Use a different map to distinguish eval contexts

src/flag-definitions.h

Index: src/flag-definitions.h
diff --git a/src/flag-definitions.h b/src/flag-definitions.h
index 0667735e59acafc10b0fbb27c3b20d724ec5f6b7..e6d6baa7670ae76889d98976d6a09d17f42992a4 100644
--- a/src/flag-definitions.h
+++ b/src/flag-definitions.h
@@ -1136,6 +1136,9 @@ DEFINE_BOOL(test_secondary_stub_cache, false,
 DEFINE_BOOL(test_primary_stub_cache, false,
             "test primary stub cache by disabling the secondary one")
 
+DEFINE_BOOL(test_small_max_function_context_stub_size, false,

[adamk, 2016/12/14 12:23:31]

Can you elaborate on why a flag was needed here? Did you try creating a test
case that would exceed the real maximum slots and find it was too slow?

[Dan Ehrenberg, 2016/12/14 22:53:31]

Yes, I found that it consistently made my Z620 run out of memory and freeze up.
It also failed in automation. I manually verified that if I turn on the flag and
introduce a bug that makes the bytecode generator always use FUNCTION_SCOPE for
the overflow path, then one of the newly added tests fails. I didn't add a
comment in the source because none of the other flags have comments.

[adamk, 2016/12/15 08:46:52]

Don't think you need to add a comment, this response is what I meant by
"elaborate". Thanks :)

+            "enable testing the function context size overflow path "
+            "by making the maximum size smaller")
 
 // codegen-ia32.cc / codegen-arm.cc
 DEFINE_BOOL(print_code, false, "print generated code")