Hi
On Thu, Aug 17, 2017 at 1:44 PM, Markus Armbruster <arm...@redhat.com> wrote:
> Marc-André Lureau <marcandre.lur...@redhat.com> writes:
>
>> The qapi schema has per-target definitions. Move qapi objects in the
>> per-target build, so they can be configured at compile time.
>
> Suggest something like:
>
> QAPI can't do target-specific conditionals (the symbols are
> poisoned), and the work-around is to pretend the target-specific
> stuff is target-independent, with stubs for the other targets.
> Makes the target-specifity invisible in introspection.
>
> To unpoison the symbols, we need to move the generated QAPI code to
> the per-target build.
>
>> Keep qapi-types.o qapi-visit.o in util-obj as they are necessary for
>> common code, but they will be overwritten during the target link.
>
> --verbose: how are they supposed to even compile in the
> target-independent build once we generate #if defined(TARGET_FOO) into
> them?
>
For common code, they will compile without TARGET_FOO.
>> Add
>> some stubs for block events, in code shared by tools & qemu.
>
> Sounds awkward.
I guess the alternative is to make the objects that depend on it per-target.
>
>> The following patch will configure the schema to conditionally remove
>> per-target disabled features.
>
> "The following patches", right?
yes, fixed
>
>> Signed-off-by: Marc-André Lureau <marcandre.lur...@redhat.com>
>> ---
>> stubs/qapi-event.c | 74
>> ++++++++++++++++++++++++++++++++++++++
>> tests/test-qobject-input-visitor.c | 1 -
>> Makefile.objs | 9 +----
>> Makefile.target | 4 +++
>> stubs/Makefile.objs | 1 +
>> trace/Makefile.objs | 2 +-
>> 6 files changed, 81 insertions(+), 10 deletions(-)
>> create mode 100644 stubs/qapi-event.c
>>
>> diff --git a/stubs/qapi-event.c b/stubs/qapi-event.c
>> new file mode 100644
>> index 0000000000..9415299f3a
>> --- /dev/null
>> +++ b/stubs/qapi-event.c
>> @@ -0,0 +1,74 @@
>> +#include "qemu/osdep.h"
>> +#include "qapi-event.h"
>> +
>> +void qapi_event_send_device_tray_moved(const char *device, const char *id,
>> + bool tray_open, Error **errp)
>> +{
>> +}
>> +
>> +void qapi_event_send_quorum_report_bad(QuorumOpType type, bool has_error,
>> + const char *error, const char
>> *node_name,
>> + int64_t sector_num,
>> + int64_t sectors_count, Error **errp)
>> +{
>> +}
>> +
>> +void qapi_event_send_quorum_failure(const char *reference, int64_t
>> sector_num,
>> + int64_t sectors_count, Error **errp)
>> +{
>> +}
>> +
>> +void qapi_event_send_block_job_cancelled(BlockJobType type, const char
>> *device,
>> + int64_t len, int64_t offset,
>> + int64_t speed, Error **errp)
>> +{
>> +}
>> +
>> +void qapi_event_send_block_job_completed(BlockJobType type, const char
>> *device,
>> + int64_t len, int64_t offset,
>> + int64_t speed, bool has_error,
>> + const char *error, Error **errp)
>> +{
>> +}
>> +
>> +void qapi_event_send_block_job_error(const char *device,
>> + IoOperationType operation,
>> + BlockErrorAction action, Error **errp)
>> +{
>> +}
>> +
>> +void qapi_event_send_block_job_ready(BlockJobType type, const char *device,
>> + int64_t len, int64_t offset, int64_t
>> speed,
>> + Error **errp)
>> +{
>> +}
>> +
>> +void qapi_event_send_block_io_error(const char *device, const char
>> *node_name,
>> + IoOperationType operation,
>> + BlockErrorAction action, bool
>> has_nospace,
>> + bool nospace, const char *reason,
>> + Error **errp)
>> +{
>> +}
>> +
>> +void qapi_event_send_block_image_corrupted(const char *device,
>> + bool has_node_name,
>> + const char *node_name,
>> + const char *msg, bool has_offset,
>> + int64_t offset, bool has_size,
>> + int64_t size, bool fatal,
>> + Error **errp)
>> +{
>> +}
>> +
>> +void qapi_event_send_block_write_threshold(const char *node_name,
>> + uint64_t amount_exceeded,
>> + uint64_t write_threshold,
>> + Error **errp)
>> +{
>> +}
>> +
>> +void qapi_event_send_device_deleted(bool has_device, const char *device,
>> + const char *path, Error **errp)
>> +{
>> +}
>
> Yup, awkward.
