NOTREACHED should use DCHECK if DCHECK_ALWAYS_ON is set on cros.

base/logging.h

Index: base/logging.h
diff --git a/base/logging.h b/base/logging.h
index 54ff57970abe2b5a63507ce77e11f5ae698ef457..f6d2bd649c4145cc30d6771d6e13fee3da4d0921 100644
--- a/base/logging.h
+++ b/base/logging.h
@@ -701,7 +701,7 @@ const LogSeverity LOG_DCHECK = LOG_INFO;
 #define DCHECK_GT(val1, val2) DCHECK_OP(GT, > , val1, val2)
 #define DCHECK_IMPLIES(val1, val2) DCHECK(!(val1) || (val2))
 
-#if defined(NDEBUG) && defined(OS_CHROMEOS)
+#if defined(NDEBUG) && !defined(DCHECK_ALWAYS_ON) && defined(OS_CHROMEOS)

[Peter Kasting, 2014/10/27 21:36:18]

This seems fragile.  We're slowly redeclaring all the conditions that control
DCHECK.

Can we instead do something like:

#if defined(NDEBUG) && defined(OS_CHROMEOS)
#define NOTREACHED() do { LOG(ERROR) << "blah"; DCHECK(false); } while (false)
#else
#define NOTREACHED() DCHECK(false)
#endif

Or maybe even better, make DCHECK failure in ChromeOS release builds be
LOG(ERROR) instead of nothing, eliminating the need to define something here,
and catching even more cases?

[scottmg, 2014/10/27 21:38:04]

(do while(0) won't work with notreached() << "stuff";)

[dmichael (off chromium), 2014/10/27 21:40:25]

do while won't, but it might be OK to use:
DCHECK(false) << "NOTREACHED() hit in..."
so that NOTREACHED will expand to just another <<. Then you could still not have
to rely on DCHECK_ALWAYS_ON

[danakj, 2014/10/27 21:42:34]

Why not #if DCHECK_IS_ON?

[oshima, 2014/10/27 22:04:16]

looking at the bug crbug.com/288014, I wonder if we want to keep this.
zork@, zel@, you guys added this line. What do you want to keep this or ok to
remove?

[oshima, 2014/10/27 22:04:16]

IIRC, DCHECK is no-op on normal rel build.

[oshima, 2014/10/27 22:04:16]

Sure done.

 #define NOTREACHED() LOG(ERROR) << "NOTREACHED() hit in " << \
     __FUNCTION__ << ". "
 #else