Re: [Qemu-devel] [PATCH RFC v4 02.5/32] qapi: Hide internal data members of schema objects.

[Markus Armbruster]

Eric Blake <address@hidden> writes:

> We have a few fields that exist mainly to hold information from
> __init__() until check() (matching the fact that parsing is
> two-pass; the first to find type names, the second to associate
> types together while honoring forward references), or which should
> only be used through accessor methods.  We should not use these
> fields directly in other files after check() has run, so use the
> python convention of naming these fields with leading underscore
> to mark their internal usage, and to check that no one else was
> using them.
>
> Exception: our crazy handling of simple unions (with a C member
> 'kind' matching the QMP wire 'type') requires peeking through
> the hidden field.  This leaky abstraction will be cleaned up in
> a later patch.

There are two more peeks: QAPISchemaAlternateType peeks when it asserts
it got simple variants, and test-qapi.py peeks when it prints a flat
union's tag.  You missed the latter, and neglected to mention the
former.

> Signed-off-by: Eric Blake <address@hidden>
> ---
>
> Technically, I wrote this patch after 32; if you decide to rebase
> it into the series, you'll have to split it among 2, 10, 11, and 30.
> Up to you if you want to squash this in during your spin of v5, or
> if you want me to keep it as a separate patch for inclusion after
> your series.

I doubt the improvement worth the churn as a follow-up patch.  Instead,
I'll squash it in, less the rename of .tag_name, plus a few more renames
elsewhere.

Thanks!