Syntax error coverage should now be complete.
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>
Reviewed-by: NEric Blake <eblake@redhat.com>

The first check ensures the second one can't trigger.  Drop the first
one, because the second one is in a more logical place, and emits a
nicer error message.
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>
Reviewed-by: NEric Blake <eblake@redhat.com>

Reproducer: with

    { 'command': 'user_def_cmd4', 'returns': { 'a': 'int' } }

added to qapi-schema-test.json, qapi-commands.py dies when it tries to
generate the command handler function

    Traceback (most recent call last):
      File "/work/armbru/qemu/scripts/qapi-commands.py", line 359, in <module>
        ret = generate_command_decl(cmd['command'], arglist, ret_type) + "\n"
      File "/work/armbru/qemu/scripts/qapi-commands.py", line 29, in generate_command_decl
        ret_type=c_type(ret_type), name=c_name(name),
      File "/work/armbru/qemu/scripts/qapi.py", line 927, in c_type
        assert isinstance(value, str) and value != ""
    AssertionError

because the return type doesn't exist.

Simply outlaw this usage, and drop or dumb down test cases accordingly.
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>
Reviewed-by: NEric Blake <eblake@redhat.com>

A command's or event's 'data' must be a struct type, given either as a
dictionary, or as struct type name.

Commit dd883c6f tightened the checking there, but not enough: we still
accept 'union'.  Fix to reject it.

We may want to support union types there, but we'll have to extend
qapi-commands.py and qapi-events.py for it.
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>
Reviewed-by: NEric Blake <eblake@redhat.com>

A command's 'data' must be a struct type, given either as a
dictionary, or as struct type name.

Existing test case data-int.json covers simple type 'int'.  Add test
cases for type names referring to union and alternate types.

The latter is caught (good), but the former is not (bug).

Events have the same problem, but since they get checked by the same
code, we don't bother to duplicate the tests.
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>
Reviewed-by: NEric Blake <eblake@redhat.com>

Since every schema entity has 'data', the data- prefix conveys no
information.  These tests actually exercise commands.  Only commands
have arguments, so change the prefix to to args-.
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>
Reviewed-by: NEric Blake <eblake@redhat.com>

Test case added in commit 2fc00432, and messed up in commit 5223070c.
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>
Reviewed-by: NEric Blake <eblake@redhat.com>

When event FOO's 'data' is a struct with a base, we consider only the
struct's direct members, and ignore its base.  The generated
qapi_event_send_foo() doesn't take arguments for base members.

No such events currently exist in the QMP schema.
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>
Reviewed-by: NEric Blake <eblake@redhat.com>

We generate a declaration, but no definition.

The QMP schema has two: Qcow2OverlapChecks and BlockdevRef.  Neither
visit_type_Qcow2OverlapChecksKind() nor visit_type_BlockdevRefKind()
is actually used.
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>
Reviewed-by: NEric Blake <eblake@redhat.com>

The visit_type_implicit_FOO() are generated on demand, right before
their first use.  Used by visit_type_STRUCT_fields() when STRUCT has
base FOO, and by visit_type_UNION() when flat UNION has member a FOO.

If the schema defines FOO after its first use as struct base or flat
union member, visit_type_implicit_FOO() calls
visit_type_implicit_FOO() before its definition, which doesn't
compile.

Rearrange qapi-schema-test.json to demonstrate the bug.

Fix by generating the necessary forward declaration.
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>
Reviewed-by: NEric Blake <eblake@redhat.com>

Signed-off-by: NMarkus Armbruster <armbru@redhat.com>
Reviewed-by: NEric Blake <eblake@redhat.com>

The new test demonstrates another generator crash.
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>
Reviewed-by: NEric Blake <eblake@redhat.com>

We print the name as it appears in the include expression.  Tools
processing error messages want it relative to the working directory.
Make it so.
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>
Reviewed-by: NEric Blake <eblake@redhat.com>

Enhance the testsuite to cover downstream events and commands.
Events worked without more tweaks, but commands needed a few final
updates in the generator to mangle names in the appropriate places.
In making those tweaks, it was easier to drop type_visitor() and
inline its actions instead.
Signed-off-by: NEric Blake <eblake@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

