Work around linux resolver problem where changes to resolv.conf go unnoticed in some distributions.

Work around linux resolver problem where changes to resolv.conf go unnoticed in some distributions.
BUG=11380
TEST=Start chrome with empty resolv.conf, fix resolv.conf; chrome should
be able to resolve sites.

[Evan Martin, 2009-05-27 17:30:43]

Darin says that this is the right fix.  I just have stylistic concerns.

http://codereview.chromium.org/113904/diff/3/1002
File net/base/host_resolver.cc (right):

http://codereview.chromium.org/113904/diff/3/1002#newcode12
Line 12: #include <resolv.h>
I think the netdb and socket should still be in a POSIX block, and then the
linux-specific include in its own block.

http://codereview.chromium.org/113904/diff/3/1002#newcode85
Line 85: int err = getaddrinfo(host.c_str(), port.c_str(), &hints, out);
I guess this line is shared by both branches?  So rather than the #else, just
put it above?

[Craig Schlenter, 2009-05-27 18:21:39]

On 2009/05/27 17:30:43, Evan Martin wrote:
> Darin says that this is the right fix.  I just have stylistic concerns.
> 
> http://codereview.chromium.org/113904/diff/3/1002
> File net/base/host_resolver.cc (right):
> 
> http://codereview.chromium.org/113904/diff/3/1002#newcode12
> Line 12: #include <resolv.h>
> I think the netdb and socket should still be in a POSIX block, and then the
> linux-specific include in its own block.

Done.
 
> http://codereview.chromium.org/113904/diff/3/1002#newcode85
> Line 85: int err = getaddrinfo(host.c_str(), port.c_str(), &hints, out);
> I guess this line is shared by both branches?  So rather than the #else, just
> put it above?

Yes ... much better. Thanks!

[Evan Martin, 2009-05-27 18:24:21]

LGTM, will commit

[darin (slow to review), 2009-05-27 19:23:11]

http://codereview.chromium.org/113904/diff/2001/3001
File net/base/host_resolver.cc (right):

http://codereview.chromium.org/113904/diff/2001/3001#newcode81
Line 81: if (err && !res_init())
please use res_ninit instead, which is threadsafe.

see the mozilla code here:
http://mxr.mozilla.org/seamonkey/source/netwerk/dns/src/nsHostResolver.cpp

[Evan Martin, 2009-05-27 19:31:45]

Upon more thought, we should probably follow Mozilla more closely (including the
timeout behavior).  I will file a follow-up bug.

[Evan Martin, 2009-05-27 19:35:34]

Craig: http://code.google.com/p/chromium/issues/detail?id=12740

[Craig Schlenter, 2009-05-27 19:49:18]

> http://codereview.chromium.org/113904/diff/2001/3001#newcode81
> Line 81: if (err && !res_init())
> please use res_ninit instead, which is threadsafe.

Ah, yes ... thanks for spotting that. I have commented in the bug Evan created
and will fix tomorrow after some more reading.

Thank you.