diff --git a/src/ds/bits.h b/src/ds/bits.h
index 4aa8e11..68df89e 100644
--- a/src/ds/bits.h
+++ b/src/ds/bits.h
@@ -452,5 +452,17 @@ namespace snmalloc
         return (size_t)1 << (m_e + LOW_BITS);
       }
     }
+
+    template <typename T>
+    constexpr inline T min(T t1, T t2)
+    {
+      return t1 < t2 ? t1 : t2;
+    }
+
+    template <typename T>
+    constexpr inline T max(T t1, T t2)
+    {
+      return t1 > t2 ? t1 : t2;
+    }
   }
 }
diff --git a/src/ds/mpscq.h b/src/ds/mpscq.h
index e7395f1..1993e94 100644
--- a/src/ds/mpscq.h
+++ b/src/ds/mpscq.h
@@ -1,9 +1,7 @@
 #pragma once
 
 #include "bits.h"
-
-#include <stdlib.h>
-#include <utility>
+#include "helpers.h"
 
 namespace snmalloc
 {
diff --git a/src/mem/alloc.h b/src/mem/alloc.h
index efd7d4e..a344ec8 100644
--- a/src/mem/alloc.h
+++ b/src/mem/alloc.h
@@ -21,8 +21,6 @@
 #include "sizeclasstable.h"
 #include "slab.h"
 
-#include <array>
-
 namespace snmalloc
 {
   enum Boundary
diff --git a/src/mem/globalalloc.h b/src/mem/globalalloc.h
index 9471264..5a4921a 100644
--- a/src/mem/globalalloc.h
+++ b/src/mem/globalalloc.h
@@ -49,6 +49,7 @@ namespace snmalloc
       }
     }
 
+#ifdef USE_SNMALLOC_STATS
     void print_all_stats(std::ostream& o, uint64_t dumpid = 0)
     {
       auto alloc = Parent::iterate();
@@ -59,6 +60,12 @@ namespace snmalloc
         alloc = Parent::iterate(alloc);
       }
     }
+#else
+    void print_all_stats(void*& o, uint64_t dumpid = 0)
+    {
+      UNUSED(o); UNUSED(dumpid);
+    }
+#endif
 
     void cleanup_unused()
     {
diff --git a/src/mem/largealloc.h b/src/mem/largealloc.h
index ab89d11..6ac38cf 100644
--- a/src/mem/largealloc.h
+++ b/src/mem/largealloc.h
@@ -8,8 +8,6 @@
 #include "baseslab.h"
 #include "sizeclass.h"
 
-#include <utility>
-
 namespace snmalloc
 {
   template<class PAL>
@@ -61,7 +59,7 @@ namespace snmalloc
     size_t bump;
     size_t remaining;
 
-    std::pair<void*, size_t> reserve_block() noexcept
+    void new_block()
     {
       size_t size = SUPERSLAB_SIZE;
       void* r = reserve<false>(&size, SUPERSLAB_SIZE);
@@ -70,14 +68,9 @@ namespace snmalloc
         error("out of memory");
 
       ((PAL*)this)->template notify_using<NoZero>(r, OS_PAGE_SIZE);
-      return std::make_pair(r, size);
-    }
 
-    void new_block()
-    {
-      auto r_size = reserve_block();
-      bump = (size_t)r_size.first;
-      remaining = r_size.second;
+      bump = (size_t)r;
+      remaining = size;
     }
 
     /**
diff --git a/src/mem/pagemap.h b/src/mem/pagemap.h
index 9fae7a3..1ca4688 100644
--- a/src/mem/pagemap.h
+++ b/src/mem/pagemap.h
@@ -1,9 +1,10 @@
 #pragma once
 
 #include "../ds/bits.h"
+#include "../ds/helpers.h"
 
-#include <algorithm>
 #include <atomic>
+#include <utility>
 
 namespace snmalloc
 {
@@ -274,7 +275,7 @@ namespace snmalloc
         auto leaf_ix = get_leaf_index<true>(p, success);
         size_t ix = leaf_ix.second;
 
