Anthony Williams wrote:
Thomas Witt <witt@acm.org> writes:
Any input on this one. Seems like too complicated a fix for 1.34.1
I believe the patch below will fix Johan's immediate problems; I would be grateful if he could try it.
There was a typo at (patched) line 82: "#if define(BOOST_HAS_WINTHREADS)" instead of "#if defined(BOOST_HAS_WINTHREADS)". IIRC we earlier discussed checking for a valid TSS index before calling TlsSetValue when in the cleanup_slots context. Further, I'd suggest statically initializing tss_data_native_key to 0xFFFFFFFF instead of having it zero initialized. The attached patch file (against RC_1_34_0) has these things fixed/added. Please review. I've applied the patch and successfully run all tests under libs/thread/test for current RC_1_34_0 under WinXP SP2 (SMP + HT). My own tests also keep running successfully with the patch applied. Even though the patch "couldn't possibly" have any effects under non-windows platforms I also applied the patch and ran the tests under OpenSUSE 10.2 using gcc 4.1.2. Adding a named constant instead of hardcoding 0xFFFFFFFF all over the place would also be great. However, as I believe this is the last Boost release containing the "original" thread code I guess the point is moot? [A bit off-topic] Is it intentional to have the 1.34.1 updates go under the RC_1_34_0 branch? Thanks / Johan