tre-config.h contains additional defines that their configure puts in config.h. We have remapped the POSIX entry points with a tre_ suffix: without it there was confusion with entry points in OS libraries on some OSes (e.g. Mac OS X). Pro tem alloca has been disabled: the declarations used do not work on Windows, and most likely not on FreeBSD. The way we compile it, TRE works internally in one of three modes STR_BYTE, using bytes STR_MBS , using wchar_t for matching, I/O in MBCS STR_WIDE, using wchar_t We've added interfaces tre_regcompb, tre_regncompb, tre_regexecb, tre_regnexecb and regaexecb to force STR_BYTE. Unless forced, bytes are used in a single-byte locale and wchar_t in a MBCS. We need to force for useBytes=TRUE, for support for UTF-8 encoded strings in a single-byte locale (principally on Windows) and when dealing with raw vectors. You do need to ensure that regcomp and regexec use the same type. The offsets in the regmatch_t structure are in characters for STR_WIDE, bytes for the other two modes. On Windows (MinGW?), iswctype is missing 'blank'. REG_LITERAL was not matching "" in some cases, hence a change at line 1652 of tre-parse.c. We fixed stack trampling in tre_tnfa_run_approx() and converted tre_ast_to_tnfa() from recursion to iteration over a TRE stack. We commented out in tre-match-approx.c /* These seem compiler/OS-specific, but unexplained On Linux the first is intended to be used only with GCC. #define __USE_STRING_INLINES #undef __NO_INLINE__ */ since it failed on clang https://stat.ethz.ch/pipermail/r-devel/2010-February/056728.html Nov 2011: incorporated minor fixes from https://github.com/GerHobbelt/libtre (see http://laurikari.net/tre/website-issues-and-future-plans/#comments).