Re: bonding options rework experiment (dcbw/bondopts)

[Dan Winship <danw gnome org>]

On 11/14/2013 08:57 PM, Dan Williams wrote:
> Hi,
>
> I've pushed my experiments here to dcbw/bondopts, enough that it
> compiles and passes 'make check' in libnm-util, although we certainly
> don't have enough testcase coverage of bonding to make sure it works.
>
> My main attempts were to use the normal GObject property ParamSpecs for
> min/max validation and all defaults, and to reduce some of the legacy
> property complexity by just exposing all the properties in the old
> 'options' property.

Looks great

  - NM_SETTING_BOND_PRIMARY_RESELECT is defined with a "_" rather than
    a "-" in its value. (I have no idea how I happened to notice this.)

  - what's the point of __NM_SETTING_BOND_MODE_IS_balance_rr(), etc?
    (they don't get used anywhere)

  - I'd rename "validate_list" to "validate_string", since that matches
    the TYPE_ enum better, and doesn't sound like it's validating a
    list of things.

  - I assume that if you remove the miimon/arp_interval magic, then
    you'll need to update nmcli to know to clear one when setting the
    other. (I'm assuming this because jklimes is the one who added that
    magic, not because I've actually looked at the nmcli code in
    question.)

  - ifcfg-rh/writer.c clearly doesn't compile as-is, so I guess you
    haven't tried to build src/ yet?