make Thread.sm_this a plain TLS value

[MartinNowak]

• this became possible, because the TLS emulation
  no longer uses the Thread.getThis() instance
• Windows use impersonate_thread to set sm_this
  when attaching a foreign thread

[braddr]

This pull request seems to have broken osx 32 and 64. It was pulled before the platform completed (because it was hung in the first druntime tests). Consequently, the master branch pulls are also hung on the same tests. It looks like infinite recursion.. which ought to lead to an eventual fault, but...

(gdb) bt
#0 0x000e993a in D2rt4deh213__eh_finddataFPvPyS2rt4deh29FuncTablePyS2rt4deh29FuncTableZPyS2rt4deh29FuncTable ()
#1 0x000e98f1 in D2rt4deh213__eh_finddataFPvZPyS2rt4deh29FuncTable17__foreachbody3418MFKS2rt12sections_osx12SectionGroupZi ()
#2 0x000efbb9 in D2rt12sections_osx12SectionGroup7opApplyFMDFKS2rt12sections_osx12SectionGroupZiZi ()
#3 0x000e9898 in D2rt4deh213__eh_finddataFPvZPyS2rt4deh29FuncTable ()
#4 0x000e99ec in _d_throwc ()
#5 0x0008967a in onAssertError ()
#6 0x000e9e66 in _d_assertm ()
#7 0x00079d99 in D2rt12sections_osx8__assertFiZv ()
#8 0x000effb0 in __tls_get_addr ()
#9 0x000e9a9e in _d_throwc ()
#10 0x0008967a in onAssertError ()
#11 0x000e9e66 in _d_assertm ()
#12 0x00079d99 in D2rt12sections_osx8__assertFiZv ()
...

[alexrp]

@braddr @AndrejMitrovic's revert is in now, sorry about the mess

[braddr]

No worries.. these sorts of failures are just rare enough that I haven't taught the auto tester how to watch for hung tests and deal with them yet. If only there were more hours in the day. Anyone wanna take a stab at adding it?

[MartinNowak]

Anyone wanna take a stab at adding it?

Yep, I'll put it on my task list. This one here https://github.com/braddr/d-tester?

[braddr]

yes.. we can discuss the details more, but not in this pull request.