Re: [boost] Re: Re: Zero Unnecessary Temporary Objects: A Revolution

From: Gabriel Dos Reis (gdr_at_[hidden])
Date: 2002-10-13 12:51:13

Terje Slettebø <tslettebo_at_[hidden]> writes:

| >From: "Gabriel Dos Reis" <gdr_at_[hidden]>
| 
| >Terje Slettebø <tslettebo_at_[hidden]> writes:
| 
| >| Could someone point me to the part of the standard that allows for RVO
| and
| >| NRVO?
| 
| >12.8/15.
| 
| Thanks. :) It says:

[...]

| It appears from this that RVO and NRVO are treated in the same way - both
| are allowed to elide copying, even if the copy would have a side effect.

In theory, yes; in practice, no.  GCC, for example, has been able to
elide unnamed temporary looong before it implemented the "true" NRVO
(and that inflences part of its standard library implementation; I
didn't have the time to go through it and take advantages of NRVO;
maybe something to do in GCC-3.4)

| >From this, as mentioned in another posting, could this also hold for a value
| parameter (it's a local object, as well)? I.e. like the following:
| 
| inline Test operator+(Test t1,const Test &t2)
| {
|   t1+=t2;
| 
|   return t1; // Can NRVO be applied here?
| }

Here, an implementation may elide the copy-constructor call to
initialize t1.  While in some sense a parameter maybe be viewed as a
local variable, it is a local variable that is initialized in the
caller site, so it is not clear whether NVRO can be blatantly applied.
However, the same 12.8/15 maybe be applied if the implementation can
prove it to elide the second copy-constructor call.  But we're getting
lots of "if". :-)

-- Gaby