[v11,28/28] qapi: Detect base class loops

It should be fairly obvious that qapi base classes need to
form an acyclic graph, since QMP cannot specify the same
key more than once, while base classes are included as flat
members alongside other members added by the child.  But the
old check_member_clash() parser function was not prepared to
check for this, and entered an infinite recursion (at least
until python gives up, complaining about nesting too deep).

Now that check_member_clash() has been recently removed,
attempts at self-inheritance trigger an assertion failure
introduced by commit ac88219a.  The obvious fix is to turn
the assertion into a conditional.

This patch includes both the test and the fix, since the .err
file output for the unfixed case is not useful (particularly
when it was warning about unbounded recursion, as that limit
may be platform-specific).

We don't need to worry about cycles in flat unions (neither
the base nor a variant class can be a union) nor in alternates
(alternate branches cannot themselves be an alternate).  And
even the case of using the same class for a flat union base
class and one its variants is already caught by the fact that
both uses will introduce the same member name (or will be okay
if the class is empty).

Signed-off-by: Eric Blake <eblake@redhat.com>

---
v11 (no v10): rename base-cycle to base-cycle-indirect, and add
base-cycle-direct; touch up commit message
v9: no change
v8: improve commit message
v7: improve commit message
v6: rebase to earlier info changes
---
 scripts/qapi.py                            | 6 +++++-
 tests/Makefile                             | 2 ++
 tests/qapi-schema/base-cycle-direct.err    | 1 +
 tests/qapi-schema/base-cycle-direct.exit   | 1 +
 tests/qapi-schema/base-cycle-direct.json   | 2 ++
 tests/qapi-schema/base-cycle-direct.out    | 0
 tests/qapi-schema/base-cycle-indirect.err  | 1 +
 tests/qapi-schema/base-cycle-indirect.exit | 1 +
 tests/qapi-schema/base-cycle-indirect.json | 3 +++
 tests/qapi-schema/base-cycle-indirect.out  | 0
 10 files changed, 16 insertions(+), 1 deletion(-)
 create mode 100644 tests/qapi-schema/base-cycle-direct.err
 create mode 100644 tests/qapi-schema/base-cycle-direct.exit
 create mode 100644 tests/qapi-schema/base-cycle-direct.json
 create mode 100644 tests/qapi-schema/base-cycle-direct.out
 create mode 100644 tests/qapi-schema/base-cycle-indirect.err
 create mode 100644 tests/qapi-schema/base-cycle-indirect.exit
 create mode 100644 tests/qapi-schema/base-cycle-indirect.json
 create mode 100644 tests/qapi-schema/base-cycle-indirect.out

diff --git a/tests/qapi-schema/base-cycle-indirect.out b/tests/qapi-schema/base-cycle-indirect.out
new file mode 100644
index 0000000..e69de29

Eric Blake <eblake@redhat.com> writes:

> It should be fairly obvious that qapi base classes need to
> form an acyclic graph, since QMP cannot specify the same
> key more than once, while base classes are included as flat
> members alongside other members added by the child.  But the
> old check_member_clash() parser function was not prepared to
> check for this, and entered an infinite recursion (at least
> until python gives up, complaining about nesting too deep).

Nitpick: Python.

>
> Now that check_member_clash() has been recently removed,
> attempts at self-inheritance trigger an assertion failure
> introduced by commit ac88219a.  The obvious fix is to turn
> the assertion into a conditional.

Exactly.  Like all the assertions in check() methods, it was always
meant as a placeholder for a semantic check.

> This patch includes both the test and the fix, since the .err
> file output for the unfixed case is not useful (particularly
> when it was warning about unbounded recursion, as that limit
> may be platform-specific).
>
> We don't need to worry about cycles in flat unions (neither
> the base nor a variant class can be a union)

(neither the base type nor the type of a variant ...)

Note that if they could, this check would report the cycle just fine.

>                                              nor in alternates
> (alternate branches cannot themselves be an alternate).

If they could, all it took to detect the cycle would be a v.type.check()
in QAPISchemaAlternateType.check().

>                                                          And
> even the case of using the same class for a flat union base
> class and one its variants is already caught by the fact that
> both uses will introduce the same member name (or will be okay
> if the class is empty).

That's not a cycle, that's a diamond, isn't it?

