Fixed Windows DLL builds of test targets

This is a heavily modified version of
https://github.com/abseil/abseil-cpp/pull/1445,
which adds some missing test libraries to the test DLL.

Unlike #1445, this change moves several global variables out of
headers that did not need to be in headers.

For instance, cord_btree_exhaustive_validation was a global
defined/declared in cord_internal, but only used in cord_rep_btree
and its test.

cordz_handle defined a queue in its header even though it wasn't needed,
which also led to ODR problems.

The Spinlock used in CordzHandle is replaced with a Mutex. This was
originally a Mutex, but Chromium asked us to change it to a Spinlock
to avoid a static initializer. After this change, the static
initializer is no longer an issue.

#1407

PiperOrigin-RevId: 531516991
Change-Id: I0e431a193698b20ba03fac6e414c26f153f330a7

1 files changed, 0 insertions, 6 deletions

diff --git a/absl/strings/internal/cord_internal.h b/absl/strings/internal/cord_internal.h
index e6f0d544..8d9836ba 100644
--- a/absl/strings/internal/cord_internal.h
+++ b/absl/strings/internal/cord_internal.h
@@ -69,12 +69,6 @@ enum CordFeatureDefaults {
 extern std::atomic<bool> cord_ring_buffer_enabled;
 extern std::atomic<bool> shallow_subcords_enabled;
 
-// `cord_btree_exhaustive_validation` can be set to force exhaustive validation
-// in debug assertions, and code that calls `IsValid()` explicitly. By default,
-// assertions should be relatively cheap and AssertValid() can easily lead to
-// O(n^2) complexity as recursive / full tree validation is O(n).
-extern std::atomic<bool> cord_btree_exhaustive_validation;
-
 inline void enable_cord_ring_buffer(bool enable) {
   cord_ring_buffer_enabled.store(enable, std::memory_order_relaxed);
 }