[ruby-core:97630] [Ruby master Bug#15935] Memory leak triggered by String#encode, possibly elsewhere too

Issue #15935 has been updated by usa (Usaku NAKAMURA).

Backport changed from 2.4: UNKNOWN, 2.5: REQUIRED, 2.6: DONE to 2.4: UNKNOWN, 2.5: DONE, 2.6: DONE

ruby_2_5 r67860 merged revision(s) 78ef2d0f331c3e056ee367214710b41722de2fe0.

----------------------------------------
Bug #15935: Memory leak triggered by String#encode, possibly elsewhere too
https://bugs.ruby-lang.org/issues/15935#change-84823

* Author: luke-gru (Luke Gruber)
* Status: Closed
* Priority: Normal
* Backport: 2.4: UNKNOWN, 2.5: DONE, 2.6: DONE
----------------------------------------
Hi, I've found a leak that can be reproduced in the following way:

```ruby
loop do
    puts "running..."

    50.times do
        File.open("./test/rexml/data/utf16.xml", external_encoding: 'UTF-16LE', binmode: true) do |f| # must be in ruby's srcdir
            while line = f.readline()
                line.encode("UTF-8", "UTF-16LE")
            end
        end
    rescue EOFError
    end
end
```

It seems to be due to `rb_enc_associate_index()`, in encoding.c, which can call `rb_str_change_terminator_length()` with
the given string. Under certain conditions, `str_make_independent_expand()` is called in this function. This function
can allocate a new heap buffer if the string is large enough to not be embeddable, but does not free the previous one, if
necessary.

The following patch seems to fix the leak:
```c
    oldptr = RSTRING_PTR(str);
    if (oldptr) {
	memcpy(ptr, oldptr, len);
    }
    if (!str_dependent_p(str) && !FL_TEST(str, STR_NOFREE) && FL_TEST(str, STR_NOEMBED)) {
        xfree(oldptr);
    }
```

I can add PR if you want, or you can fix it as you see fit with whichever code you prefer.

NOTE: I found this leak by adding more debug assertions to string.c. Basically I added the same code as above, except instead of `xfree` I asserted that the string should never
have a freeable buffer, as this is the assumption the function seemed to make.

Then, when running make test-all, I run into this failed assertion a bunch, especially in rexml tests. This also causes the leak:

```ruby
loop do
  File.open("./test/rexml/data/utf16.xml") do |f|
    REXML::Document.new(f)
  end
end
```

Thanks for your time :)



-- 
https://bugs.ruby-lang.org/

Unsubscribe: <mailto:[e-mail]?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>