> Signed-off-by: Eric Blake <eblake@redhat.com>
>
> ---
> v11 (no v10): rename base-cycle to base-cycle-indirect, and add
> base-cycle-direct; touch up commit message
> v9: no change
> v8: improve commit message
> v7: improve commit message
> v6: rebase to earlier info changes
> ---
>  scripts/qapi.py                            | 6 +++++-
>  tests/Makefile                             | 2 ++
>  tests/qapi-schema/base-cycle-direct.err    | 1 +
>  tests/qapi-schema/base-cycle-direct.exit   | 1 +
>  tests/qapi-schema/base-cycle-direct.json   | 2 ++
>  tests/qapi-schema/base-cycle-direct.out    | 0
>  tests/qapi-schema/base-cycle-indirect.err  | 1 +
>  tests/qapi-schema/base-cycle-indirect.exit | 1 +
>  tests/qapi-schema/base-cycle-indirect.json | 3 +++
>  tests/qapi-schema/base-cycle-indirect.out  | 0
>  10 files changed, 16 insertions(+), 1 deletion(-)
>  create mode 100644 tests/qapi-schema/base-cycle-direct.err
>  create mode 100644 tests/qapi-schema/base-cycle-direct.exit
>  create mode 100644 tests/qapi-schema/base-cycle-direct.json
>  create mode 100644 tests/qapi-schema/base-cycle-direct.out
>  create mode 100644 tests/qapi-schema/base-cycle-indirect.err
>  create mode 100644 tests/qapi-schema/base-cycle-indirect.exit
>  create mode 100644 tests/qapi-schema/base-cycle-indirect.json
>  create mode 100644 tests/qapi-schema/base-cycle-indirect.out
>
> diff --git a/scripts/qapi.py b/scripts/qapi.py
> index 08a366e..e60c1d8 100644
> --- a/scripts/qapi.py
> +++ b/scripts/qapi.py
> @@ -941,7 +941,11 @@ class QAPISchemaObjectType(QAPISchemaType):
>          self.members = None
>
>      def check(self, schema):
> -        assert self.members is not False        # not running in cycles
> +        if self.members is False:               # check for cycles
> +            assert self._base_name
> +            raise QAPIExprError(self.info,
> +                                "Object %s cyclically depends on %s"
> +                                % (self.name, self._base_name))

Well, it cyclically depends on *itself*.

If the cycle is of length one, this message becomes

    Object Foo cyclically depends on Foo

Sounds awkward.

If the cycle is longer, it becomes

    Object Foo cyclically depends on Bar

where Foo -> Bar happens to be the beginning of the cycle.  Borders on
misleading.

We could print the complete cycle, but that feels like too much work for
too little gain.  Let's dumb down the message to something like "Type
Foo contains itself"[*].
  
>          if self.members:
>              return
>          self.members = False                    # mark as being checked
> diff --git a/tests/Makefile b/tests/Makefile
> index cdff7a4..2706126 100644
> --- a/tests/Makefile
> +++ b/tests/Makefile
> @@ -253,6 +253,8 @@ qapi-schema += bad-ident.json
>  qapi-schema += bad-type-bool.json
>  qapi-schema += bad-type-dict.json
>  qapi-schema += bad-type-int.json
> +qapi-schema += base-cycle-direct.json
> +qapi-schema += base-cycle-indirect.json
>  qapi-schema += command-int.json
>  qapi-schema += comments.json
>  qapi-schema += double-data.json
> diff --git a/tests/qapi-schema/base-cycle-direct.err b/tests/qapi-schema/base-cycle-direct.err
> new file mode 100644
> index 0000000..46fc436
> --- /dev/null
> +++ b/tests/qapi-schema/base-cycle-direct.err
> @@ -0,0 +1 @@
> +tests/qapi-schema/base-cycle-direct.json:2: Object Foo cyclically depends on Foo
> diff --git a/tests/qapi-schema/base-cycle-direct.exit b/tests/qapi-schema/base-cycle-direct.exit
> new file mode 100644
> index 0000000..d00491f
> --- /dev/null
> +++ b/tests/qapi-schema/base-cycle-direct.exit
> @@ -0,0 +1 @@
> +1
> diff --git a/tests/qapi-schema/base-cycle-direct.json b/tests/qapi-schema/base-cycle-direct.json
> new file mode 100644
> index 0000000..dff6fab
> --- /dev/null
> +++ b/tests/qapi-schema/base-cycle-direct.json
> @@ -0,0 +1,2 @@
> +# we reject a loop in base classes
> +{ 'struct': 'Foo', 'base': 'Foo', 'data': {} }

I gather you didn't like 'Loopy'.  Pity...

> diff --git a/tests/qapi-schema/base-cycle-direct.out b/tests/qapi-schema/base-cycle-direct.out
> new file mode 100644
> index 0000000..e69de29
> diff --git a/tests/qapi-schema/base-cycle-indirect.err b/tests/qapi-schema/base-cycle-indirect.err
> new file mode 100644
> index 0000000..9a1ca44
> --- /dev/null
> +++ b/tests/qapi-schema/base-cycle-indirect.err
> @@ -0,0 +1 @@
> +tests/qapi-schema/base-cycle-indirect.json:2: Object Base1 cyclically depends on Base2
> diff --git a/tests/qapi-schema/base-cycle-indirect.exit b/tests/qapi-schema/base-cycle-indirect.exit
> new file mode 100644
> index 0000000..d00491f
> --- /dev/null
> +++ b/tests/qapi-schema/base-cycle-indirect.exit
> @@ -0,0 +1 @@
> +1
> diff --git a/tests/qapi-schema/base-cycle-indirect.json b/tests/qapi-schema/base-cycle-indirect.json
> new file mode 100644
> index 0000000..2866772
> --- /dev/null
> +++ b/tests/qapi-schema/base-cycle-indirect.json
> @@ -0,0 +1,3 @@
> +# we reject a loop in base classes
> +{ 'struct': 'Base1', 'base': 'Base2', 'data': {} }
> +{ 'struct': 'Base2', 'base': 'Base1', 'data': {} }
> diff --git a/tests/qapi-schema/base-cycle-indirect.out b/tests/qapi-schema/base-cycle-indirect.out
> new file mode 100644
> index 0000000..e69de29


[*] I'm tempted to make it "Foo is a type, not a Klein bottle", but I
acknowledge our users may not share my sense of humor ;)