Previously, working with alternates required two lookup arrays
and some indirection: for type Foo, we created Foo_qtypes[]
which maps each qtype to a value of the generated FooKind enum,
then look up that value in FooKind_lookup[] like we do for other
union types.
This has a couple of subtle bugs. First, the generator was
creating a call with a parameter '(int *) &(*obj)->type' where
type is an enum type; this is unsafe if the compiler chooses
to store the enum type in a different size than int, where
assigning through the wrong size pointer can corrupt data or
cause a SIGBUS.
Related bug, not not fixed in this patch: qapi-visit.py's
gen_visit_enum() generates a cast of its enum * argument to
int *. Marked FIXME.
Second, since the values of the FooKind enum start at zero, all
entries of the Foo_qtypes[] array that were not explicitly
initialized will map to the same branch of the union as the
first member of the alternate, rather than triggering a desired
failure in visit_get_next_type(). Fortunately, the bug seldom
bites; the very next thing the input visitor does is try to
parse the incoming JSON with the wrong parser, which normally
fails; the output visitor is not used with a C struct in that
state, and the dealloc visitor has nothing to clean up (so
there is no leak).
However, the second bug IS observable in one case: parsing an
integer causes unusual behavior in an alternate that contains
at least a 'number' member but no 'int' member, because the
'number' parser accepts QTYPE_QINT in addition to the expected
QTYPE_QFLOAT (that is, since 'int' is not a member, the type
QTYPE_QINT accidentally maps to FooKind 0; if this enum value
is the 'number' branch the integer parses successfully, but if
the 'number' branch is not first, some other branch tries to
parse the integer and rejects it). A later patch will worry
about fixing alternates to always parse all inputs that a
non-alternate 'number' would accept, for now this is still
marked FIXME in the updated test-qmp-input-visitor.c, to
merely point out that new undesired behavior of 'ans' matches
the existing undesired behavior of 'asn'.
This patch fixes the default-initialization bug by deleting the
indirection, and modifying get_next_type() to directly assign a
QTypeCode parameter. This in turn fixes the type-casting bug,
as we are no longer casting a pointer to enum to a questionable
size. There is no longer a need to generate an implicit FooKind
enum associated with the alternate type (since the QMP wire
format never uses the stringized counterparts of the C union
member names). Since the updated visit_get_next_type() does not
know which qtypes are expected, the generated visitor is
modified to generate an error statement if an unexpected type is
encountered.
Callers now have to know the QTYPE_* mapping when looking at the
discriminator; but so far, only the testsuite was even using the
C struct of an alternate types. I considered the possibility of
keeping the internal enum FooKind, but initialized differently
than most generated arrays, as in:
typedef enum FooKind {
FOO_KIND_A = QTYPE_QDICT,
FOO_KIND_B = QTYPE_QINT,
} FooKind;
to create nicer aliases for knowing when to use foo->a or foo->b
when inspecting foo->type; but it turned out to add too much
complexity, especially without a client.
There is a user-visible side effect to this change, but I
consider it to be an improvement. Previously,
the invalid QMP command:
{"execute":"blockdev-add", "arguments":{"options":
{"driver":"raw", "id":"a", "file":true}}}
failed with:
{"error": {"class": "GenericError",
"desc": "Invalid parameter type for 'file', expected: QDict"}}
(visit_get_next_type() succeeded, and the error comes from the
visit_type_BlockdevOptions() expecting {}; there is no mention of
the fact that a string would also work). Now it fails with:
{"error": {"class": "GenericError",
"desc": "Invalid parameter type for 'file', expected: BlockdevRef"}}
(the error when the next type doesn't match any expected types for
the overall alternate).
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1449033659-25497-5-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Now that we guarantee the user doesn't have any enum values
beginning with a single underscore, we can use that for our
own purposes. Renaming ENUM_MAX to ENUM__MAX makes it obvious
that the sentinel is generated.
This patch was mostly generated by applying a temporary patch:
|diff --git a/scripts/qapi.py b/scripts/qapi.py
|index e6d014b..b862ec9 100644
|--- a/scripts/qapi.py
|+++ b/scripts/qapi.py
|@@ -1570,6 +1570,7 @@ const char *const %(c_name)s_lookup[] = {
| max_index = c_enum_const(name, 'MAX', prefix)
| ret += mcgen('''
| [%(max_index)s] = NULL,
|+// %(max_index)s
| };
| ''',
| max_index=max_index)
then running:
$ cat qapi-{types,event}.c tests/test-qapi-types.c |
sed -n 's,^// \(.*\)MAX,s|\1MAX|\1_MAX|g,p' > list
$ git grep -l _MAX | xargs sed -i -f list
The only things not generated are the changes in scripts/qapi.py.
