Re: [PATCH v2 3/8] qnum: QNumValue type for QNum value literals

[Eduardo Habkost]

On Thu, Nov 19, 2020 at 11:24:52AM +0100, Markus Armbruster wrote:
> Marc-André Lureau <marcandre.lureau@gmail.com> writes:
> 
> > On Tue, Nov 17, 2020 at 6:42 PM Eduardo Habkost <ehabkost@redhat.com> wrote:
> >
> >> On Tue, Nov 17, 2020 at 12:37:56PM +0400, Marc-André Lureau wrote:
> >> > On Tue, Nov 17, 2020 at 2:43 AM Eduardo Habkost <ehabkost@redhat.com>
> >> wrote:
> >> >
> >> > > Provide a separate QNumValue type that can be used for QNum value
> >> > > literals without the referencing counting and memory allocation
> >> > > features provided by QObject.
> >> > >
> >> > > Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
> >> > > ---
> >> > > Changes v1 -> v2:
> >> > > * Fix "make check" failure, by updating check-qnum unit test to
> >> > >   use the new struct fields
> >> > > ---
> >> > >  include/qapi/qmp/qnum.h | 40 +++++++++++++++++++--
> >> > >  qobject/qnum.c          | 78 ++++++++++++++++++++---------------------
> >> > >  tests/check-qnum.c      | 14 ++++----
> >> > >  3 files changed, 84 insertions(+), 48 deletions(-)
> >> > >
> >> > > diff --git a/include/qapi/qmp/qnum.h b/include/qapi/qmp/qnum.h
> >> > > index 55c27b1c24..62fbdfda68 100644
> >> > > --- a/include/qapi/qmp/qnum.h
> >> > > +++ b/include/qapi/qmp/qnum.h
> >> > > @@ -46,20 +46,56 @@ typedef enum {
> >> > >   * in range: qnum_get_try_int() / qnum_get_try_uint() check range and
> >> > >   * convert under the hood.
> >> > >   */
> >> > > -struct QNum {
> >> > > -    struct QObjectBase_ base;
> >> > > +
> >> > > +/**
> >> > > + * struct QNumValue: the value of a QNum
> >> > > + *
> >> > > + * QNumValue literals can be constructed using the `QNUM_VAL_INT`,
> >> > > + * `QNUM_VAL_UINT`, and `QNUM_VAL_DOUBLE` macros.
> >> > > + */
> >> > > +typedef struct QNumValue {
> >> > > +    /* private: */
> 
> Do we care?

Are you asking if we want to make them private, or if we want to
document them as private (assuming they are).

The answer to the latter is yes ("private:" is an indication to
kernel-doc).  The answer to the former seems to be "no", based on
your other comments.

Or maybe `kind` should be public and `u` should be private?

