Given the RTF input "{\rtf{\b h\'e6}}", unrtf fails to make the escaped 0xe6 (æ) bold. The expected output HTML is "<b>hæ</b>", but unrtf writes "<b>h</b>æ". When there are additional unescaped characters after it (or another word, e.g. a dummy "{\*\0}") unrtf produces the expected output.

The issue seems to be that when the hex sequence is parsed, it is written to a buffer that is not flushed until after the attribute end tag has been written. In this minimal example I am able to fix the problem by moving "flush_iconv_input()" a few lines up in convert.c, but I am not sure if I am breaking anything else. I ran tests/test_html.sh and all files were converted successfully. I diffed the output files and, as far as I can see, the only difference between them is that attribute end tags are placed after the escaped characters. In other words, it looks to me like this is a safe change, but I am not 100 % sure.

Patch (for 0.21.10):
--- src/convert.c
+++ src/convert.c
@@ -4503,6 +4503,10 @@
  within_picture = FALSE;
  }

+ // Flush iconv input
+ flush_iconv_input();
+ iconv_cur = 0;
+
  /* Undo font attributes UNLESS we're doing table cells
  * since they would appear between </td> and </tr>.
  */
@@ -4525,10 +4529,6 @@

  attrstack_drop();

- // Flush iconv input
- flush_iconv_input();
- iconv_cur = 0;
-
  // If there is an encoding in the stacks, restore it, else
  // restore default.
  desc = my_iconv_close(desc);