Implement Linux sys_string_conversions using the system APIs.

Implement Linux sys_string_conversions using the system APIs.

BUG=14826

[Dean McNamee, 2009-06-22 14:28:17]

Gnarly unix-hacker code.  I bet Adam will find a bug.

Wow, this APIs are awful!

[Evan Martin, 2009-06-22 15:22:56]

In general, I think this needs a commetn somewhere about what this special case
you're worried about is.

http://codereview.chromium.org/140062/diff/1/2
File base/sys_string_conversions_linux.cc (right):

http://codereview.chromium.org/140062/diff/1/2#newcode36
Line 36: // Use a temp buf since a output of NULL does not do what we want.
s/a/an/
This comment doesn't really explain anything (where does "output of NULL" come
from?)

http://codereview.chromium.org/140062/diff/1/2#newcode38
Line 38: // We don't want wcrtomb to do it's funkiness for embedded NULLs.
s/it's/its/

See above request for a descriptive comment.

http://codereview.chromium.org/140062/diff/1/2#newcode49
Line 49: num_out_chars += res;
Is it correct to do this if res is negative but not -1?
I think more common for this kind of API is

if (res < 0) {
  // error handling, bailing out
}

if (res == 0) {
  special case
} else {
  normal case
}

[Dean McNamee, 2009-06-22 15:26:02]

http://codereview.chromium.org/140062/diff/1/2
File base/sys_string_conversions_linux.cc (right):

http://codereview.chromium.org/140062/diff/1/2#newcode36
Line 36: // Use a temp buf since a output of NULL does not do what we want.
On 2009/06/22 15:22:56, Evan Martin wrote:
> s/a/an/
> This comment doesn't really explain anything (where does "output of NULL" come
> from?)

Ok, I'll expand on it.  Basically:

       A third case is when s is NULL.  In this case wc is ignored, and the
function effectively returns wcrtomb(buf,L’\0’,ps) where buf is
       an internal anonymous buffer.

Which is not the behavior we want at all.  So we need to do the write to a
temporary buffer and just ignore the output, so we can the size value correct.

http://codereview.chromium.org/140062/diff/1/2#newcode49
Line 49: num_out_chars += res;
On 2009/06/22 15:22:56, Evan Martin wrote:
> Is it correct to do this if res is negative but not -1?
> I think more common for this kind of API is
> 
> if (res < 0) {
>   // error handling, bailing out
> }

res is a size_t (unsigned).

> 
> if (res == 0) {
>   special case
> } else {
>   normal case
> }

[agl, 2009-06-22 18:21:49]

LGTM