Re: [Qemu-devel] [RFC v4 02/13] qapi: qobject_compare() helper

[Eric Blake]

On 08/14/2017 04:57 PM, Eduardo Habkost wrote:
> The helper function will be useful when writing support code to
> deal with device slot information.
> 
> TODO: documentation is incomplete and unclear, needs to be
> improved.
> 
> Signed-off-by: Eduardo Habkost <address@hidden>
> ---
>  include/qapi/util.h    | 39 +++++++++++++++++++++++++++++
>  qapi/qapi-util.c       | 66 
> ++++++++++++++++++++++++++++++++++++++++++++++++++
>  tests/test-qapi-util.c | 53 ++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 158 insertions(+)
> 

> +/**
> + * qobject_compare:
> + *
> + * Compare the value of @a and @b.
> + *
> + * If @a and @b have the same type and the same value (see list
> + * of supported types below), return 0.
> + *
> + * If @a and @b are both strings, return strcmp(a, b).
> + *
> + * If @a and @b are numbers, return a negative value if a < b,
> + * and a positive value if a > b.
> + *
> + * Otherwise (if @a and @b are not the same, have different types,
> + * are of an unsupported type, or are different), return a non-zero value.

Is this number going to be commutative and distributive, in order to
provide stable qsort()ing?  That is, if comparing a and b gives a
positive number, then comparing b and a should give a negative number;
and if comparing a and b then b and c results in two positive numbers,
then comparing a and c should also give a positive number.  It is
unclear from the documentation whether you are able to make this
guarantee; and without it, it is unsafe to use this comparator in places
that require stability.

> + *
> + * Note that this function doesn't support some types, and may
> + * return false if the types are unsupported, or if the types don't
> + * match exactly.

How is a return of false (== 0, which also means equivalent) correct?

> + *
> + * Supported types:
> + * - QTYPE_QNULL
> + * - QTYPE_QSTRING
> + * - QTYPE_QBOOL
> + * - QTYPE_QNUM (integers only)
> + * - QTYPE_QLIST
> + *
> + * Unsupported (always return false):
> + * - QTYPE_QNUM (non-integer values)
> + * - QTYPE_QDICT
> + *
> + * TODO: rewrite documentation to be clearer.
> + * TODO: support non-integer QTYPE_NUM values and QTYPE_QDICT.

There's another patch series pending for qobject_is_equal(); should
these two patches share approaches or even code?

https://lists.gnu.org/archive/html/qemu-devel/2017-07/msg01134.html
https://lists.gnu.org/archive/html/qemu-devel/2017-08/msg02459.html

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org

signature.asc
Description: OpenPGP digital signature