Enhance the testsuite to cover downstream alternates, including
whether the branch name or type is downstream.  Update the
generator to mangle alternate names in the appropriate places.
Signed-off-by: NEric Blake <eblake@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

Enhance the testsuite to cover downstream flat unions, including
the base type, discriminator name and type, and branch name and
type.  Update the generator to mangle the union names in the
appropriate places.
Signed-off-by: NEric Blake <eblake@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

Enhance the testsuite to cover downstream simple unions, including
when a union branch is a downstream name.  Update the generator to
mangle the union names in the appropriate places.
Signed-off-by: NEric Blake <eblake@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

Enhance the testsuite to cover downstream structs, including struct
members and base structs.  Update the generator to mangle the
struct names in the appropriate places.
Signed-off-by: NEric Blake <eblake@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

Enhance the testsuite to cover a downstream enum type and enum
string.  Update the generator to mangle the enum name in the
appropriate places.
Signed-off-by: NEric Blake <eblake@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

Our type inheritance for both 'struct' and for flat 'union' merges
key/value pairs from the base class with those from the type in
question.  Although the C code currently boxes things so that there
is a distinction between which member is referred to, the QMP wire
format does not allow passing a key more than once in a single
object.  Besides, if we ever change the generated C code to not be
quite so boxy, we'd want to avoid duplicate member names there,
too.

Fix a testsuite entry added in an earlier patch, as well as adding
a couple more tests to ensure we have appropriate coverage.  Ensure
that collisions are detected, regardless of whether there is a
difference in opinion on whether the member name is optional.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

The handling of \ inside QAPI strings was less than ideal, and
really only worked JSON's \/, \\, \", and our extension of \'
(an obvious extension, when you realize we use '' instead of ""
for strings).  For other things, like '\n', it resulted in a
literal 'n' instead of a newline.

Of course, at the moment, we really have no use for escaped
characters, as QAPI has to map to C identifiers, and we currently
support ASCII only for that.  But down the road, we may add
support for default values for string parameters to a command
or struct; if that happens, it would be nice to correctly support
all JSON escape sequences, such as \n or \uXXXX.  This gets us
closer, by supporting Unicode escapes in the ASCII range.

Since JSON does not require \OCTAL or \xXX escapes, and our QMP
implementation does not understand them either, I intentionally
reject it here, but it would be an easy addition if we desired it.
Likewise, intentionally refusing the NUL byte means we don't have
to worry about C strings being shorter than the qapi input.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus 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
(see previous commit messages for more details why); but existing
use of inline nested structs conflicts with that goal. Now that
all commands have been changed to avoid inline nested structs,
nuke support for them, and turn it into a hard error. Update the
testsuite to reflect tighter parsing rules.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus 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: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus 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: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

Referring to "type" as both a meta-type (built-in, enum, union,
alternate, or struct) and a specific type (the name that the
schema uses for declaring structs) is confusing.  Do the bulk of
the conversion to "struct" in qapi schema, with a fairly
mechanical:

for f in `find -name '*.json'; do sed -i "s/'type'/'struct'/"; done

followed by manually filtering out the places where we have a
'type' embedded in 'data'.  Then tweak a couple of tests whose
output changes slightly due to longer lines.

I also verified that the generated files for QMP and QGA (such
as qmp-commands.h) are the same before and after, as assurance
that I didn't leave in any accidental member name changes.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

Referring to "type" as both a meta-type (built-in, enum, union,
alternate, or struct) and a specific type (the name that the
schema uses for declaring structs) is confusing.  The confusion
is only made worse by the fact that the generator mostly already
refers to struct even when dealing with expr['type'].  This
commit changes the generator to consistently refer to it as
struct everywhere, plus a single back-compat tweak that allows
accepting the existing .json files as-is, so that the meat of
this change is separate from the mindless churn of that change.

