines,
emit error message if it is and bash compat is off.