Description of problem:
The behaviour of pthread_detach has changed from previous versions, and seems
wrong. (Without the posix standard, which I don't have, I can't say for
certain). If the thread being detached has already exited, then pthread_detach
fails with ESRCH, where it used to succeed (and succeeds on Solaris). 

Version-Release number of selected component (if applicable):
glibc-2.3.2-11.9

How reproducible:
Every time.

Steps to Reproduce:
uudecode, gunzip, compile, and run the following code

begin 644 threadtest.c.gz
M'XL("(W3D3X``W1H<F5A9'1E<W0N8P"%4DUOPC`,O?=7>""F!'43VY&._0*T
MVVZ34-2F-%))D..B21/_?4X3"F43Y!`YSQ_OV?'4V++M*@UO>VI0J^JY><^F
M`]A9X^D*8\"X`&4'9RJ8U\Z):"G<2OC)@,\>C:5:3-@)GA32EYW(HG?Y5NN]
M>$FOR\#*67V.0TT=6OCX7*^+[)C8:F5:P1F@$1WF4#KK"<I&(<QW?J"O4UG6
MRH$Y3$)>AWHY\\M9Q3?3Y,`)^:F0)^PMT=\R:=#?AL13T,H"`NU.&=OS<Z]E
MGHA#XX>A\SC'C2+"#8V>Q2A@\$4X%$7MNY:*+`(UB`C`:ES56%;U>`E)>%C!
MXB2AGT`85$SG]O^DG\=\O,%6\DU:#$7#.?'F_Z(]P]C%/SL&PH^.`)F-K7M=
F1%E7#<2U>I6WIE=I4F4CTE/>GU=,N&)*B[D(2_$+01;,?D`#````
`
end

   
Actual results:
$ cc     threadtest.c  -lpthread -o threadtest
$ ./threadtest
foo start
foo done
failure:pthread_detach:3:No such process

Expected results:
$ cc     threadtest.c  -lpthread -o threadtest
$ ./threadtest
foo start
foo done

Additional info:
A cursory look at the code from the nptl in the glibc source RPM suggests that
the tid field in the pthread structure is set to zero when the thread
exits, and pthread_detach doesn't handle that situation. (Joining the thread
after it exits works fine, and pthread_join uses a different test which differs
only by the tid==0 case).