Re: Remove all do_inplace* logic from C++

[John W. Eaton]

On 05/15/2013 11:36 AM, Jordi Gutiérrez Hermoso wrote:
> Now that I've actually gotten bitten by this, I want to fix it:
>
>      https://savannah.gnu.org/bugs/?38466
>
> But now I remember why I did it this way in the first place. An
> erstwhile Octave developer had decided to dubiously optimise Octave's
> in-place operators (i.e. "a OP= b" instead of "a = a OP b") to
> actually also be written in-place in C++. However, after doing some
> testing by compiling with g++ -O2 and decompiling the results, it
> looks like gcc is smart enough to compile both "a OP= b" and "a = b OP
> a" into the same damn thing. My test was simplistic, but I don't think
> there is any actual benefit to using in-place operators in C++, and
> they're needlessly cluttering our code:
>
>      http://codepad.org/YtkJflcV
>
> In particular, they're making it more difficult for me to patch
> #38466. Note that Octave's in-place operators are meant to be pure
> syntactic sugar, and they are supposed to behave exactly like their
> expanded versions.
>
> Any objection to completely removing all of the inplace logic from C++?

I thought Jaroslav would have posted some timings related to this 
change, but as you and I discussed on #octave, I can't seem to find any. 
 OTOH, there is this bug report:

  https://savannah.gnu.org/bugs/?35400

So now I'm a bit confused because the way I read the assignment 
functions in ov.cc, A op= B is handled by first computing a value (call 
it V) for A (including any indexing operations), then computing V op B 
and assigning it to A (again with any indexing that is part of the 
expression A).  So at this point, I'm not sure where the in-place 
operation is recognized.  But apparently it is, because there is some 
difference in speed as pointed out by bug report 35400.

Also, since Octave is an array language, one thing to keep in mind is 
that an in-place operation can save the memory allocation of the 
temporary storage required to hold the RHS value that is the result of 
computing V op B.

Anyway, what about the bug report 38466 are you trying to fix?

jwe