Fix the testsuite fallout for error messages that change, and
in some cases, become more legible.  Improve comments to better
match our intentions where a struct (rather than any complex
type) is required.  Note that in some cases, an error message
now refers to 'struct' while the schema still refers to 'type';
that will be cleaned up in the later commit to the schema.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

Now that we have a way to validate every type, we can also be
stricter about enforcing that callers that want to bypass
type safety in generated code.  Prior to this patch, it didn't
matter what value was associated with the key 'gen', but it
looked odd that 'gen':'yes' could result in bypassing the
generated code.  These changes also enforce the changes made
earlier in the series for documentation and consolidation of
using '**' as the wildcard type, as well as 'gen':false as the
canonical spelling for requesting type bypass.

Note that 'gen':false is a one-way switch away from the default;
we do not support 'gen':true (similar for 'success-response').
In practice, this doesn't matter.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

...or an array of dictionaries.  Although we have to cater to
existing commands, returning a non-dictionary means the command
is not extensible (no new name/value pairs can be added if more
information must be returned in parallel).  By making the
whitelist explicit, any new command that falls foul of this
practice will have to be self-documenting, which will encourage
developers to either justify the action or rework the design to
use a dictionary after all.

It's a little bit sloppy that we share a single whitelist among
three clients (it's too permissive for each).  If this is a
problem, a future patch could tighten things by having the
generator take the whitelist as an argument (as in
scripts/qapi-commands.py --legacy-returns=...), or by having
the generator output C code that requires explicit use of the
whitelist (as in:
 #ifndef FROBNICATE_LEGACY_RETURN_OK
 # error Command 'frobnicate' should return a dictionary
 #endif
then having the callers define appropriate macros).  But until
we need such fine-grained separation (if ever), this patch does
the job just fine.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

Previous commits demonstrated that the generator overlooked various
bad naming situations:
- types, commands, and events need a valid name
- enum members must be valid names, when combined with prefix
- union and alternate branches cannot be marked optional

Valid upstream names match [a-zA-Z][a-zA-Z0-9_-]*; valid downstream
names match __[a-zA-Z][a-zA-Z0-9._-]*.  Enumerations match the
weaker [a-zA-Z0-9._-]+ (in part thanks to QKeyCode picking an enum
that starts with a digit, which we can't change now due to
backwards compatibility).  Rather than call out three separate
regex, this patch just uses a broader combination that allows both
upstream and downstream names, as well as a small hack that
realizes that any enum name is merely a suffix to an already valid
name prefix (that is, any enum name is valid if prepending _ fits
the normal rules).

We could reject new enumeration names beginning with a digit by
whitelisting existing exceptions.  We could also be stricter
about the distinction between upstream names (no leading
underscore, no use of dot) and downstream (mandatory leading
double underscore), but it is probably not worth the bother.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

Now that we know every expression is valid with regards to
its keys, we can add further tests that those keys refer to
valid types.  With this patch, all uses of a type (the 'data':
of command, type, union, alternate, and event; the 'returns':
of command; the 'base': of type and union) must resolve to an
appropriate subset of metatypes  declared by the current qapi
parse; this includes recursing into each member of a data
dictionary.  Dealing with '**' and nested anonymous structs
will be done in later patches.

Update the testsuite to match improved output.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

Demonstrate that the qapi generator silently parses confusing
types, which may cause other errors later on. Later patches
will update the expected results as the generator is made stricter.

Most of the new tests focus on blatant errors.  But
returns-whitelist is a case where we have historically allowed
returning something other than a JSON object from particular
commands; we have to keep that behavior to avoid breaking clients,
but it would be nicer to avoid adding such commands in the future,
because any return that is not an (array of) object cannot be
easily extended if future qemu wants to return additional
information.  The QMP protocol already documents that clients
should ignore unknown dictionary keys, but does not require
clients to have to handle more than one type of JSON object.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

For a few QMP commands, we are forced to pass an arbitrary type
without tracking it properly in QAPI.  Among the existing clients,
this unnamed type was spelled 'dict', 'visitor', and '**'; this
patch standardizes on '**', matching the documentation changes
earlier in the series.

Meanwhile, for the 'gen' key, we have been ignoring the value,
although the schema consistently used "'no'" ('success-response'
was hard-coded to checking for 'no').  But now that we can support
a literal "false" in the schema, we might as well use that rather
than ignoring the value or special-casing a random string.  Note
that these are one-way switches (use of 'gen':true is not the same
as omitting 'gen'). Also, the use of '**' requires 'gen':false,
but the use of 'gen':false does not mandate the use of '**'.

