=begin
I didn't think of this, but the patch can also apply to 1.8.7. The line numbers may differ, but the code is the same.

I'm not sure where the 1.8.7 branch is in git or how to submit to that. I'd be happy to submit that to Shyouhei as well.

Ciao!
=end

=begin
Hi,

At Wed, 11 Nov 2009 04:50:33 +0900,
Christian Höltje wrote in [ruby-core:26668]:

The environ manipulations that are done in the fall-back code
in hash.c (the code that runs if !WIN32 and !HAVE_SETENV)
causes crashes in Solaris some times. In addition, it fails
hard if you use the binary on a newer version of Solaris.

What does this note mean?

  * Note: This does leak memory and there isn't anything we can do
  * about it.

putenv()ed string is stored in environ but isn't freed?

--
Nobu Nakada

=end

=begin
Because new variables are allocated by ALLOC_N() but the unsetenv code cannot de-allocate the memory: we don't know if it is initially allocated memory by the kernel (which I think is stack memory and absolutely cannot be free()d) or if it is something we did a putenv() on (which we cannot free() either, depending on the platform).

Ciao!
=end

=begin
Hi,

At Wed, 11 Nov 2009 11:32:37 +0900,
Christian Höltje wrote in [ruby-core:26675]:

Because new variables are allocated by ALLOC_N() but the
unsetenv code cannot de-allocate the memory: we don't know if
it is initially allocated memory by the kernel (which I think
is stack memory and absolutely cannot be free()d) or if it is
something we did a putenv() on (which we cannot free()
either, depending on the platform).

Does this work?

Index: hash.c¶

--- hash.c (revision 25717)
+++ hash.c (working copy)
@@ -2012,5 +2012,16 @@ rb_env_path_tainted(void)
}

-#if !defined(_WIN32) && !(defined(HAVE_SETENV) && defined(HAVE_UNSETENV))
+#if defined(_WIN32) || (defined(HAVE_SETENV) && defined(HAVE_UNSETENV))
+#elif defined sun
+static int
+in_origenv(const char *str)
+{

• char **env;
• for (env = origenviron; *env; ++env) {
• if (*env == str) return 1;
• }
• return 0;
  +}
  +#else
  static int
  envix(const char *nam)
  @@ -2058,4 +2069,19 @@ ruby_setenv(const char *name, const char
  else
  unsetenv(name);
  +#elif defined sun
• size_t len = strlen(name);
• char **env_ptr, *str;
• for (env_ptr = GET_ENVIRON(environ); (str = *env_ptr) != 0; ++env_ptr) {
• if (!strncmp(str, name, len) && str[len] == '=') {

•  if (!in_origenv(str)) free(str);
  

•  while ((env_ptr[0] = env_ptr[1]) != 0) env_ptr++;
  

•  break;
  

• }
• }
• if (value) {
• str = malloc(len += strlen(value) + 2);
• snprintf(str, len, "%s=%s", name, value);
• putenv(str);
• }
  #else /* WIN32 */
  size_t len;
  

--
Nobu Nakada

=end

=begin
I'm not sure the reclaiming is actually working. I ran some tests and it at least appears that it may not be reclaiming much or any. I'm not entirely sure how to test that.

I think I would recommend using the original style unsetenv code I used and just live with the memory leak. It's not usually that much and is usually done right before an exec()...

=end