Don't read past end of string in Guess ctor

<https://ci.libreoffice.org//job/lo_ubsan/1082/>:
> ==26422==ERROR: AddressSanitizer: heap-buffer-overflow on address 0x604000accc72 at pc 0x2ae43e63f4b6 bp 0x2ae43e600510 sp 0x2ae43e600508
> READ of size 1 at 0x604000accc72 thread T70 (cppu_threadpool)
>     #0 0x2ae43e63f4b5 in Guess::Guess(char const*) /lingucomponent/source/languageguessing/guess.cxx:95:28
>     #1 0x2ae43e667f2f in SimpleGuesser::GetManagedLanguages(char) /lingucomponent/source/languageguessing/simpleguesser.cxx:169:19
>     #2 0x2ae43e668420 in SimpleGuesser::GetAvailableLanguages() /lingucomponent/source/languageguessing/simpleguesser.cxx:179:12
>     #3 0x2ae43e64a18e in LangGuess_Impl::getEnabledLanguages() /lingucomponent/source/languageguessing/guesslang.cxx:229:24
[...]
> 0x604000accc72 is located 0 bytes to the right of 34-byte region [0x604000accc50,0x604000accc72)
> allocated by thread T70 (cppu_threadpool) here:
[...]
>     #7 0x2ae43e65350a in std::string::operator+=(char const*) /home/tdf/lode/opt_private/lib/gcc/x86_64-unknown-linux-gnu/5.2.0/../../../../include/c++/5.2.0/bits/basic_string.h:3355:16
>     #8 0x2ae43e667e6e in SimpleGuesser::GetManagedLanguages(char) /lingucomponent/source/languageguessing/simpleguesser.cxx:168:21
>     #9 0x2ae43e668420 in SimpleGuesser::GetAvailableLanguages() /lingucomponent/source/languageguessing/simpleguesser.cxx:179:12
>     #10 0x2ae43e64a18e in LangGuess_Impl::getEnabledLanguages() /lingucomponent/source/languageguessing/guesslang.cxx:229:24
[...]

shows, during UITest_librelogo, the Guess ctor making wrong assumptions about
the structure of guess_str and skipping over the terminating NUL.  Locally I
could see that while most inputs do have the expected syntax of starting with
"[" and containing two "-", one input is indeed just "[haw-utf8" without a
second "-".

I don't know where the strings passed into the Guess ctor in the two places in
lingucomponent/source/languageguessing/simpleguesser.cxx ultimately come from,
and what their guaranteed syntax and their semantics is.  So from the existing
code and the non--well-formed "[haw-utf8" sample (where the second segment shall
apparently designate an encoding, not a country), construct rules how to
robustly parse any input into potential language/country/encoding parts.  (What
is obvious from the call sites is that for one each input will start with "[",
and for another the item to parse need neither be "]"- nor NUL-terminated.)

(Guess::encoding_str and the local enc variable have effectively been unused
ever since the code's introduction in 0762811922c4e1a3474bdd91daf308baafbc18e4
"INTEGRATION: CWS languageguessing".  Guess::encoding_str, but not the local
enc variable, got later removed with b275246c30ce3796cd22f72cd82c58b5cf4c86f0
"loplugin:unusedfields in formula..registry".)

Change-Id: Icbedc05ed5b119ee4efbc3118cc17076a4d80c74
Reviewed-on: https://gerrit.libreoffice.org/62390
Tested-by: Jenkins
Reviewed-by: Stephan Bergmann <sbergman@redhat.com>

0 files changed, 0 insertions, 0 deletions