For char * strings we manually manage stack buffers and allocated buffers, use strcpy(), memcpy(), strcat(), ... The udata.c code even uses a "tiny string" poor man's string "class", all written in C.

Wouldn't it be nice if we could use STL string?

Status

As of ICU 4.6 Milestone 1 (4.5.1), the old CharString class has been turned into a minimalist but fully usable C++ char * string class for ICU-internal use. (The old CharString class only served for invariant-character conversion from UnicodeString to const char *.)

See ticket 7496 and its commits.

Old design idea

It should be possible to define

class icu_string : public UMemory, stl::basic_string<char, char_traits<char>, icu_allocator<char> > {};

Note:

• Inheriting UMemory means that we can do "new icu_string" and not use global new and delete.
• Providing our own icu_allocator means that when the icu_string allocates its internal buffer, it won't use global new and delete but rather our own uprv_malloc() and uprv_free().
• We could actually provide our own traits class which might use int32_t for string indexes instead of streampos.

Question: What should happen in an out-of-memory situation? The icu_allocator could

• either throw a bad_alloc exception like the std::allocator, and all icu_string-using code would have to catch that exception,
• or return NULL and let the icu_string implementation crash.
  

References

• http://www.sgi.com/tech/stl/basic_string.html
• http://www.cplusplus.com/reference/string/
• http://www.cplusplus.com/reference/std/memory/allocator/
• http://www.sgi.com/tech/stl/character_traits.html