Re: [PATCH v3 3/7] qapi/expr: Split check_expr out from check_exprs

[Markus Armbruster <armbru@redhat.com>]

Another observation...

John Snow <jsnow@redhat.com> writes:

> Primarily, this reduces a nesting level of a particularly long
> block. It's mostly code movement, but a new docstring is created.
>
> It also has the effect of creating a fairly convenient "catch point" in
> check_exprs for exception handling without making the nesting level even
> worse.
>
> Signed-off-by: John Snow <jsnow@redhat.com>
>
> ---
>
> This patch was originally written as part of my effort to factor out
> QAPISourceInfo from this file by having expr.py raise a simple
> exception, then catch and wrap it at the higher level.
>
> This series doesn't do that anymore, but reducing the nesting level
> still seemed subjectively nice. It's not crucial.
>
> Signed-off-by: John Snow <jsnow@redhat.com>
> ---
>  scripts/qapi/expr.py | 179 +++++++++++++++++++++++--------------------
>  1 file changed, 95 insertions(+), 84 deletions(-)
>
> diff --git a/scripts/qapi/expr.py b/scripts/qapi/expr.py
> index 5a1782b57ea..b56353bdf84 100644
> --- a/scripts/qapi/expr.py
> +++ b/scripts/qapi/expr.py
> @@ -595,6 +595,99 @@ def check_event(expr: _JSONObject, info: QAPISourceInfo) -> None:
>      check_type(args, info, "'data'", allow_dict=not boxed)
>  
>  
> +def check_expr(expr_elem: _JSONObject) -> None:

[...]

>  def check_exprs(exprs: List[_JSONObject]) -> List[_JSONObject]:
>      """
>      Validate and normalize a list of parsed QAPI schema expressions.

The typing is kind of wrong.

The list is the value of QAPISchemaParser.exprs, which is (losely) typed
as List[Dict[str, object]].  It is actually a dict mapping

   'expr' -> _ExprValue
   'info' -> QAPISourceInfo
   'doc'  -> Optional[QAPIDoc]

Thet's not what _JSONObject is!  Its doc string describes it as
"Deserialized JSON objects as returned by the parser", i.e. the same as
_ExprValue.

Works only because _JSONObject is a dict mapping arbitrary keys to
_JSONObject, str or bool.

This patch spreads the flawed typing to the new function.

Gets healed later in the series.

[...]