Well, it looks like we should yank exception specifications. On Mon, Sep 5, 2011 at 3:54 AM, Jonathan Wakely <jwakely.gcc@xxxxxxxxx> wrote: > On 5 September 2011 05:13, Jeffrey Walton <noloader@xxxxxxxxx> wrote: >> On Sun, Sep 4, 2011 at 6:50 PM, Jeffrey Walton <noloader@xxxxxxxxx> wrote: >>> Hi All, >>> >>> I'm seeing the following on a negative test case: >>> >>> terminate called after throwing an instance of >>> 'esapi::NoSuchAlgorithmException' >>> what(): Algorithm 'Foo' is not supported >>> >>> Any ideas on what might cause the exception to pass uncaught? I am >>> familiar with http://gcc.gnu.org/faq.html#dso. >>> >>> [SNIP] >> >> Found it.... MessageDigest md1("Foo") uses 'createInstance(...)' in a >> PIMPL. However, createInstance(...) [incorrectly] declared it throws a >> "InvalidArgumentException". In reality, the function throws a >> "NoSuchAlgorithmException". >> >> Visual Studio ran fine because MSVC ignores the exception >> specification (http://msdn.microsoft.com/en-us/library/wfa0edys%28v=VS.90%29.aspx). >> >> I'm running with -Wall, -Wextra, which did not help me. Would anyone >> know the warning I am looking for? I did not see it at >> http://gcc.gnu.org/onlinedocs/gcc/Warning-Options.html. > > There's no such warning, C++ exception specifications are not > statically checked, ever. C++ is not Java. > > The only good exception specification is throw() (renamed noexcept in > C++11), best practice is not to use non-empty exception specs. Pretty > much the only effect non-empty specs have is to cause unwanted > terminations, as you've discovered, and they're deprecated in C++11. > If you want to document what exceptions can be thrown, use a comment. > If it's wrong you won't terminate immediately. > > http://www.gotw.ca/publications/mill22.htm >
