[v4,0/4] qapi: generalize special features

Message

Daniel P. Berrangé Feb. 5, 2025, 12:35 p.m. UTC

This series is a spin-off from

  https://lists.nongnu.org/archive/html/qemu-devel/2024-06/msg00807.html

That series introduced a pragma allowing a schema to declare extra
features that would be exposed to code.

Following Markus' suggestion:

  https://lists.nongnu.org/archive/html/qemu-devel/2024-07/msg03765.html

I've changed impl such that we expose all features to the code
regardless of whether they are special, and don't require any pragma.

I've split it from the QGA patches since it makes more sense to work
on this bit in isolation.

Changed in v4:

 * Apply fixes to make mypy happy

Changed in v3:

 * "make -C python check-tox" is now clean
 * Added test case for "too many features" error scenario
 * Don't sort the features enum, just rely on ordered dict
 * Add 'features' accessor on QAPISchema instead of touching
   private data via 'schema._features.values()'

Changed in v2:

 * Reference QapiSpecialFeature enums constants when defining
   QapiFeature enums constants for deprecated & unstable

 * Don't expose qapi-features.h in qapi-types.h, to avoid
   global namespace pollution, instead pull it into only the
   .c files that need it. This avoids the need to add a 'prefix'
   to enum constants

 * Collect all features during parse time, rather than
   generate time, to allow earlier error reporting of the
   64 feature limit

Daniel P. Berrangé (4):
  qapi: cope with  feature names containing a '-'
  qapi: change 'unsigned special_features' to 'uint64_t features'
  qapi: rename 'special_features' to 'features'
  qapi: expose all schema features to code

 include/qapi/compat-policy.h             |  2 +-
 include/qapi/qmp/dispatch.h              |  4 +-
 include/qapi/util.h                      |  2 +-
 include/qapi/visitor-impl.h              |  4 +-
 include/qapi/visitor.h                   | 12 +++---
 meson.build                              |  1 +
 qapi/qapi-forward-visitor.c              |  8 ++--
 qapi/qapi-util.c                         |  6 +--
 qapi/qapi-visit-core.c                   | 12 +++---
 qapi/qmp-dispatch.c                      |  2 +-
 qapi/qmp-registry.c                      |  4 +-
 qapi/qobject-input-visitor.c             |  4 +-
 qapi/qobject-output-visitor.c            |  6 +--
 scripts/qapi/commands.py                 |  5 ++-
 scripts/qapi/features.py                 | 51 ++++++++++++++++++++++++
 scripts/qapi/gen.py                      |  9 +++--
 scripts/qapi/main.py                     |  2 +
 scripts/qapi/schema.py                   | 31 +++++++++++++-
 scripts/qapi/types.py                    | 19 +++++----
 scripts/qapi/visit.py                    | 17 ++++----
 tests/meson.build                        |  2 +
 tests/qapi-schema/features-too-many.err  |  2 +
 tests/qapi-schema/features-too-many.json | 13 ++++++
 tests/qapi-schema/features-too-many.out  |  0
 tests/qapi-schema/meson.build            |  1 +
 25 files changed, 163 insertions(+), 56 deletions(-)
 create mode 100644 scripts/qapi/features.py
 create mode 100644 tests/qapi-schema/features-too-many.err
 create mode 100644 tests/qapi-schema/features-too-many.json
 create mode 100644 tests/qapi-schema/features-too-many.out

Comments

Markus Armbruster Feb. 10, 2025, 8:25 a.m. UTC | #1

Daniel P. Berrangé <[e-mail]> writes:

> This series is a spin-off from
>
>   https://lists.nongnu.org/archive/html/qemu-devel/2024-06/msg00807.html
>
> That series introduced a pragma allowing a schema to declare extra
> features that would be exposed to code.
>
> Following Markus' suggestion:
>
>   https://lists.nongnu.org/archive/html/qemu-devel/2024-07/msg03765.html
>
> I've changed impl such that we expose all features to the code
> regardless of whether they are special, and don't require any pragma.
>
> I've split it from the QGA patches since it makes more sense to work
> on this bit in isolation.

Series
Reviewed-by: Markus Armbruster <[e-mail]>

Markus Armbruster Feb. 10, 2025, 8:37 a.m. UTC | #2

Queued with imports tidied up.  Thanks!