-        auto last = std::min(LEAF_MASK + 1, ix + length);
+        auto last = bits::min(LEAF_MASK + 1, ix + length);
 
         auto diff = last - ix;
 
diff --git a/src/mem/sizeclasstable.h b/src/mem/sizeclasstable.h
index 64e0347..877c39d 100644
--- a/src/mem/sizeclasstable.h
+++ b/src/mem/sizeclasstable.h
@@ -30,7 +30,7 @@ namespace snmalloc
           bits::from_exp_mant<INTERMEDIATE_BITS, MIN_ALLOC_BITS>(sizeclass);
 
         size_t alignment =
-          std::min((size_t)1 << bits::ctz_const(size[sizeclass]), OS_PAGE_SIZE);
+          bits::min((size_t)1 << bits::ctz_const(size[sizeclass]), OS_PAGE_SIZE);
         cache_friendly_mask[sizeclass] = (alignment - 1);
         inverse_cache_friendly_mask[sizeclass] = ~(alignment - 1);
       }
diff --git a/src/mem/superslab.h b/src/mem/superslab.h
index 9105b8b..1c127a6 100644
--- a/src/mem/superslab.h
+++ b/src/mem/superslab.h
@@ -3,8 +3,7 @@
 #include "../ds/helpers.h"
 #include "allocslab.h"
 #include "metaslab.h"
-
-#include <cstring>
+#include <new>
 
 namespace snmalloc
 {
diff --git a/src/mem/threadalloc.h b/src/mem/threadalloc.h
index 8914aec..032fa39 100644
--- a/src/mem/threadalloc.h
+++ b/src/mem/threadalloc.h
@@ -7,6 +7,10 @@
 #error At most one out of SNMALLOC_USE_THREAD_CLEANUP and SNMALLOC_USE_THREAD_DESTRUCTOR may be defined.
 #endif
 
+#if !defined(_WIN32) && !defined(FreeBSD_KERNEL)
+# include "pthread.h"
+#endif
+
 namespace snmalloc
 {
   extern "C" void _malloc_thread_cleanup(void);
diff --git a/src/override/malloc.cc b/src/override/malloc.cc
index 768b619..18ddab6 100644
--- a/src/override/malloc.cc
+++ b/src/override/malloc.cc
@@ -2,6 +2,7 @@
 #include "../snmalloc.h"
 
 #include <errno.h>
+#include <string.h>
 
 using namespace snmalloc;
 
@@ -89,7 +90,7 @@ extern "C"
     if (p != nullptr)
     {
       assert(p == Alloc::external_pointer<Start>(p));
-      sz = (std::min)(size, sz);
+      sz = bits::min(size, sz);
       memcpy(p, ptr, sz);
       SNMALLOC_NAME_MANGLE(free)(ptr);
     }
@@ -135,7 +136,7 @@ extern "C"
       return nullptr;
     }
 
-    size = (std::max)(size, alignment);
+    size = bits::max(size, alignment);
     uint8_t sc = size_to_sizeclass(size);
     if (sc >= NUM_SIZECLASSES)
     {
diff --git a/src/pal/pal_apple.h b/src/pal/pal_apple.h
index 4c5e643..f1c92d2 100644
--- a/src/pal/pal_apple.h
+++ b/src/pal/pal_apple.h
@@ -5,10 +5,11 @@
 #  include "../mem/allocconfig.h"
 
 #  include <pthread.h>
-#  include <stdio.h>
 #  include <strings.h>
 #  include <sys/mman.h>
 
+int puts ( const char * str );
+
 namespace snmalloc
 {
   /**
diff --git a/src/pal/pal_freebsd.h b/src/pal/pal_freebsd.h
index 122b8b9..1c8b04e 100644
--- a/src/pal/pal_freebsd.h
+++ b/src/pal/pal_freebsd.h
@@ -4,7 +4,6 @@
 #  include "../ds/bits.h"
 #  include "../mem/allocconfig.h"
 
-#  include <pthread.h>
 #  include <stdio.h>
 #  include <strings.h>
 #  include <sys/mman.h>
diff --git a/src/pal/pal_linux.h b/src/pal/pal_linux.h
index f1e2d01..5ded0a9 100644
--- a/src/pal/pal_linux.h
+++ b/src/pal/pal_linux.h
@@ -4,10 +4,11 @@
 #  include "../ds/bits.h"
 #  include "../mem/allocconfig.h"
 
-#  include <stdio.h>
 #  include <string.h>
 #  include <sys/mman.h>
 
+int puts ( const char * str );
+
 namespace snmalloc
 {
   class PALLinux