[Qemu-devel] [PATCH] qom: avoid reporting errors for NULL error object

[Qemu-devel] [PATCH] qom: avoid reporting errors for NULL error object

[Daniel P. Berrangé]

When debugging QEMU it is often useful to put a breakpoint on the
error_setg_internal method impl.

Unfortunately the object_property_add / object_class_property_add
methods call object_property_find / object_class_property_find methods
to check if a property exists already before adding the new property.

As a result there are a huge number of calls to error_setg_internal
on startup of most QEMU commands, making it very painful to set a
breakpoint on this method.

This puts a minor optimization on the code so that we avoid calling
error_setg() when errp is NULL. Functionally there's no difference
since error_setg() is a no-op when errp is NULL, but this lets us
use breakpoints in GDB in a practical way.

Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
---
 qom/object.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/qom/object.c b/qom/object.c
index 547dcf97c3..ddd5e7a30e 100644
--- a/qom/object.c
+++ b/qom/object.c
@@ -1087,7 +1087,12 @@ ObjectProperty *object_property_find(Object *obj, const char *name,
         return prop;
     }
 
-    error_setg(errp, "Property '.%s' not found", name);
+    /* Optimized to avoid calling error_setg if errp == NULL
+     * otherwise every property add call hits error_setg
+     * making it impratical to set breakpoints in GDB */
+    if (errp) {
+        error_setg(errp, "Property '.%s' not found", name);
+    }
     return NULL;
 }
 
@@ -1133,7 +1138,10 @@ ObjectProperty *object_class_property_find(ObjectClass *klass, const char *name,
     }
 
     prop = g_hash_table_lookup(klass->properties, name);
-    if (!prop) {
+    /* Optimized to avoid calling error_setg if errp == NULL
+     * otherwise every property add call hits error_setg
+     * making it impratical to set breakpoints in GDB */
+    if (!prop && errp) {
         error_setg(errp, "Property '.%s' not found", name);
     }
     return prop;
-- 
2.19.1

Re: [Qemu-devel] [PATCH] qom: avoid reporting errors for NULL error object

[Marc-André Lureau]

Hi

On Mon, Nov 19, 2018 at 5:59 PM Daniel P. Berrangé <berrange@redhat.com> wrote:
>
> When debugging QEMU it is often useful to put a breakpoint on the
> error_setg_internal method impl.
>
> Unfortunately the object_property_add / object_class_property_add
> methods call object_property_find / object_class_property_find methods
> to check if a property exists already before adding the new property.
>
> As a result there are a huge number of calls to error_setg_internal
> on startup of most QEMU commands, making it very painful to set a
> breakpoint on this method.
>
> This puts a minor optimization on the code so that we avoid calling
> error_setg() when errp is NULL. Functionally there's no difference
> since error_setg() is a no-op when errp is NULL, but this lets us
> use breakpoints in GDB in a practical way.
>
> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>

Or maybe change the error_setg* macros? After all, if errp is NULL,
it's an error qemu can handle.

> ---
>  qom/object.c | 12 ++++++++++--
>  1 file changed, 10 insertions(+), 2 deletions(-)
>
> diff --git a/qom/object.c b/qom/object.c
> index 547dcf97c3..ddd5e7a30e 100644
> --- a/qom/object.c
> +++ b/qom/object.c
> @@ -1087,7 +1087,12 @@ ObjectProperty *object_property_find(Object *obj, const char *name,
>          return prop;
>      }
>
> -    error_setg(errp, "Property '.%s' not found", name);
> +    /* Optimized to avoid calling error_setg if errp == NULL
> +     * otherwise every property add call hits error_setg
> +     * making it impratical to set breakpoints in GDB */
> +    if (errp) {
> +        error_setg(errp, "Property '.%s' not found", name);
> +    }
>      return NULL;
>  }
>
> @@ -1133,7 +1138,10 @@ ObjectProperty *object_class_property_find(ObjectClass *klass, const char *name,
>      }
>
>      prop = g_hash_table_lookup(klass->properties, name);
> -    if (!prop) {
> +    /* Optimized to avoid calling error_setg if errp == NULL
> +     * otherwise every property add call hits error_setg
> +     * making it impratical to set breakpoints in GDB */
> +    if (!prop && errp) {
>          error_setg(errp, "Property '.%s' not found", name);
>      }
>      return prop;
> --
> 2.19.1
>
>


-- 
Marc-André Lureau

Re: [Qemu-devel] [PATCH] qom: avoid reporting errors for NULL error object

[Daniel P. Berrangé]

