Bug 5930 – cas doesn't work when used in code compiled with -D

$ cat cas_test.d import core.atomic; import std.exception; import std.stdio; unittest { shared bool init = false; writefln("init = %s", init); enforce(cas(&init, false, true)); writefln("init = %s", init); } void main() {} $ dmd -unittest cas_test.d && ./cas_test init = false init = true $ dmd -unittest -D cas_test.d && ./cas_test init = false object.Exception@cas_test.d(9): Enforcement failed ---------------- ./cas_test(_D3std9exception7bailOutFAyaixAaZv+0x5b) [0x80a0dcb] ./cas_test(_D3std9exception45__T7enforceTbVAyaa10_6361735f746573742e64Vi9Z7enforceFbLAxaZb+0x2b) [0x80985ff] ./cas_test(_D8cas_test11__unittest1FZv+0x39) [0x80954dd] ./cas_test(_D8cas_test9__modtestFZv+0x8) [0x80987bc] ./cas_test(_D4core7runtime18runModuleUnitTestsUZb16__foreachbody247MFKPS6object10ModuleInfoZi+0x24) [0x80a3e1c] ./cas_test(_D6object10ModuleInfo7opApplyFMDFKPS6object10ModuleInfoZiZi+0x46) [0x809a872] ./cas_test(runModuleUnitTests+0x87) [0x80a3d37] ./cas_test(_D2rt6dmain24mainUiPPaZi6runAllMFZv+0x20) [0x809b718] ./cas_test(_D2rt6dmain24mainUiPPaZi7tryExecMFMDFZvZv+0x20) [0x809b678] ./cas_test(main+0x94) [0x809b624] /lib/i386-linux-gnu/libc.so.6(__libc_start_main+0xe7) [0x126e37] ./cas_test() [0x80953f1] ------------------------- Does this have to do with core/atomic.d having version(D_Doc) bool cas(...)(...) {return false; } ? The work around is not to use object files compiled with the -D flag!

Yep. This is because core.atomic has stub functions for documentation purposes inside version(D_Ddoc) blocks. There was some talk about Phobos and Druntime moving to a different convention for documentation though. I'll see about fixing this for 2.054, since 2.053 is almost in the can.

phobos now uses version StdDdoc instead of D_Ddoc to fix similar issues. Grep tells me core.atomic is the only module in druntime that uses version(D_ddoc). What needs to be done is: - add -version=DruntimeDdoc or similar to DOCFMT in the makefiles - use version(DruntimeDdoc) instead of version(D_Ddoc)

pull https://github.com/D-Programming-Language/druntime/pull/146 Uses the solution proposed by Rainer Schuetze. Instead of DruntimeDdoc, it uses CoreDdoc.

*** Issue 7528 has been marked as a duplicate of this issue. ***

(In reply to comment #4) > *** Issue 7528 has been marked as a duplicate of this issue. *** As mentioned in 7528, this also has the unpleasant side-effect that attempting to use atomic operations (through core.atomic.atomicOp) silently results in incorrect behaviour, as the operations do nothing and return init.

This bug really needs to get fixed for 2.059. It causes incorrect code to be generated and code to silently fail, in ways that may be difficult to track down and occur rarely (namely, when atomicOp is used).

Wouldn't it be better to put a "static assert(false);" in the body of those Ddoc function stubs instead of "return false;"? Then it would become immediately obvious what is wrong if they are used.

Doc functions shouldn't need a body at all.