Yes, this seems nice:

Foo.Builder builder;
builder.setBar(builder.getBar().toBuilder().setA(...).build());

But I guess I'm a little worried about its hidden cost if I use it a
lot. If message Bar was very large, then would converting it to a
builder, setting one value, and converting it back to a message again
incur a large cost? That is, is the time cost of setting one of a
builder's sub-message's values proportional to the total size of the
sub-message? That would be quite different than the cost of setting
one of a builder's values, which is constant.

If I know that a message of type Foo contains some Foo-specific
information, then I'd want to be able to write code that reads a
message of type Foo and gives me back that information. Like in your
example, I know that a Foo message has some "a" information and some
"b" information. I could write a method GetA(Foo) : int32 ... well,
Option<int32> anyway. (It might be null.) Then if I later decided to
move the field "a" from Bar to Foo I wouldn't have to change client
code of GetA(Foo) at all. It would be just as fast, too, because there
really isn't any difference between msg.getBar().getA() and msg.getA
(). I can do this right now, which is great.

But writing is a different story. Let's say I moved "a" from Bar to
Foo. I know that I can write to a builder of type Foo.Builder some "a"
information and some "b" information. I could write a method SetA
(Foo.Builder, int32) : unit that just invokes the setA method in
Foo.Builder. But then, if I decided to move "a" back to Bar, then my
implementation of SetA(Foo.Builder, int32) : unit has to change a lot.
I might have to convert Foo's bar from a message to a builder, set "a"
in it, and rebuild the updated Bar message. That doesn't change the
type signature of SetA but it might have a big effect on the cost of
calling the method.

If builders could contain builders, however, then I'd know for sure
that I could change where "a" lives without significantly affecting
the performance of methods like SetA!

And I do those kinds of changes to my protocol a lot--I'm always
rearranging how information relates to each other as my applications
evolve. Is it just me?

On Dec 16, 11:55 am, Jason Hsueh <jas...@google.com> wrote:
> You should be able to pass a builder to some method, and have the method
> modify some value of the builder just fine.
>
> Modifying a value in a sub-message is a bit more inconvenient, but still
> doable. If you have some type Bar in type Foo like:
> message Bar {
>   optional int32 a = 1;
>   optional int32 b = 2;
>
> }
>
> message Foo {
>   optional Bar bar = 1;
>
> }
>
> Foo.Builder builder;
> builder.setBar(builder.getBar().toBuilder().setA(...).build());
>
> Would that work for you?
>
>
>
> On Wed, Dec 16, 2009 at 9:24 AM, Mark <mjsc...@gmail.com> wrote:
> > I wish I could pass a builder object to a method and have the method
> > modify either a value of the builder or a value of a sub-message in
> > the builder!
>
> > I came across this thread, which described exactly the problem I have.
> > The Car/Engine example in the thread is perfectly illustrative of my
> > scenario.
>
> >http://groups.google.com/group/protobuf/browse_thread/thread/16997910...
>
> > Has anyone else been in this same situation? What have you done to
> > ameliorate the problem?
>
> > --
>
> > You received this message because you are subscribed to the Google Groups
> > "Protocol Buffers" group.
> > To post to this group, send email to proto...@googlegroups.com.
> > To unsubscribe from this group, send email to
> > protobuf+unsubscr...@googlegroups.com<protobuf%2bunsubscr...@googlegroups.c 
> > om>
> > .
> > For more options, visit this group at
> >http://groups.google.com/group/protobuf?hl=en.

--

You received this message because you are subscribed to the Google Groups 
"Protocol Buffers" group.
To post to this group, send email to proto...@googlegroups.com.
To unsubscribe from this group, send email to 
protobuf+unsubscr...@googlegroups.com.
For more options, visit this group at 
http://groups.google.com/group/protobuf?hl=en.


Reply via email to