about summary refs log tree commit diff stats
path: root/071deep_copy.cc
diff options
context:
space:
mode:
authorKartik K. Agaram <vc@akkartik.com>2018-01-03 00:31:10 -0800
committerKartik K. Agaram <vc@akkartik.com>2018-01-03 00:44:09 -0800
commitacce384bcc88d5b300b913c14b9872081a182155 (patch)
treea21c33d342c44382b08e37a212a2e79416baca45 /071deep_copy.cc
parentc8eb6c1a64d76dc9a1005571c4eb71ddc6d8f2a9 (diff)
downloadmu-acce384bcc88d5b300b913c14b9872081a182155.tar.gz
4179 - experiment: rip out memory reclamation
I have a plan for a way to avoid use-after-free errors without all the
overheads of maintaining refcounts. Has the nice side-effect of
requiring manual memory management. The Mu way is to leak memory by
default and build tools to help decide when and where to expend effort
plugging memory leaks. Arguably programs should be distributed with
summaries of their resource use characteristics.

Eliminating refcount maintenance reduces time to run tests by 30% for
`mu edit`:

              this commit                 parent
  mu test:         3.9s                        4.5s
  mu test edit:  2:38                        3:48

Open questions:
  - making reclamation easier; some sort of support for destructors
  - reclaiming local scopes (which are allocated on the heap)
    - should we support automatically reclaiming allocations inside them?
Diffstat (limited to '071deep_copy.cc')
-rw-r--r--071deep_copy.cc15
1 files changed, 2 insertions, 13 deletions
diff --git a/071deep_copy.cc b/071deep_copy.cc
index ffeb00bd..f5f0f1eb 100644
--- a/071deep_copy.cc
+++ b/071deep_copy.cc
@@ -3,9 +3,6 @@
 //
 // Invariant: After a deep-copy its ingredient and result will point to no
 // common addresses.
-// Implications: Refcounts of all data pointed to by the original ingredient
-// will remain unchanged. Refcounts of all data pointed to by the (newly
-// created) result will be 1, in the absence of cycles.
 //
 // We do handle cycles in the ingredient, however. All cycles are translated
 // to new cycles in the product.
@@ -60,11 +57,6 @@ def main [
 +mem: storing 0 in location 10
 # however, the contents are identical
 +mem: storing 1 in location 11
-# the result of deep-copy gets a refcount of 1
-# (its address 202 = 200 base + 2 for temporary space inside deep-copy)
-+run: {2: ("address" "number")} <- copy {0: "literal"}
-+mem: decrementing refcount of 202: 1 -> 0
-+abandon: saving 202 in free-list of size 2
 
 :(scenario deep_copy_address_to_container)
 % Memory_allocated_until = 200;
@@ -107,8 +99,7 @@ def main [
 def main [
   # avoid all memory allocations except the implicit ones inside deep-copy, so
   # that the result is deterministic
-  100:num <- copy 1  # pretend refcount
-  101:num <- copy 3  # pretend array length
+  100:num <- copy 3  # pretend array length
   1:&:@:num <- copy 100/unsafe  # pretend allocation
   put-index *1:&:@:num, 0, 34
   put-index *1:&:@:num, 1, 35
@@ -234,7 +225,7 @@ vector<double> deep_copy(const reagent& in) {
   vector<double> result = deep_copy(in, addresses_copied, tmp);
   // reclaim Mu memory allocated for tmp
   trace(9991, "run") << "deep-copy: reclaiming temporary" << end();
-  abandon(tmp.value, payload_type(tmp.type), payload_size(tmp));
+  abandon(tmp.value, payload_size(tmp));
   // reclaim host memory allocated for tmp.type when tmp goes out of scope
   return result;
 }
@@ -263,8 +254,6 @@ int deep_copy_address(const reagent& canonized_in, map<int, int>& addresses_copi
   if (contains_key(addresses_copied, in_address)) {
     int out = get(addresses_copied, in_address);
     trace(9991, "run") << "deep-copy: copy already exists: " << out << end();
-    assert(contains_key(Memory, out));  // refcount must already be incremented
-    ++get(Memory, out);
     return out;
   }
   int out = allocate(payload_size(canonized_in));