Re: [Qemu-devel] [PATCH v2 2/3] qapi: Add enum_table[] parameter to start_alternate

[Eric Blake]

On 05/05/2017 03:11 PM, Eduardo Habkost wrote:
> The new parameter will be used by the string input visitor to detect
> alternate types that can't be parsed unambiguously.
> 
> Signed-off-by: Eduardo Habkost <address@hidden>
> ---
> Changes v1 -> v2:
> * (new patch added to series)
> ---

> +++ b/include/qapi/visitor.h
> @@ -411,13 +411,21 @@ void visit_end_list(Visitor *v, void **list);
>   * @supported_qtypes is a bit mask indicating which QTypes are supported
>   * by the alternate.
>   *
> + * @enum_table contains the enum value lookup table, in case
> + * strings in the input are going to be parsed as enums. Visitors
> + * aren't required to validate string input according to
> + * enum_table, as visit_type_enum() will be called automatically
> + * if (*obj)->type is QTYPE_QSTRING.

Presumably, enum_table will be NULL if the alternate type does not
include an enum?  [reads ahead] yes. Should be documented.

I'm less convinced we need this patch, if we can instead guarantee at
QAPI-generation time that alternates are not possible if they would
cause an ambiguity.


> +++ b/scripts/qapi-visit.py
> @@ -166,6 +166,12 @@ def gen_visit_alternate(name, variants):
>      supported_qtypes = ' | '.join(qtypes)
>      ret = ''
>  
> +    enum_table = 'NULL'
> +    for var in variants.variants:
> +        if isinstance(var.type, QAPISchemaEnumType):
> +            enum_table = '%s_lookup' % (var.type.c_name())
> +            break
> +
>      ret += mcgen('''
>  
>  void visit_type_%(c_name)s(Visitor *v, const char *name, %(c_name)s **obj, 
> Error **errp)
> @@ -174,7 +180,7 @@ void visit_type_%(c_name)s(Visitor *v, const char *name, 
> %(c_name)s **obj, Error
>      uint32_t supported_qtypes = %(supported_qtypes)s;
>  
>      visit_start_alternate(v, name, (GenericAlternate **)obj, sizeof(**obj),
> -                          supported_qtypes, &err);
> +                          supported_qtypes, %(enum_table)s, &err);

So there's where you populate the enum through.

I'll have to see where 3/3 goes before deciding if this is worth having.

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

signature.asc
Description: OpenPGP digital signature