> 
> >> > >      QNumKind kind;
> >> > >      union {
> >> > >          int64_t i64;
> >> > >          uint64_t u64;
> >> > >          double dbl;
> >> > >      } u;
> >> > > +} QNumValue;
> >> > > +
> >> > > +#define QNUM_VAL_INT(value) \
> >> > > +    { .kind = QNUM_I64, .u.i64 = value }
> >> > > +#define QNUM_VAL_UINT(value) \
> >> > > +    { .kind = QNUM_U64, .u.u64 = value }
> >> > > +#define QNUM_VAL_DOUBLE(value) \
> >> > > +    { .kind = QNUM_DOUBLE, .u.dbl = value }
> >> > > +
> >> > > +struct QNum {
> >> > > +    struct QObjectBase_ base;
> >> > > +    QNumValue value;
> >> > >  };
> >> > >
> >> > > +/**
> >> > > + * qnum_from_int(): Create a new QNum from a QNumValue
> >> > >
> >> >
> >> > Copy-pasta qnum_from_int() -> qnum_from_value()
> >>
> >> Oops!  I will fix it in v3, or submit a fixup patch if that's the
> >> only needed change.
> >>
> >> >
> >> > I wonder if there is a check for that kind of mistake, as it may be
> >> common.
> >>
> >> It looks like kernel-doc ignores what's before the colon in the
> >> summary line.  It probably shouldn't.
> >>
> >> >
> >> > + * @value: QNumValue
> >> > > + *
> >> > > + * Return strong reference.
> >> > > + */
> >> > > +QNum *qnum_from_value(QNumValue value);
> >> > >
> >> > +
> >> > >  QNum *qnum_from_int(int64_t value);
> >> > >  QNum *qnum_from_uint(uint64_t value);
> >> > >  QNum *qnum_from_double(double value);
> >> > >
> >> > > +/**
> >> > > + * qnum_get_value(): Get QNumValue from QNum
> >> > > + * @qn: QNum object
> >> > > + */
> >> > > +static inline const QNumValue *qnum_get_value(const QNum *qn)
> >> > > +{
> >> > > +    return &qn->value;
> >> > > +}
> >> > > +
> >> > >
> >> >
> >> > Nothing uses that function in this patch. Should be put into use.
> >>
> >> It is used in patch 5/8.  Why do you think it's necessary to use
> >> it in internal code too?
> >>
> >
> > Not necessarily, just want to make sure we don't introduce dead code. Fine
> > it's used later, perhaps worth noting in the commit message.
> 
> Yes, use later in the same series suffices.
> 
> The whole patch makes sense only because later patches put QNumValue to
> use.  The commit message could perhaps be more explicit about that.  No
> need to be explicit about every new function, though.
> 
> That said, I wouldn't bother with a getter function.  Yes, we have
> similar trivial getters for other QFoo.  I don't care for them, either.
> The actual structure of these data types is trivial, and not worth
> hiding.
> 
> Also, I'm wary of inline functions in headers.  *Especially* when they
> require additional #include (which this one doesn't).  I accept them
> when they provide a worthwhile performance improvement.  Guesses don't
> count as evidence :)
> 
> >>
> >> >
> >> >  bool qnum_get_try_int(const QNum *qn, int64_t *val);
> >> > >  int64_t qnum_get_int(const QNum *qn);
> >> > >
> >> > > diff --git a/qobject/qnum.c b/qobject/qnum.c
> >> > > index 69fd9a82d9..f80d4efd76 100644
> >> > > --- a/qobject/qnum.c
> >> > > +++ b/qobject/qnum.c
> >> > > @@ -15,6 +15,15 @@
> >> > >  #include "qemu/osdep.h"
> >> > >  #include "qapi/qmp/qnum.h"
> >> > >
> >> > > +QNum *qnum_from_value(QNumValue value)
> >> > >
> >> >
> >> > I wonder if it shouldn't be made "inline" in header too (if that can help
> >> > avoid argument copy).
> >>
> >> I'm unsure.  I would make it inline (in a separate patch) if
> >> there's evidence it's worth it.  I expect the g_new() call to be
> >> more expensive than any argument copying, though.
> >
> > ok
> 
> I'm with Eduardo here.
> 
> >> >
> >> > +{
> >> > > +    QNum *qn = g_new(QNum, 1);
> >> > > +
> >> > > +    qobject_init(QOBJECT(qn), QTYPE_QNUM);
> >> > > +    qn->value = value;
> >> > > +    return qn;
> >> > > +}
> >> > > +
> >> > >  /**
> >> > >   * qnum_from_int(): Create a new QNum from an int64_t
> >> > >   * @value: int64_t value
> >> > > @@ -23,13 +32,7 @@
> >> > >   */
> >> > >  QNum *qnum_from_int(int64_t value)
> >> > >  {
> >> > > -    QNum *qn = g_new(QNum, 1);
> >> > > -
> >> > > -    qobject_init(QOBJECT(qn), QTYPE_QNUM);
> >> > > -    qn->kind = QNUM_I64;
> >> > > -    qn->u.i64 = value;
> >> > > -
> >> > > -    return qn;
> >> > > +    return qnum_from_value((QNumValue) QNUM_VAL_INT(value));
> 
> No space between between (type) and its operand, please.
> 
> Could we lift the cast into the macro somehow?

I think we can.  I had thought the cast in the macro would break
usage as static variable initializers.  I was wrong.

> 
> >> > >  }
> >> > >
> >> > >  /**
> >> > > @@ -40,13 +43,7 @@ QNum *qnum_from_int(int64_t value)
> >> > >   */
> >> > >  QNum *qnum_from_uint(uint64_t value)
> >> > >  {
> >> > > -    QNum *qn = g_new(QNum, 1);
> >> > > -
> >> > > -    qobject_init(QOBJECT(qn), QTYPE_QNUM);
> >> > > -    qn->kind = QNUM_U64;
> >> > > -    qn->u.u64 = value;
> >> > > -
> >> > > -    return qn;
> >> > > +    return qnum_from_value((QNumValue) QNUM_VAL_UINT(value));
> >> > >  }
> >> > >
> >> > >  /**
> >> > > @@ -57,13 +54,7 @@ QNum *qnum_from_uint(uint64_t value)
> >> > >   */
> >> > >  QNum *qnum_from_double(double value)
> >> > >  {
> >> > > -    QNum *qn = g_new(QNum, 1);
> >> > > -
> >> > > -    qobject_init(QOBJECT(qn), QTYPE_QNUM);
> >> > > -    qn->kind = QNUM_DOUBLE;
> >> > > -    qn->u.dbl = value;
> >> > > -
> >> > > -    return qn;
> >> > > +    return qnum_from_value((QNumValue) QNUM_VAL_DOUBLE(value));
> >> > >  }
> >> > >
> >> > >  /**
> >> > > @@ -75,15 +66,17 @@ QNum *qnum_from_double(double value)
> >> > >   */
> >> > >  bool qnum_get_try_int(const QNum *qn, int64_t *val)
> >> > >  {
> >> > > -    switch (qn->kind) {
> >> > > +    const QNumValue *qv = &qn->value;
> >> > > +
> >> > > +    switch (qv->kind) {
> >> > >      case QNUM_I64:
> >> > > -        *val = qn->u.i64;
> >> > > +        *val = qv->u.i64;
> >> > >          return true;
> >> > >      case QNUM_U64:
> >> > > -        if (qn->u.u64 > INT64_MAX) {
> >> > > +        if (qv->u.u64 > INT64_MAX) {
> >> > >              return false;
> >> > >          }
> >> > > -        *val = qn->u.u64;
> >> > > +        *val = qv->u.u64;
> >> > >          return true;
> >> > >      case QNUM_DOUBLE:
> >> > >          return false;
> 
> Here you add a new variable to shorten "qn->value." to "qv->".  In
> tests/check-qnum.c you don't.  I'm not sure the variable is worthwhile.

