[wasm] Prepare WasmCompilationUnit for lazy compilation

src/vector.h

Index: src/vector.h
diff --git a/src/vector.h b/src/vector.h
index eb5808322c731f404c3ff2b2d1cfcc2490b78526..03e5d6cb1ea2f42162b878ead3773c46b68335fb 100644
--- a/src/vector.h
+++ b/src/vector.h
@@ -35,7 +35,7 @@ class Vector {
   // spanning from and including 'from', to but not including 'to'.
   Vector<T> SubVector(int from, int to) const {
     DCHECK(0 <= from);
-    SLOW_DCHECK(from < to);
+    SLOW_DCHECK(from <= to);

[titzer, 2017/03/03 09:21:44]

Any reason this is a SLOW_DCHECK? I wonder if it's not faster just to make this
a DCHECK, since it's single comparison. Same below.

[Clemens Hammacher, 2017/03/03 09:29:53]

I don't know the reason for this. Maybe just to reduce code size (assuming that
this function gets inlined a lot)? SLOW_DCHECK compiles to nothing if not
compiled with v8_enable_slow_dchecks.
Also note that the first SLOW_DCHECK is redundant with the DCHECK in the Vector
constructor.
Adding jochen@ who added the SLOW_DCHECKs more than two years ago.

[jochen (gone - plz use gerrit), 2017/05/03 10:47:22]

Note that this change breaks gcc compilation.

(I just renamed the macro, it was a slow dcheck before as well)

[Clemens Hammacher, 2017/05/03 11:01:55]

For reference: You are referring to this bug: http://crbug.com/v8/6341

How does my change cause this? Wouldn't gcc have complained about "num_chars - 1
< num_chars" as well?

     SLOW_DCHECK(static_cast<unsigned>(to) <= static_cast<unsigned>(length_));
     return Vector<T>(start() + from, to - from);
   }