[Patch] libstdc++/25626: valarray vs non-POD

[Gabriel Dos Reis]

On Sun, 15 Jan 2006, Paolo Carlini wrote:

| Gabriel Dos Reis wrote:
| > I much prefer the use of __values().
| >
| Honestly, I don't, for various reasons, and, not hearing from you,
| already committed the last patch.

I had to go and deliver a talk at POPL and for the most part I had no
connection.  I tried hard to be away for the minimum amount of time
possible, given the circumstances.
I would not have committed a patch for sso_string, for instance, without
hearing from you, unless it causes obvious compilation problem.

| Among the reasons, we end up with both friendship of _Array and
| __values() in valarray (we cannot change the type of the _M_array member
| of valarray for ABI reasons, sorry for not clearly explaining that);

__values() does not change ABI.

| inconsistency among the various constructors about the use/not use of
| _Array.

the apparent consistency hides a deep obscurity in the design.

| Also, not secondary, I trust the last version much more because

you could have asked me, I would have tried to higher up your confidence in
the other version :-)

| uses completely for the real work existing facilities and calling
| conventions, in a way completely consistent with the rest of the
| library: add maybe a minimum of clean-up and someone coming from the
| outside would not even imagine we had the bug, at some point.

ChangeLog is for archeology :-)

| I would
| leave to you, much more knowledgeable in this area, further
| improvements... Only, as we already discussed, let's agree in advance
| about binary compatibility breakages...

-- Gaby