There is no difference to the generated code.  Add some tests on
what we'd like to guarantee, although it will take later patches
to clean up test results and actually enforce the use of a bool
parameter.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

In the near term, we will use it for a sensible-looking
'gen':false inside command declarations, instead of the
current ugly 'gen':'no'.

In the long term, it will allow conversion from shorthand
with defaults mentioned only in side-band documentation:
 'data':{'*flag':'bool', '*string':'str'}
into an explicit default value documentation, as in:
 'data':{'flag':{'type':'bool', 'optional':true, 'default':true},
         'string':{'type':'str', 'optional':true, 'default':null}}

We still don't parse integer values (also necessary before
we can allow explicit defaults), but that can come in a later
series.

Update the testsuite to match an improved error message.
Signed-off-by: NFam Zheng <famz@redhat.com>
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

The previous commit demonstrated that the generator overlooked
duplicate expressions:
- a complex type or command reusing a built-in type name
- redeclaration of a type name, whether by the same or different
metatype
- redeclaration of a command or event
- collision of a type with implicit 'Kind' enum for a union
- collision with an implicit MAX enum constant

Since the c_type() function in the generator treats all names
as being in the same namespace, this patch adds a global array
to track all known names and their source, to prevent collisions
before it can cause further problems.  While valid .json files
won't trigger any of these cases, we might as well be nicer to
developers that make a typo while trying to add new QAPI code.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

Demonstrate that the qapi generator doesn't deal very well with
redefined expressions.  At the parse level, they are silently
accepted; and while the testsuite just stops at parsing, I've
further tested that many of them cause generator crashes or
invalid C code if they were appended to qapi-schema-test.json.
A later patch will tighten things up and adjust the testsuite
to match.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

The previous commit demonstrated that the generator overlooked some
fairly basic broken expressions:
- missing metataype
- metatype key has a non-string value
- unknown key in relation to the metatype
- conflicting metatype (this patch treats the second metatype as an
unknown key of the first key visited, which is not necessarily the
first key the user typed)

Add check_keys to cover these situations, and update testcases to
match.  A couple other tests (enum-missing-data, indented-expr) had
to change since the validation added here occurs so early.
Conversely, changes to ident-with-escape results show that we still
have problems where our handling of escape sequences differs from
true JSON, which will matter down the road if we allow arbitrary
default string values for optional parameters (but for now is not
too bad, as we currently can avoid unicode escaping as we don't
need to represent anything beyond C identifier material).

While valid .json files won't trigger any of these cases, we might
as well be nicer to developers that make a typo while trying to add
new QAPI code.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

Demonstrate that the qapi generator doesn't deal well with
expressions that aren't up to par. Later patches will improve
the expected results as the generator is made stricter.  Only
a few of the the added tests actually behave sanely at
rejecting obvious problems or demonstrating success.

Note that in some cases, we reject bad QAPI merely because our
pseudo-JSON parser does not yet know how to parse numbers.  This
series does not address that, but when a later series adds support
for numeric defaults of integer fields, the testsuite will ensure
that we don't lose the error (and hopefully that the error
message quality is improved).
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

Previous patches have led up to the point where I create the
new meta-type "'alternate':'Foo'".  See the previous patches
for documentation; I intentionally split as much work into
earlier patches to minimize the size of this patch, but a lot
of it is churn due to testsuite fallout after updating to the
new type.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus 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: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>

Special-casing 'discriminator == {}' for handling anonymous unions
is getting awkward; since this particular type is not always a
dictionary on the wire, it is easier to treat it as a completely
different class of type, "alternate", so that if a type is listed
in the union_types array, we know it is not an anonymous union.

This patch just further segregates union handling, to make sure that
anonymous unions are not stored in union_types, and splitting up
check_union() into separate functions.  A future patch will change
the qapi grammar, and having the segregation already in place will
make it easier to deal with the distinct meta-type.
Signed-off-by: NEric Blake <eblake@redhat.com>
Reviewed-by: NMarkus Armbruster <armbru@redhat.com>
Signed-off-by: NMarkus Armbruster <armbru@redhat.com>