>
>> diff --git a/tests/test-qobject-input-visitor.c
>> b/tests/test-qobject-input-visitor.c
>> index 4da5d02c35..0a9352c5c1 100644
>> --- a/tests/test-qobject-input-visitor.c
>> +++ b/tests/test-qobject-input-visitor.c
>> @@ -1266,7 +1266,6 @@ static void
>> test_visitor_in_qmp_introspect(TestInputVisitorData *data,
>> const void *unused)
>> {
>> do_test_visitor_in_qmp_introspect(data, &test_qmp_schema_qlit);
>> - do_test_visitor_in_qmp_introspect(data, &qmp_schema_qlit);
>> }
>>
>> int main(int argc, char **argv)
>
> Either squash this change into PATCH 20, or mention it in the commit
> message.
done
>
> See also my review of PATCH 04.
>
>> diff --git a/Makefile.objs b/Makefile.objs
>> index 24a4ea08b8..2664720f9b 100644
>> --- a/Makefile.objs
>> +++ b/Makefile.objs
>> @@ -2,7 +2,7 @@
>> # Common libraries for tools and emulators
>> stub-obj-y = stubs/ crypto/
>> util-obj-y = util/ qobject/ qapi/
>> -util-obj-y += qmp-introspect.o qapi-types.o qapi-visit.o qapi-event.o
>> +util-obj-y += qapi-types.o qapi-visit.o
>>
>> chardev-obj-y = chardev/
>>
>> @@ -72,13 +72,6 @@ common-obj-y += chardev/
>> common-obj-$(CONFIG_SECCOMP) += qemu-seccomp.o
>>
>> common-obj-$(CONFIG_FDT) += device_tree.o
>> -
>> -######################################################################
>> -# qapi
>> -
>> -common-obj-y += qmp-marshal.o
>> -common-obj-y += qmp-introspect.o
>> -common-obj-y += qmp.o hmp.o
>> endif
>>
>> #######################################################################
>> diff --git a/Makefile.target b/Makefile.target
>> index 2baec9252f..a97dd056ad 100644
>> --- a/Makefile.target
>> +++ b/Makefile.target
>> @@ -154,6 +154,10 @@ endif
>>
>> GENERATED_FILES += hmp-commands.h hmp-commands-info.h
>>
>> +obj-y += qmp-introspect.o qapi-types.o qapi-visit.o qapi-event.o
>> +obj-y += qmp-marshal.o qmp-introspect.o
>> +obj-y += qmp.o hmp.o
>> +
>
> Moving so much code from "compile once" to "compile per target" is kind
> of sad. With the full series applied, I see
>
> $ wc qapi*c qmp*c
> 1528 3089 34576 qapi-event.c
> 5097 9126 107004 qapi-types.c
> 18848 44862 469514 qapi-visit.c
> 12407 32395 404704 qmp-introspect.c
> 6883 14997 182063 qmp-marshal.c
> 44763 104469 1197861 total
>
> Is there any way to split stuff so we recompile less? Note that this is
> a valid question even without your patches: changing one little thing in
> the QAPI schema commonly triggers a lengthy recompile. In large part
> because our undisciplined use of #include. But also because the
> generator's output is monolithic.
>
> I'm not expecting you to answer this question now, I just want to toss
> it out :)
>
Yes, I also think it's the next logical thing to do. We would benefit
from a modular schema, especially common/target split.
>> endif # CONFIG_SOFTMMU
>>
>> # Workaround for http://gcc.gnu.org/PR55489, see configure.
>> diff --git a/stubs/Makefile.objs b/stubs/Makefile.objs
>> index f5b47bfd74..1b2bef99c9 100644
>> --- a/stubs/Makefile.objs
>> +++ b/stubs/Makefile.objs
>> @@ -21,6 +21,7 @@ stub-obj-y += machine-init-done.o
>> stub-obj-y += migr-blocker.o
>> stub-obj-y += monitor.o
>> stub-obj-y += notify-event.o
>> +stub-obj-y += qapi-event.o
>> stub-obj-y += qtest.o
>> stub-obj-y += replay.o
>> stub-obj-y += runstate-check.o
>> diff --git a/trace/Makefile.objs b/trace/Makefile.objs
>> index afd571c3ec..6447729d60 100644
>> --- a/trace/Makefile.objs
>> +++ b/trace/Makefile.objs
>> @@ -56,4 +56,4 @@ util-obj-$(CONFIG_TRACE_SIMPLE) += simple.o
>> util-obj-$(CONFIG_TRACE_FTRACE) += ftrace.o
>> util-obj-y += control.o
>> target-obj-y += control-target.o
>> -util-obj-y += qmp.o
>> +target-obj-y += qmp.o
>
--
Marc-André Lureau
