as sberg points out, it needs a massive blacklist, which is going
to be a pain to maintain. Until I can think of a way to make this
more resilient, move it where it can't bother people.
Change-Id: I64fc498146a3cb4ea9c1e37150c4ef50023b7bd5
The problem being that any lib later loaded via osl_loadModule (e.g.,
libgcc3_uno.so) would not bind to the same global (RTTI-related) symbols as
libsofficeapp.so and its dependencies (so, e.g., -fsanitize=function would
erroneously assume that bridges/source/cpp_uno/shared/component.cxx's
uno_initEnvironment is called with a different, non-matching uno_Environment
type).
Change-Id: I08b0cbc1f9eb74641eb617c46587a0a528a56c31
...to special member function declarations that were left undefined. Helps
compilers do a better job at identifiying unused class members. This plugin
uses heuristics and whitelists do identify applicable declarations, but is not
appropriate for "unattended" use so is placed into store/.
The following commits contain the results of running this plugin, per module:
* Declarations of undefined special member functions are mmarked
SAL_DELETED_FUNCTION (aka "= delete", which is deemed superior to deriving the
class from boost::noncopyable, cf. Howard Hinnant's reply to
<http://stackoverflow.com/questions/7823990/what-are-the-advantages-of-boostnoncopyable>.
* Any redundant "explicit" or SAL_DLLPRIVATE markers are removed from the
deleted definitions.
* Some redundant declarations of undefined default ctors are simply removed;
smelled like clueless cargo-cult to have them declared at all.
* Some declarations of undefined operator == etc. are left in (and marked
SAL_DELETED_FUNCTION) for now, to be on the safe side, though they are likely
clueless cargo-cult, too.
* Most "static-only" classes are replaced with namespaces (and some where that
would be non-trivial due to private members are marked TODO for later).
* Newly identified unused class members are removed.
Change-Id: Ibeaae4fd579d7a0971a2c2a654a2263acd13414a
we don't need to run them on an ongoing basis, and the current code
does not compile with older versions of clang.
Change-Id: I07ccacf7ff7b00e8e2453fff91a3f487dd5abed9
Where we can prove that the virtual method is never overriden.
In the case of pure-virtual methods, we remove the method entirely.
Sometimes this leads to entire methods and fields being
eliminated.
Change-Id: I138ef81c95f115dbd8c023a83cfc7e9d5d6d14ae
Stumbled across such redundant visibility re-specifications when looking at the
odd case of cppu_unsatisfied_iquery_msg declared CPPU_DLLPUBLIC in
cppu/source/cppu/cppu_opt.cxx and used in inline code in
include/com/sun/star/uno/Reference.hxx with only a declaration lacking
CPPU_DLLPUBLIC visible, and wondering how that actually works on Windows.
However, this plugin is probably not worth it being run all the time, so
committing it to compilerplugins/clang/store/.
Change-Id: Ibc3c4e7499213de1b419ce7eb85455cb832e1510
...some paths trough clang::Expr::isIntegerConstantExpr (esp. if
non-CPlusPlus11) assert the assumption that the given expr is not
value-dependent, so it appears to be a prereq
Change-Id: Ibc5fe472ea3f91b31c8cb7f06c4b7c7f4d6831a3
(Original idea from Kendy)
Look for code that is calling std::find on a sorted container
(set/map/vector) and warn about it - the code should be using
the find method on the container itself, since that is considerably faster.
Change-Id: Ib74e5d3faa836eeb0df16a736d202696626bdfd2
...and is apparently a leftover from temporary debug output in
e36badb98d0bb5866a297cb51c3e95cdce62d8da "Fix workaround for bug in Clang 3.2
FunctionDecl::isInlined."
Change-Id: I3213981c5d236a7b67083014692566f75a2bcd51
