On Mon, Sep 10, 2018 at 02:56:48PM +0200, Phil Sutter wrote: > Hi, > > On Wed, Aug 29, 2018 at 02:19:22AM +0200, Pablo Neira Ayuso wrote: > > On Tue, Aug 28, 2018 at 10:26:56PM +0200, Phil Sutter wrote: > > > Unless there are objections, I will later follow-up with a patch to > > > convert remaining human readable values into numeric ones in JSON > > > regardless of numeric output or not. This is just a quick fix to unbreak > > > the build. > > > > Please, go ahead. > > After making the changes and looking at the results, I'm not so sure > anymore whether it really is a good idea: It means no symbol table > lookups are done anymore, possibly exposing internal or at least not > well-known values. A few examples are: > > * NFT_REJECT_ICMPX_*, which are unrelated to ICMP/ICMPv6 codes. > * RTN_*, declared without explicit value in rtnetlink.h. > * NFPROTO_*, similar but unrelated to ETHERTYPE_*. > * ct states, calculated from IP_CT_* > > What are your thoughts, should I still perform the changes or maybe > rather just fix inet_service_type_json() to make it consistent with > inet_service_type_print() again? I think built-in definition should still remain strings, while anything else that is user-defined, including inet_service_type_print() which depends on /etc/services, should default on numeric values. Question here is if you consider it useful to get json in sync with human-readable output and the new -l options which explicitly requests literal values to be printed. I'm ambivalent regarding behaviour or json and -l.