Rejecting enum members named 'MAX' is now useless, and will be dropped
in the next patch.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1447836791-369-23-git-send-email-eblake@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
[Rebased to current master, commit message tweaked]
Signed-off-by: Markus Armbruster <armbru@redhat.com>
The testsuite was only covering that we could output the 'int'
branch of an alternate (no additional allocation/cleanup required).
Add a test of the 'str' branch, to make sure that things still
work even when a branch involves allocation.
Update to modern style of g_new0() over g_malloc0() while
touching it.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1446791754-23823-9-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
By using &error_abort, we can avoid a local err variable in
situations where we expect success. It also has the nice
effect that if the test breaks, the error message from
error_abort tends to be nicer than that of g_assert().
This patch has an additional bonus of fixing several call sites that
were passing &err to two different functions without checking it in
between. In general that is unsafe practice; because if the first
function sets an error, the second function could abort() if it tries to
set a different error. We got away with it because we were asserting
that err was NULL through the entire chain, but switching to
&error_abort avoids the questionable practice up front.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1446791754-23823-7-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Make valgrind happy with the current state of the tests, so that
it is easier to see if future patches introduce new memory problems
without being drowned in noise. Many of the leaks were due to
calling a second init without tearing down the data from an earlier
visit. But since teardown is already idempotent, and we already
register teardown as part of input_visitor_test_add(), it is nicer
to just make init() safe to call multiple times than it is to have
to make all tests call teardown.
Another common leak was forgetting to clean up an error object,
after testing that an error was raised.
Another leak was in test_visitor_in_struct_nested(), failing to
clean the base member of UserDefTwo. Cleaning that up left
check_and_free_str() as dead code (since using the qapi_free_*
takes care of recursion, and we don't want double frees).
A final leak was in test_visitor_out_any(), which was reassigning
the qobj local variable to a subset of the overall structure
needing freeing; it did not result in a use-after-free, but
was not cleaning up all the qdict.
test-qmp-event and test-qmp-commands were already clean.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1446791754-23823-6-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Make each list element different, to ensure that order is
preserved, and use the generated free function instead of
hand-rolling our own to ensure (under valgrind) that the
list is properly cleaned.
Suggested-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1446791754-23823-3-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Commit d88f5fd and friends first introduced the various test-qmp-*
tests in 2011, with duplicated hand-rolled TestStruct machinery,
to make sure the qapi visitor interface was tested. Later, commit
4f193e3 in 2013 added a .json file for further testing use by the
files, but without consolidating any of the existing hand-rolled
visitors. And with four copies, subtle differences have crept in,
between the tests themselves (mainly whitespace differences, but
also a question of whether to use NULL or "TestStruct" when
calling visit_start_struct()) and from what the generator produces
(the hand-rolled versions did not cater to partially-allocated
objects, because they did not have a deallocation usage).
Of course, just because the visitor interface is tested does not
mean it is a sane interface; and future patches will be changing
some of the visitor contracts. Rather than having to duplicate
the cleanup work in each copy of the TestStruct visitor, and keep
each hand-rolled copy in sync with what the generator supplies, we
might as well just test what the generator should give us in the
first place.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1446791754-23823-2-git-send-email-eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
We have two issues with our qapi union layout:
1) Even though the QMP wire format spells the tag 'type', the
C code spells it 'kind', requiring some hacks in the generator.
2) The C struct uses an anonymous union, which places all tag
values in the same namespace as all non-variant members. This
leads to spurious collisions if a tag value matches a non-variant
member's name.
Make the conversion to the new layout for testsuite code.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1445898903-12082-15-git-send-email-eblake@redhat.com>
[Commit message tweaked slightly]
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Rather than storing a base class as a pointer to a box, just
store the fields of that base class in the same order, so that
a child struct can be directly cast to its parent. This gives
less malloc overhead, less pointer dereferencing, and even less
generated code. Compare to the earlier commit 1e6c1616a "qapi:
Generate a nicer struct for flat unions" (although that patch
had fewer places to change, as less of qemu was directly using
qapi structs for flat unions). It also allows us to turn on
automatic type-safe wrappers for upcasting to the base class
of a struct.
Changes to the generated code look like this in qapi-types.h:
| struct SpiceChannel {
|- SpiceBasicInfo *base;
|+ /* Members inherited from SpiceBasicInfo: */
|+ char *host;
|+ char *port;
|+ NetworkAddressFamily family;
|+ /* Own members: */
| int64_t connection_id;
as well as additional upcast functions like qapi_SpiceChannel_base().