The variable is worthwhile if it becomes a parameter to
qnum_value_get_try_int().

If we don't create a qnum_value_get_try_int() function, I still
think the variable makes the code easier to read, but I'm not too
attached to it.

> 
> >> > > @@ -116,15 +109,17 @@ int64_t qnum_get_int(const QNum *qn)
> >> > >   */
> >> > >  bool qnum_get_try_uint(const QNum *qn, uint64_t *val)
> >> > >  {
> >> > > -    switch (qn->kind) {
> >> > > +    const QNumValue *qv = &qn->value;
> >> > > +
> >> > > +    switch (qv->kind) {
> >> > >      case QNUM_I64:
> >> > > -        if (qn->u.i64 < 0) {
> >> > > +        if (qv->u.i64 < 0) {
> >> > >              return false;
> >> > >          }
> >> > > -        *val = qn->u.i64;
> >> > > +        *val = qv->u.i64;
> >> > >          return true;
> >> > >      case QNUM_U64:
> >> > > -        *val = qn->u.u64;
> >> > > +        *val = qv->u.u64;
> >> > >          return true;
> >> > >      case QNUM_DOUBLE:
> >> > >          return false;
> >> > > @@ -156,13 +151,15 @@ uint64_t qnum_get_uint(const QNum *qn)
> >> > >   */
> >> > >  double qnum_get_double(const QNum *qn)
> >> > >  {
> >> > > -    switch (qn->kind) {
> >> > > +    const QNumValue *qv = &qn->value;
> >> > > +
> >> > > +    switch (qv->kind) {
> >> > >      case QNUM_I64:
> >> > > -        return qn->u.i64;
> >> > > +        return qv->u.i64;
> >> > >      case QNUM_U64:
> >> > > -        return qn->u.u64;
> >> > > +        return qv->u.u64;
> >> > >      case QNUM_DOUBLE:
> >> > > -        return qn->u.dbl;
> >> > > +        return qv->u.dbl;
> >> > >      }
> >> > >
> >> > >      assert(0);
> >> > > @@ -171,14 +168,15 @@ double qnum_get_double(const QNum *qn)
> >> > >
> >> > >  char *qnum_to_string(QNum *qn)
> >> > >  {
> >> > > +    const QNumValue *qv = &qn->value;
> >> > >
> >> >
> >> > qnum_get_value() ?
> >>
> >> I prefer to not hide this behind a function call, in internal
> >> code.  But I don't mind changing it if you think it's important.
> 
> Me too.  Even in external code.

Understood.

> 
> > no, it's ok to me
> >
> >
> >>
> >> >
> >> >      char *buffer;
> >> > >      int len;
> >> > >
> >> > > -    switch (qn->kind) {
> >> > > +    switch (qv->kind) {
> >> > >      case QNUM_I64:
> >> > > -        return g_strdup_printf("%" PRId64, qn->u.i64);
> >> > > +        return g_strdup_printf("%" PRId64, qv->u.i64);
> >> > >      case QNUM_U64:
> >> > > -        return g_strdup_printf("%" PRIu64, qn->u.u64);
> >> > > +        return g_strdup_printf("%" PRIu64, qv->u.u64);
> >> > >      case QNUM_DOUBLE:
> >> > >          /* FIXME: snprintf() is locale dependent; but JSON requires
> >> > >           * numbers to be formatted as if in the C locale. Dependence
> >> > > @@ -189,7 +187,7 @@ char *qnum_to_string(QNum *qn)
> >> > >           * rounding errors; we should be using DBL_DECIMAL_DIG (17),
> >> > >           * and only rounding to a shorter number if the result would
> >> > >           * still produce the same floating point value.  */
> >> > > -        buffer = g_strdup_printf("%f" , qn->u.dbl);
> >> > > +        buffer = g_strdup_printf("%f" , qv->u.dbl);
> >> > >          len = strlen(buffer);
> >> > >          while (len > 0 && buffer[len - 1] == '0') {
> >> > >              len--;
> >> > > @@ -221,8 +219,10 @@ char *qnum_to_string(QNum *qn)
> >> > >   */
> >> > >  bool qnum_is_equal(const QObject *x, const QObject *y)
> >> > >  {
> >> > > -    QNum *num_x = qobject_to(QNum, x);
> >> > > -    QNum *num_y = qobject_to(QNum, y);
> >> > > +    const QNum *qnum_x = qobject_to(QNum, x);
> >> > > +    const QNum *qnum_y = qobject_to(QNum, y);
> 
> A bit of drive-by constification.  Okay.

