Re: [nft PATCH 10/28] datatype: Increase symbolic constant printer robustness

[Phil Sutter <phil@nwl.cc>]

On Thu, Oct 30, 2025 at 10:56:04PM +0100, Pablo Neira Ayuso wrote:
> On Thu, Oct 30, 2025 at 12:00:48PM +0100, Phil Sutter wrote:
> > On Wed, Oct 29, 2025 at 07:36:34PM +0100, Pablo Neira Ayuso wrote:
> > > On Thu, Oct 23, 2025 at 06:13:59PM +0200, Phil Sutter wrote:
> > > > Do not segfault if passed symbol table is NULL.
> > >
> > > Is this a fix, or a cleanup?
> >
> > It is a fix but for a case which normally doesn't happen. It is
> > triggered by the macro in patch 26 due to the ad-hoc struct output_ctx
> > definition not populating the symbol tables.
> 
> For the debug thing that is unused? ie. #define expr_print_debug(expr)

Yes, exactly. It basically does:

| {
| 	struct output_ctx octx = { .output_fp = stdout };
| 
| 	expr_print(expr, &octx);
| }

Depending on type of 'expr', this may then call
symbolic_constant_print(NULL, ...).

> This cannot ever happen in the code:
> 
> src/datatype.c: return symbolic_constant_print(octx->tbl.mark, expr, true, octx);
> src/meta.c:     return symbolic_constant_print(&pkttype_type_tbl, expr, false, octx);
> src/meta.c:     return symbolic_constant_print(octx->tbl.devgroup, expr, true, octx);
> src/meta.c:     return symbolic_constant_print(&day_type_tbl, expr, true, octx);
> src/proto.c:    return symbolic_constant_print(&ethertype_tbl, expr, false, octx);
> src/rt.c:       return symbolic_constant_print(octx->tbl.realm, expr, true, octx);
> 
> And here:
> 
> void datatype_print(const struct expr *expr, struct output_ctx *octx)
> {
> ...
>                 if (dtype->sym_tbl != NULL)
>                         return symbolic_constant_print(dtype->sym_tbl, expr,
>                                                        false, octx);
> 
> Sorry but this is all sufficiently complex to add misleading hints.
> 
> If you need to extend anything, then please have a look at extending
> --debug to expose more information that you need.
> 
> Sorry, I do not see this is an improvement.

OK, no problem. I can just drop patch 26 and this one from the series.

Cheers, Phil