Meanwhile, changes to qapi-visit.c look like:
| static void visit_type_SpiceChannel_fields(Visitor *v, SpiceChannel **obj, Error **errp)
| {
| Error *err = NULL;
|
|- visit_type_implicit_SpiceBasicInfo(v, &(*obj)->base, &err);
|+ visit_type_SpiceBasicInfo_fields(v, (SpiceBasicInfo **)obj, &err);
| if (err) {
(the cast is necessary, since our upcast wrappers only deal with a
single pointer, not pointer-to-pointer); plus the wholesale
elimination of some now-unused visit_type_implicit_FOO() functions.
Without boxing, the corner case of one empty struct having
another empty struct as its base type now requires inserting a
dummy member (previously, the 'Base *base' member sufficed).
And now that we no longer consume a 'base' member in the generated
C struct, we can delete the former negative struct-base-clash-base
test.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <1445898903-12082-11-git-send-email-eblake@redhat.com>
[Commit message tweaked slightly]
Signed-off-by: Markus Armbruster <armbru@redhat.com>
It's first class, because unlike '**', it actually works, i.e. doesn't
require 'gen': false.
'**' will go away next.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
Before commit 1d10b44, it crashed. Since then, it returns NULL, with
a FIXME comment. The FIXME is valid: code that assumes QObject *
can't be null exists. I'm not aware of a way to feed this problematic
return value to code that actually chokes on null in the current code,
but the next few commits will create one, failing "make check".
Commit 481b002 solved a very similar problem by introducing a special
null QObject. Using this special null QObject is clearly the right
way to resolve this FIXME, so do that, and update the test
accordingly.
However, the patch isn't quite right: it messes up the reference
counting. After about SIZE_MAX visits, the reference counter
overflows, failing the assertion in qnull_destroy_obj(). Because
that's many orders of magnitude more visits of nulls than we expect,
we take this patch despite its flaws, to get the QMP introspection
stuff in without further delay. We'll want to fix it for real before
the release.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
Message-Id: <1442401589-24189-21-git-send-email-armbru@redhat.com>
A flat union's tag member gets renamed to 'kind' in the generated
code. Breaks when another member named 'kind' exists.
Example, adapted from qapi-schema-test.json:
{ 'struct': 'UserDefUnionBase',
'data': { 'kind': 'str', 'enum1': 'EnumOne' } }
We generate:
struct UserDefFlatUnion
{
EnumOne kind;
union {
void *data;
UserDefA *value1;
UserDefB *value2;
UserDefB *value3;
};
char *kind;
};
Kill the silly rename.
Reported-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Now that qbool is fixed, let's fix getting and setting a bool
value to a qdict member to also use C99 bool rather than int.
I audited all callers to ensure that the changed return type
will not cause any changed semantics.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Acked-by: Luiz Capitulino <lcapitulino@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
We require a C99 compiler, so let's use 'bool' instead of 'int'
when dealing with boolean values. There are few enough clients
to fix them all in one pass.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Acked-by: Luiz Capitulino <lcapitulino@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
A future patch will be using a 'name':{dictionary} entry in the
QAPI schema to specify a default value for an optional argument;
but existing use of inline nested structs conflicts with that goal.
More precisely, a definition in the QAPI schema associates a name
with a set of properties:
Example 1: { 'struct': 'Foo', 'data': { MEMBERS... } }
associates the global name 'Foo' with properties (meta-type struct)
and MEMBERS...
Example 2: 'mumble': TYPE
within MEMBERS... above associates 'mumble' with properties (type
TYPE) and (optional false) within type Foo
The syntax of example 1 is extensible; if we need another property,
we add another name/value pair to the dictionary (such as
'base':TYPE). The syntax of example 2 is not extensible, because
the right hand side can only be a type.
We have used name encoding to add a property: "'*mumble': 'int'"
associates 'mumble' with (type int) and (optional true). Nice,
but doesn't scale. So the solution is to change our existing uses
to be syntactic sugar to an extensible form:
NAME: TYPE --> NAME: { 'type': TYPE, 'optional': false }
*ONAME: TYPE --> ONAME: { 'type': TYPE, 'optional': true }
This patch fixes the testsuite to avoid inline nested types, by
breaking the nesting into explicit types; it means that the type
is now boxed instead of unboxed in C code, but makes no difference
on the wire (and if desired, a later patch could change the
generator to not do so much boxing in C). When touching code to
add new allocations, also convert existing allocations to
consistently prefer typesafe g_new0 over g_malloc0 when a type
name is involved.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
In the testsuite, UserDefTwo and UserDefNested were identical
structs other than the member names. Reduce code duplication by
having just one type, and choose names that also favor reuse.
