Re: RFC: Some expect cleanups for the threads tests

[Daniel Jacobowitz <drow@mvista.com>]

On Sat, Jul 26, 2003 at 06:55:45PM -0400, Elena Zannoni wrote:
> Daniel Jacobowitz writes:
>  > These patches fix some testsuite problems on my setup - compiler and
>  > assembler both support TLS, but the C library doesn't.  I think you get a
>  > different error (earlier) if the kernel doesn't support TLS; I'll have to
>  > check at some point.
>  > 
>  > Elena, how do the two TLS changes look?  The print-threads.exp just handles
>  > the case where we weren't switching threads before reporting the SIGTRAP, so
>  > the zombie notice was never printed.  If you say "info threads" at that
>  > point GDB crashes.  I'll get that one later today.
>  > 
>  > -- 
>  > Daniel Jacobowitz
>  > MontaVista Software                         Debian GNU/Linux Developer
>  > 
>  > 2003-07-26  Daniel Jacobowitz  <drow@mvista.com>
>  > 
>  > 	* gdb.threads/print-threads.exp (test_all_threads): Loosen KFAIL
>  > 	pattern.
>  > 	* gdb.threads/tls-shared.exp: Don't use gdb_suppress_entire_file.
>  > 	* gdb.threads/tls.exp: Recognize one case of the host library not
>  > 	supporting TLS.
>  > 
>  > Index: gdb/testsuite/gdb.threads/print-threads.exp
>  > ===================================================================
>  > RCS file: /cvs/src/src/gdb/testsuite/gdb.threads/print-threads.exp,v
>  > retrieving revision 1.4
>  > diff -u -p -r1.4 print-threads.exp
>  > --- gdb/testsuite/gdb.threads/print-threads.exp	29 Jun 2003 19:29:42 -0000	1.4
>  > +++ gdb/testsuite/gdb.threads/print-threads.exp	26 Jul 2003 16:00:40 -0000
>  > @@ -94,7 +94,7 @@ proc test_all_threads { name kill } {
>  >  		fail "all threads ran once ($name) (total $i threads ran)"
>  >  	    }
>  >  	}
>  > -	-re "Program received signal SIGTRAP.*Thread \[0-9\]* \\(zombie\\).*$gdb_prompt $" {
>  > +	-re "Program received signal SIGTRAP.*(Thread \[0-9\]* \\(zombie\\)|0x00000000 in ).*$gdb_prompt $" {
>  >  	    if { $kill == 1 } {
>  >  		kfail "gdb/1265" "Running threads ($name) (zombie thread)"
>  >  	    } else {
>  > Index: gdb/testsuite/gdb.threads/tls-shared.exp
>  > ===================================================================
>  > RCS file: /cvs/src/src/gdb/testsuite/gdb.threads/tls-shared.exp,v
>  > retrieving revision 1.1
>  > diff -u -p -r1.1 tls-shared.exp
>  > --- gdb/testsuite/gdb.threads/tls-shared.exp	22 Jul 2003 15:49:45 -0000	1.1
>  > +++ gdb/testsuite/gdb.threads/tls-shared.exp	26 Jul 2003 16:00:40 -0000
>  > @@ -78,7 +78,7 @@ if { ($gcc_compiled 
>  >  }
>  >  
>  >  if {[gdb_compile_pthreads "${objdir}/${subdir}/${testfile}.o ${objdir}/${subdir}/${libfile}.so" "${binfile}" executable [list debug $additional_flags]] != ""} {
>  > -    gdb_suppress_entire_file "Testcase compile failed, so all tests in this file will automatically fail."
>  > +    return -1
>  >  }
>  >  
>  >  
> 
> Doh, of course.
> 
>  > Index: gdb/testsuite/gdb.threads/tls.exp
>  > ===================================================================
>  > RCS file: /cvs/src/src/gdb/testsuite/gdb.threads/tls.exp,v
>  > retrieving revision 1.2
>  > diff -u -p -r1.2 tls.exp
>  > --- gdb/testsuite/gdb.threads/tls.exp	23 Jul 2003 03:08:26 -0000	1.2
>  > +++ gdb/testsuite/gdb.threads/tls.exp	26 Jul 2003 16:00:40 -0000
>  > @@ -172,6 +172,10 @@ gdb_test "b [gdb_get_line_number "before
>  >  
>  >  send_gdb "continue\n"
>  >  gdb_expect {
>  > +    -re ".*Program received signal SIGSEGV.*a_thread_local = 0;.*$gdb_prompt $" {
>  > +        unsupported "continue to first thread: system does not support TLS"
>  > +        return -1
>  > +    }
>  >      -re ".*Program exited normally.*$gdb_prompt $" {
>  >          fail "continue to first thread: program runaway"
>  >      }
> 
> 
> Yuck. Is this the failure mode? Nothing detects this earlier? Sigh.

Yeah :((

If the kernel does not support TLS and glibc does, then the binary
won't start; you get an error about set_thread_area, I think.  But if
the kernel _does_ support TLS and glibc doesn't, nothing goes wrong
until we access a thread local.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer