On Sun, 7 Mar 2021 22:00:41 GMT, Sergey Bylokhov <s...@openjdk.org> wrote:
>> Yes, including c++ standard library headers like <ostream> means you need to >> deal with C++ exceptions thrown from library functions, and the code needs >> to be compiled with unwind capabilities. If its not switched on, and a C++ >> exception happens, the behavior is undefined. In my experience it results in >> the process being terminated. >> >> I wondered why C++ std headers are even used. The source code looks C-ish; >> but "8196681: Java Access Bridge logging and debug flags dynamically >> controlled" added some coding, adding a bunch of C++11x semantics and >> included C++ std headers. Using "/Ehsc" had even been discussed: >> https://mail.openjdk.java.net/pipermail/awt-dev/2018-December/014847.html >> but not done. >> >> Adding /Ehsc is fine of course, but I think thats not enough. Now C++ >> exceptions can pass through the code, but if they do, then what? E.g. this >> function `getTimeStamp()` added with 8196681: >> >> https://github.com/openjdk/jdk/blob/113b5184cfc78fde057a7fe4d5872b463930da00/src/jdk.accessibility/windows/native/common/AccessBridgeDebug.cpp#L85 >> >> calls into the C++ standard library to get a time stamp. Can't these >> function not throw C++ exceptions? Is that not what the compiler is warning >> about? If they throw, exceptions propagate through the code unbounded, until >> they either end the process or cause havoc at the next upper >> C-only-interface. >> >> Or, maybe, rewrite this coding to use standard C- and Windows-APIs. I think >> 8196681 could had been done with traditional windows- or standard C APIs. In >> particular, `getTimeStamp()` could probably be done simply with >> `GetTickCount` or `GetSystemTime` or functions from time.h. >> >> Cheers, Thomas > >> I wondered why C++ std headers are even used. The source code looks C-ish; >> but "8196681: Java Access Bridge logging and debug flags dynamically >> controlled" added some coding, adding a bunch of C++11x semantics and >> included C++ std headers. Using "/Ehsc" had even been discussed: >> https://mail.openjdk.java.net/pipermail/awt-dev/2018-December/014847.html >> but not done. > > After that discussion the usage of "string.h"/etc was dropped and "C string > manipulation API" was used instead, and that suppressed the warning. I do not > know why it was not complaining about "getTimeStamp". I think the fix should > refactor the code again. Thank you for comments! I refactored `getTimeStamp()` not to use STL functions in new commit. The error messages have gone without /EHsc. I use `11644473600000ULL` to rebase Epoch from 1601 to 1970, it comes from ProcessHandleImpl_win.c: https://github.com/openjdk/jdk/blob/22a3117d229cba10c690a4e66baf9c754a09e57c/src/java.base/windows/native/libjava/ProcessHandleImpl_win.c#L342 ------------- PR: https://git.openjdk.java.net/jdk/pull/2859
