1 files changed, 58 insertions, 0 deletions
diff --git a/MEMORY_BUGS.md b/MEMORY_BUGS.md
new file mode 100644
index 00000000..12e0b854
--- /dev/null
+++ b/MEMORY_BUGS.md
@@ -0,0 +1,58 @@
+# Memory Bugs
+
+This is a list of all of the memory bugs that were found in *released* versions
+of `bc`, `dc`, or `bcl`. (Non-released commits with memory bugs do not count.)
+
+I made this list for two reasons: first, so users can know what versions of
+`bc`, `dc`, and `bcl` have vulnerabilities, and two, I once had a perfect record
+and then found a couple, but forgot and claimed I still had a perfect record
+right after, which was embarrassing.
+
+This list is sorted by the first version a bug exists in, not the last it
+existed in.
+
+* In versions `1.1.0` until `6.2.0` (inclusive) of `bc` and `dc`, there is a
+  out of bounds read and write in history when pressing ctrl+r (or any other
+  unused letter) then inserting two characters.
+
+  The first version without this bug is `6.2.1`.
+
+* In versions `3.0.0` until `6.0.1` (inclusive) of `bc` and `dc`, there is a
+  double-free on `SIGINT` when using command-line expressions with `-e` and
+  `-f`. This was caused by not properly ending a jump series.
+
+  The first version without this bug is `6.0.2`.
+
+* In versions `5.0.0` until `6.0.4` (inclusive) of `bc`, there is an
+  out-of-bounds access if a non-local (non-`auto`) variable is set to a string
+  with `asciify()`, then the function is redefined with a use of the same
+  non-local variable.
+
+  This happened because strings were stored per-function, and the non-local
+  variable now had a reference to the string in the old function, which could be
+  at a higher index than exists in the new function. Strings are stored globally
+  now, and they are *not* freed once not used.
+
+  The first version without this bug is `6.1.0`.
+
+* In versions `5.0.0` until `6.0.4` (inclusive) of `bc`, there is another
+  out-of-bounds access if an array is passed to the `asciify()` built-in
+  function as the only argument. This happened because arrays are allowed as
+  function arguments, which allowed them to be used as arguments to `asciify()`,
+  but they should not have been allowed. However, since they were, the
+  `asciify()` code tried to access an argument that was not there.
+
+  The first version without this bug is `6.1.0`.
+
+* In version `6.0.0` of `bcl`, there are several uses of initialized data that
+  have the same root cause: I forgot to call `memset()` on the per-thread global
+  data. This is because the data used to be *actually* global, which meant that
+  it was initialized to zero by the system. This happened because I thought I
+  had properly hooked Valgrind into my `bcl` tests, but I had not.
+
+  The first version without this bug is `6.0.1`.
+
+* In version `6.0.0` until `6.2.4` (inclusive) of `bcl`, there is a possible
+  use-after-free if `bcl_init()` fails.
+
+  The first version without this bug is `6.2.5`.