On Tue, Nov 20, 2018 at 09:45:23AM +0400, Marc-André Lureau wrote:
> Hi
> 
> On Mon, Nov 19, 2018 at 5:59 PM Daniel P. Berrangé <berrange@redhat.com> wrote:
> >
> > When debugging QEMU it is often useful to put a breakpoint on the
> > error_setg_internal method impl.
> >
> > Unfortunately the object_property_add / object_class_property_add
> > methods call object_property_find / object_class_property_find methods
> > to check if a property exists already before adding the new property.
> >
> > As a result there are a huge number of calls to error_setg_internal
> > on startup of most QEMU commands, making it very painful to set a
> > breakpoint on this method.
> >
> > This puts a minor optimization on the code so that we avoid calling
> > error_setg() when errp is NULL. Functionally there's no difference
> > since error_setg() is a no-op when errp is NULL, but this lets us
> > use breakpoints in GDB in a practical way.
> >
> > Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
> 
> Or maybe change the error_setg* macros? After all, if errp is NULL,
> it's an error qemu can handle.

That would mean that all callers would get short circuited which
I don't think it is desirable in general. 

This QOM usage is special in that the calling code is using the
error scenario as a means to trigger different code flow. IOW
it doesn't consider it an error scenario at all.

> > ---
> >  qom/object.c | 12 ++++++++++--
> >  1 file changed, 10 insertions(+), 2 deletions(-)
> >
> > diff --git a/qom/object.c b/qom/object.c
> > index 547dcf97c3..ddd5e7a30e 100644
> > --- a/qom/object.c
> > +++ b/qom/object.c
> > @@ -1087,7 +1087,12 @@ ObjectProperty *object_property_find(Object *obj, const char *name,
> >          return prop;
> >      }
> >
> > -    error_setg(errp, "Property '.%s' not found", name);
> > +    /* Optimized to avoid calling error_setg if errp == NULL
> > +     * otherwise every property add call hits error_setg
> > +     * making it impratical to set breakpoints in GDB */
> > +    if (errp) {
> > +        error_setg(errp, "Property '.%s' not found", name);
> > +    }
> >      return NULL;
> >  }
> >
> > @@ -1133,7 +1138,10 @@ ObjectProperty *object_class_property_find(ObjectClass *klass, const char *name,
> >      }
> >
> >      prop = g_hash_table_lookup(klass->properties, name);
> > -    if (!prop) {
> > +    /* Optimized to avoid calling error_setg if errp == NULL
> > +     * otherwise every property add call hits error_setg
> > +     * making it impratical to set breakpoints in GDB */
> > +    if (!prop && errp) {
> >          error_setg(errp, "Property '.%s' not found", name);
> >      }
> >      return prop;
> > --
> > 2.19.1
> >
> >
> 
> 
> -- 
> Marc-André Lureau

Regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|

Re: [Qemu-devel] [PATCH] qom: avoid reporting errors for NULL error object

[Markus Armbruster]

Daniel P. Berrangé <berrange@redhat.com> writes:

> When debugging QEMU it is often useful to put a breakpoint on the
> error_setg_internal method impl.
>
> Unfortunately the object_property_add / object_class_property_add
> methods call object_property_find / object_class_property_find methods
> to check if a property exists already before adding the new property.
>
> As a result there are a huge number of calls to error_setg_internal
> on startup of most QEMU commands, making it very painful to set a
> breakpoint on this method.
>
> This puts a minor optimization on the code so that we avoid calling
> error_setg() when errp is NULL. Functionally there's no difference
> since error_setg() is a no-op when errp is NULL, but this lets us
> use breakpoints in GDB in a practical way.
>
> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
> ---
>  qom/object.c | 12 ++++++++++--
>  1 file changed, 10 insertions(+), 2 deletions(-)
>
> diff --git a/qom/object.c b/qom/object.c
> index 547dcf97c3..ddd5e7a30e 100644
> --- a/qom/object.c
> +++ b/qom/object.c
> @@ -1087,7 +1087,12 @@ ObjectProperty *object_property_find(Object *obj, const char *name,
>          return prop;
>      }
>  
> -    error_setg(errp, "Property '.%s' not found", name);
> +    /* Optimized to avoid calling error_setg if errp == NULL
> +     * otherwise every property add call hits error_setg
> +     * making it impratical to set breakpoints in GDB */
> +    if (errp) {
> +        error_setg(errp, "Property '.%s' not found", name);
> +    }
>      return NULL;
>  }
>  

In my opinion, this function's design is awkward.  Stress on *opinion*.

On success, it returns a (non-null) pointer.

On failure, it sets an error and returns null.  Note that it has just
one failure mode: "Property '.%s' not found".  Setting an error is just
a convenience for those callers that want to propagate exactly this
error to their callers.

I count 30 callers.  Only six pass a non-NULL argument to @errp.

