[Qemu-devel] Re: [PATCH 03/11] add a generic Error object

[Anthony Liguori <aliguori@us.ibm.com>]

On 03/14/2011 02:18 PM, Luiz Capitulino wrote:
> On Fri, 11 Mar 2011 15:00:41 -0600
> Anthony Liguori<aliguori@us.ibm.com>  wrote:
>
>> The Error class is similar to QError (now deprecated) except that it supports
>> propagation.  This allows for higher quality error handling.  It's losely
>> modeled after glib style GErrors.
> I think Daniel asked this, but I can't remember your answer: why don't we
> use GError then?

Because GError just uses strings and doesn't store key/values.

> Also, I think this patch needs more description regarding how this is going
> to replace QError.

Once there is no more qerror usage (we need to converting remaining HMP 
commands to QMP), qerror goes away.  This is scoped for the 0.15 release.

>   I mean, we want to deprecate QError but it seems to me
> that you're going to maintain the error declaration format, like:
>
>   "{ 'class': 'CommandNotFound', 'data': { 'name': %s } }"
>
> And the current QError class list in qerror.h. Did I get it right?

No, it will be switched to something simpler.  The QERR JSON is just an 
implementation detail.

> More comments below.
>
>> Signed-off-by: Anthony Liguori<aliguori@us.ibm.com>
>>
>> diff --git a/Makefile.objs b/Makefile.objs
>> index 0ba02c7..da31530 100644
>> --- a/Makefile.objs
>> +++ b/Makefile.objs
>> @@ -15,6 +15,7 @@ oslib-obj-$(CONFIG_POSIX) += oslib-posix.o
>>
>>   block-obj-y = cutils.o cache-utils.o qemu-malloc.o qemu-option.o module.o
>>   block-obj-y += nbd.o block.o aio.o aes.o qemu-config.o
>> +block-obj-y += error.o
>>   block-obj-$(CONFIG_POSIX) += posix-aio-compat.o
>>   block-obj-$(CONFIG_LINUX_AIO) += linux-aio.o
> You also have to do this:
>
> -CHECK_PROG_DEPS = qemu-malloc.o $(oslib-obj-y) $(trace-obj-y)
> +CHECK_PROG_DEPS = qemu-malloc.o error.o qerror.o $(oslib-obj-y) $(trace-obj-y)
>
> Otherwise you break check build.

Ah, okay.  Maybe I'll convert those over to gtester while I'm there.

>> diff --git a/error.c b/error.c
>> new file mode 100644
>> index 0000000..5d84106
>> --- /dev/null
>> +++ b/error.c
>> @@ -0,0 +1,122 @@
>> +/*
>> + * QEMU Error Objects
>> + *
>> + * Copyright IBM, Corp. 2011
>> + *
>> + * Authors:
>> + *  Anthony Liguori<aliguori@us.ibm.com>
>> + *
>> + * This work is licensed under the terms of the GNU LGPL, version 2.  See
>> + * the COPYING.LIB file in the top-level directory.
>> + */
>> +#include "error.h"
>> +#include "error_int.h"
>> +#include "qemu-objects.h"
>> +#include "qerror.h"
>> +#include<assert.h>
>> +
>> +struct Error
>> +{
>> +    QDict *obj;
> 'obj' is a bit generic and sometimes it's used to denote QObjects in the
> code, I suggest 'error_dict' or something that communicates its purpose.

Sure.

>> +const char *error_get_pretty(Error *err)
>> +{
>> +    if (err->msg == NULL) {
>> +        QString *str;
>> +        str = qerror_format(err->fmt, err->obj);
>> +        err->msg = qemu_strdup(qstring_get_str(str));
> Four comments here:
>
>   1. This is missing a QDECREF(str);

Yes, I caught this a few days ago in my tree thanks to valgrind.

>   2. Storing 'msg' looks like an unnecessary optimization to me, why don't
>      we just re-create it when error_get_pretty() is called?

Because we return a 'const char *' here so by storing msg, we can tie 
the string's life cycle to the Error object.  That means callers don't 
have to worry about freeing it.

>   3. This function is not used by this series

Yeah, it's infrastructure that needs to be here for subsequent series.

>   4. I think it's a good idea to assert on Error == NULL, specially
>      because some functions accept it

Only functions that take a double pointer, but not a bad thing to do.

>> +bool error_is_type(Error *err, const char *fmt)
>> +{
>> +    char *ptr;
>> +    char *end;
>> +    char classname[1024];
>> +
>> +    ptr = strstr(fmt, "'class': '");
>> +    assert(ptr != NULL);
>> +    ptr += strlen("'class': '");
>> +
>> +    end = strchr(ptr, '\'');
>> +    assert(end != NULL);
>> +
>> +    memcpy(classname, ptr, (end - ptr));
>> +    classname[(end - ptr)] = 0;
>> +
>> +    return strcmp(classname, error_get_field(err, "class")) == 0;
>> +}
> Not used by this series. Except for obvious stuff, I prefer to only add
> code that's going to be used right away.

That means adding a ton of stuff all at once.  Splitting it up like this 
is pretty reasonable IMHO.   Think of Error as an interface and not just 
a code dependency.

>> +
>> +void error_propagate(Error **dst_err, Error *local_err)
>> +{
>> +    if (dst_err) {
>> +        *dst_err = local_err;
>> +    } else if (local_err) {
>> +        error_free(local_err);
>> +    }
>> +}
>> +
>> +QObject *error_get_qobject(Error *err)
>> +{
>> +    QINCREF(err->obj);
>> +    return QOBJECT(err->obj);
>> +}
>> +
>> +void error_set_qobject(Error **errp, QObject *obj)
>> +{
>> +    Error *err;
>> +    if (errp == NULL) {
>> +        return;
>> +    }
>> +    err = qemu_mallocz(sizeof(*err));
>> +    err->obj = qobject_to_qdict(obj);
>> +    qobject_incref(obj);
>> +
>> +    *errp = err;
>> +}
> This is not documented. Also, I prefer the documentation&  code next to each
> other in the same file.

These are internal functions to QEMU and are documented as such.

>> diff --git a/error_int.h b/error_int.h
>> new file mode 100644
>> index 0000000..eaba65e
>> --- /dev/null
>> +++ b/error_int.h
>> @@ -0,0 +1,27 @@
>> +/*
>> + * QEMU Error Objects
>> + *
>> + * Copyright IBM, Corp. 2011
>> + *
>> + * Authors:
>> + *  Anthony Liguori<aliguori@us.ibm.com>
>> + *
>> + * This work is licensed under the terms of the GNU LGPL, version 2.  See
>> + * the COPYING.LIB file in the top-level directory.
>> + */
>> +#ifndef QEMU_ERROR_INT_H
>> +#define QEMU_ERROR_INT_H
>> +
>> +#include "qemu-common.h"
>> +#include "qobject.h"
>> +#include "error.h"
>> +
>> +/**
>> + * Internal QEMU functions for working with Error.
>> + *
>> + * These are used to convert QErrors to Errors
>> + */
>> +QObject *error_get_qobject(Error *err);
>> +void error_set_qobject(Error **errp, QObject *obj);
>> +
>> +#endif

Right here.

Regards,

Anthony Liguori