In my defense, I will argue that those are four new variables.
the variables `QNum *num_x, *num_y` don't exist anymore.

Anyway, I will probably keep the `QNumValue *num_x, *num_y`
variables untouched in the next version, and choose another name
for the new QNumValue variables (which will become
qnum_value_is_equal() parameters in the next patch).

> 
> >> > > +    const QNumValue *num_x = &qnum_x->value;
> >> > > +    const QNumValue *num_y = &qnum_y->value;
> >> > >
> >> > >      switch (num_x->kind) {
> >> > >      case QNUM_I64:
> >> > > diff --git a/tests/check-qnum.c b/tests/check-qnum.c
> >> > > index 4105015872..9499b0d845 100644
> >> > > --- a/tests/check-qnum.c
> >> > > +++ b/tests/check-qnum.c
> >> > > @@ -30,8 +30,8 @@ static void qnum_from_int_test(void)
> >> > >
> >> > >      qn = qnum_from_int(value);
> >> > >      g_assert(qn != NULL);
> >> > > -    g_assert_cmpint(qn->kind, ==, QNUM_I64);
> >> > > -    g_assert_cmpint(qn->u.i64, ==, value);
> >> > > +    g_assert_cmpint(qn->value.kind, ==, QNUM_I64);
> >> > > +    g_assert_cmpint(qn->value.u.i64, ==, value);
> >> > >      g_assert_cmpint(qn->base.refcnt, ==, 1);
> >> > >      g_assert_cmpint(qobject_type(QOBJECT(qn)), ==, QTYPE_QNUM);
> >> > >
> >> > > @@ -45,8 +45,8 @@ static void qnum_from_uint_test(void)
> >> > >
> >> > >      qn = qnum_from_uint(value);
> >> > >      g_assert(qn != NULL);
> >> > > -    g_assert_cmpint(qn->kind, ==, QNUM_U64);
> >> > > -    g_assert(qn->u.u64 == value);
> >> > > +    g_assert_cmpint(qn->value.kind, ==, QNUM_U64);
> >> > > +    g_assert(qn->value.u.u64 == value);
> >> > >      g_assert(qn->base.refcnt == 1);
> >> > >      g_assert(qobject_type(QOBJECT(qn)) == QTYPE_QNUM);
> >> > >
> >> > > @@ -60,8 +60,8 @@ static void qnum_from_double_test(void)
> >> > >
> >> > >      qn = qnum_from_double(value);
> >> > >      g_assert(qn != NULL);
> >> > > -    g_assert_cmpint(qn->kind, ==, QNUM_DOUBLE);
> >> > > -    g_assert_cmpfloat(qn->u.dbl, ==, value);
> >> > > +    g_assert_cmpint(qn->value.kind, ==, QNUM_DOUBLE);
> >> > > +    g_assert_cmpfloat(qn->value.u.dbl, ==, value);
> >> > >      g_assert_cmpint(qn->base.refcnt, ==, 1);
> >> > >      g_assert_cmpint(qobject_type(QOBJECT(qn)), ==, QTYPE_QNUM);
> >> > >
> >> > > @@ -74,7 +74,7 @@ static void qnum_from_int64_test(void)
> >> > >      const int64_t value = 0x1234567890abcdefLL;
> >> > >
> >> > >      qn = qnum_from_int(value);
> >> > > -    g_assert_cmpint((int64_t) qn->u.i64, ==, value);
> >> > > +    g_assert_cmpint((int64_t) qn->value.u.i64, ==, value);
> >> > >
> >> > >      qobject_unref(qn);
> >> > >  }
> >> > > --
> >> > > 2.28.0
> >> > >
> >> > >
> >> > >
> >> > lgtm otherwise
> >>
> >> Thanks!
> >>
> >> --
> >> Eduardo
> >>
> >>

-- 
Eduardo