I'd rather have a pair of functions similar to how Python has both
.get() and .__getitem__(): the former doesn't fail, but returns None
instead, and the latter does fail, throwing KeyError.  In QEMU, we can't
throw, so we set an error.  Here's the obvious code:

    ObjectProperty *object_property_find(Object *obj, const char *name)
    {
        ObjectProperty *prop;
        ObjectClass *klass = object_get_class(obj);

        prop = object_class_property_find(klass, name, NULL);
        if (prop) {
            return prop;
        }

        prop = g_hash_table_lookup(obj->properties, name);
        if (prop) {
            return prop;
        }

        return NULL;
    }

    ObjectProperty *object_property_find_err(Object *obj, const char *name,
                                             Error **errp)
    {
        ObjectProperty *prop = object_property_find(obj, name);

        if (!prop) {
            error_setg(errp, "Property '.%s' not found", name);
        }
        return prop;
    }

> @@ -1133,7 +1138,10 @@ ObjectProperty *object_class_property_find(ObjectClass *klass, const char *name,
>      }
>  
>      prop = g_hash_table_lookup(klass->properties, name);
> -    if (!prop) {
> +    /* Optimized to avoid calling error_setg if errp == NULL
> +     * otherwise every property add call hits error_setg
> +     * making it impratical to set breakpoints in GDB */
> +    if (!prop && errp) {
>          error_setg(errp, "Property '.%s' not found", name);
>      }
>      return prop;

Likewise, just more so: callers passing non-NULL do not exist.

Re: [Qemu-devel] [PATCH] qom: avoid reporting errors for NULL error object

[Daniel P. Berrangé]

On Tue, Nov 20, 2018 at 07:16:31PM +0100, Markus Armbruster wrote:
> Daniel P. Berrangé <berrange@redhat.com> writes:
> 
> > When debugging QEMU it is often useful to put a breakpoint on the
> > error_setg_internal method impl.
> >
> > Unfortunately the object_property_add / object_class_property_add
> > methods call object_property_find / object_class_property_find methods
> > to check if a property exists already before adding the new property.
> >
> > As a result there are a huge number of calls to error_setg_internal
> > on startup of most QEMU commands, making it very painful to set a
> > breakpoint on this method.
> >
> > This puts a minor optimization on the code so that we avoid calling
> > error_setg() when errp is NULL. Functionally there's no difference
> > since error_setg() is a no-op when errp is NULL, but this lets us
> > use breakpoints in GDB in a practical way.
> >
> > Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
> > ---
> >  qom/object.c | 12 ++++++++++--
> >  1 file changed, 10 insertions(+), 2 deletions(-)
> >
> > diff --git a/qom/object.c b/qom/object.c
> > index 547dcf97c3..ddd5e7a30e 100644
> > --- a/qom/object.c
> > +++ b/qom/object.c
> > @@ -1087,7 +1087,12 @@ ObjectProperty *object_property_find(Object *obj, const char *name,
> >          return prop;
> >      }
> >  
> > -    error_setg(errp, "Property '.%s' not found", name);
> > +    /* Optimized to avoid calling error_setg if errp == NULL
> > +     * otherwise every property add call hits error_setg
> > +     * making it impratical to set breakpoints in GDB */
> > +    if (errp) {
> > +        error_setg(errp, "Property '.%s' not found", name);
> > +    }
> >      return NULL;
> >  }
> >  
> 
> In my opinion, this function's design is awkward.  Stress on *opinion*.
> 
> On success, it returns a (non-null) pointer.
> 
> On failure, it sets an error and returns null.  Note that it has just
> one failure mode: "Property '.%s' not found".  Setting an error is just
> a convenience for those callers that want to propagate exactly this
> error to their callers.
> 
> I count 30 callers.  Only six pass a non-NULL argument to @errp.
> 
> I'd rather have a pair of functions similar to how Python has both
> .get() and .__getitem__(): the former doesn't fail, but returns None
> instead, and the latter does fail, throwing KeyError.  In QEMU, we can't
> throw, so we set an error.  Here's the obvious code:
> 
>     ObjectProperty *object_property_find(Object *obj, const char *name)
>     {
>         ObjectProperty *prop;
>         ObjectClass *klass = object_get_class(obj);
> 
>         prop = object_class_property_find(klass, name, NULL);
>         if (prop) {
>             return prop;
>         }
> 
>         prop = g_hash_table_lookup(obj->properties, name);
>         if (prop) {
>             return prop;
>         }
> 
>         return NULL;
>     }
> 
>     ObjectProperty *object_property_find_err(Object *obj, const char *name,
>                                              Error **errp)
>     {
>         ObjectProperty *prop = object_property_find(obj, name);
> 
>         if (!prop) {
>             error_setg(errp, "Property '.%s' not found", name);
>         }
>         return prop;
>     }

Yes, I'd be happy with this approach as I think it is more useful in
general


Regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|