This will also make it easier for a later patch to get rid of
inline nested types in QAPI. When touching code related to
allocations, convert g_malloc0(sizeof(Type)) to the more typesafe
g_new0(Type, 1).
Ensure that 'make check-qapi-schema check-unit' still passes.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reduce churn in the future patch that replaces anonymous unions
with a new metatype 'alternate' by changing 'AnonUnion' to
'Alternate'.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
None of the existing QMP or QGA interfaces uses a union with a
base type but no discriminator; it is easier to avoid this in the
generator to save room for other future extensions more likely to
be useful. An earlier commit added a union-base-no-discriminator
test to ensure that we eventually give a decent error message;
likewise, removing UserDefUnion outright is okay, because we moved
all the tests we wish to keep into the tests of the simple union
UserDefNativeListUnion in the previous commit. Now is the time to
actually forbid simple union with base, and remove the last
vestiges from the testsuite.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
The tests of UserDefNativeListUnion serve to validate code
generation of simple unions without a base type, except that it
did not have full coverage in the strict test. The next commits
will remove tests and support for simple unions with a base type,
so there is no real loss at repurposing that test here as
opposed to churn of adding a new test then deleting the old one.
Fix some indentation and long lines while at it.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Checks the output visitor behaviour for NULL values.
Signed-off-by: Marcel Apfelbaum <marcel.a@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
We commonly use the error API like this:
err = NULL;
foo(..., &err);
if (err) {
goto out;
}
bar(..., &err);
Every error source is checked separately. The second function is only
called when the first one succeeds. Both functions are free to pass
their argument to error_set(). Because error_set() asserts no error
has been set, this effectively means they must not be called with an
error set.
The qapi-generated code uses the error API differently:
// *errp was initialized to NULL somewhere up the call chain
frob(..., errp);
gnat(..., errp);
Errors accumulate in *errp: first error wins, subsequent errors get
dropped. To make this work, the second function does nothing when
called with an error set. Requires non-null errp, or else the second
function can't see the first one fail.
This usage has also bled into visitor tests, and two device model
object property getters rtc_get_date() and balloon_stats_get_all().
With the "accumulate" technique, you need fewer error checks in
callers, and buy that with an error check in every callee. Can be
nice.
However, mixing the two techniques is confusing. You can't use the
"accumulate" technique with functions designed for the "check
separately" technique. You can use the "check separately" technique
with functions designed for the "accumulate" technique, but then
error_set() can't catch you setting an error more than once.
Standardize on the "check separately" technique for now, because it's
overwhelmingly prevalent.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
When visit_start_struct() fails, visit_end_struct() must not be
called. Three out of four visit_type_TestStruct() call it anyway. As
far as I can tell, visit_start_struct() doesn't actually fail there.
Fix them anyway.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Since enum based discriminators provide better type-safety and
ensure that future qapi additions do not forget to adjust dependent
unions, forbid using string as discriminator from now on.
Signed-off-by: Wenchao Xia <wenchaoqemu@gmail.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
The test demonstrates a generator bug: the generated struct
UserDefFlatUnion doesn't include members for the indirect base
UserDefZero.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
error_is_set(&var) is the same as var != NULL, but it takes
whole-program analysis to figure that out. Unnecessarily hard for
optimizers, static checkers, and human readers. Dumb it down to
obvious.
Gets rid of several dozen Coverity false positives.
Note that the obvious form is already used in many places.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
The macro g_assert_not_reached is a better self documenting replacement
for assert(0) or assert(false).
Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
With the introduction of native list types, we now have types such as
int64List where the 'value' field is not a pointer, but the actual
64-bit value.
On 32-bit architectures, this can lead to situations where 'next' field
offset in GenericList does not correspond to the 'next' field in the
types that we cast to GenericList when using the visit_next_list()
interface, causing issues when we attempt to traverse linked list
structures of these types.
To fix this, pad the 'value' field of GenericList and other
schema-defined/native *List types out to 64-bits.
This is less memory-efficient for 32-bit architectures, but allows us to
continue to rely on list-handling interfaces that target GenericList to
simply visitor implementations.
In the future we can improve efficiency by defaulting to using native C
array backends to handle list of non-pointer types, which would be more
memory efficient in itself and allow us to roll back this change.
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
This exercises schema-generated visitors for native list types and does
some sanity checking on validity of serialized data.
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Amos Kong <akong@redhat.com>
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
This introduces new test reporting infrastructure based on
gtester and gtester-report.
Also, all existing tests are moved to tests/, and tests/Makefile
is reorganized to factor out the commonalities in the rules.
Signed-off-by: Anthony Liguori <aliguori@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
