Re: [24.03 RFC 0/3] Add argument manipulation library

[Bruce Richardson <bruce.richardson@intel.com>]

On Thu, Nov 02, 2023 at 10:50:29AM -0700, Stephen Hemminger wrote:
> On Thu,  2 Nov 2023 17:28:46 +0000
> Bruce Richardson <bruce.richardson@intel.com> wrote:
> 
> > DPDK has traditionally assumed that apps are written where argc/argv
> > parameters are passed directly to rte_eal_init() and then app arguments
> > are handled afterwards, as is done in the DPDK apps and examples.
> > 
> > However, based on other projects, like VPP and OVS, we know that this is
> > often not the case. Instead, the apps build up argument lists internally
> > themselves and pass those to the init function, and DPDK never directly
> > accesses the real argc/argv values.
> > 
> > Therefore, let's make this mode of operation a little easier to use, by
> > adding in an args library to allow an app to dynamically build up an
> > args array and then pass that to EAL init. This library allows things
> > like, for example, initializing a list from argv and then checking if a
> > particular parameter is present, adding it if not. All the basics of
> > such a library are included in patch 1, and I've found it of use myself
> > when writing some simple apps internally using DPDK.
> > 
> > Patches 2 and 3 go a little further than this, and allow for a slightly
> > different use-case, that where an app may want to allow mixing of EAL
> > arguments in with app args. These patches allow the user to query if a
> > particular argument is an EAL arg or not, or a valid app argument. The
> > idea here would be, that an app could go through it's argument list,
> > building up an argument list for EAL init by just picking out the EAL
> > arguments from a full argument list. I'm less convinced of the value of
> > this compared to the basics in patch 1, but I think the idea is
> > interesting so seeking feedback on it.
> > 
> > Bruce Richardson (3):
> >   args: new library to allow easier manipulation of cmdline args
> >   eal: allow export of the cmdline argument parsing options
> >   args: add functions to check parameter validity
> > 
> >  doc/api/doxy-api-index.md           |   1 +
> >  doc/api/doxy-api.conf.in            |   1 +
> >  lib/args/args.c                     | 301 ++++++++++++++++++++++++++++
> >  lib/args/meson.build                |   5 +
> >  lib/args/rte_args.h                 | 255 +++++++++++++++++++++++
> >  lib/args/version.map                |  22 ++
> >  lib/eal/common/eal_common_options.c |   9 +
> >  lib/eal/include/rte_eal.h           |  14 ++
> >  lib/eal/version.map                 |   1 +
> >  lib/meson.build                     |   2 +
> >  10 files changed, 611 insertions(+)
> >  create mode 100644 lib/args/args.c
> >  create mode 100644 lib/args/meson.build
> >  create mode 100644 lib/args/rte_args.h
> >  create mode 100644 lib/args/version.map
> > 
> > --
> > 2.39.2
> > 
> 
> It would be good to change rte_eal_init() from:
> int rte_eal_init(int argc, char **argv);
> 
> to
> int rte_eal_init(int argc, char * const *argv);
> 
> I.e it should not modify the argv's passed.
> This would save unnecessary copy in applications that build synthetic args.
> Example already exists in dumpcap.
> 

Pushed up a quick patch proposal for that:
http://patches.dpdk.org/project/dpdk/patch/20231102181148.56930-1-bruce.richardson@intel.com/