Results of macscp.exp test on cygwin

[Pierre Muller]

  Following a recent email about the testsuite that I 
ran on cygwin, where Pedro fond that the
number of failures was very high, I started to 
look into this. 

 Cygwin target uses stabs debugging format as a default,
but stabs doesn't generate any macro debugging information,
which results in 244 FAIL's (87 KFAIL plus 157 FAIL) from this unique test.

 Shouldn't we force dwarf-2 debug format, or at least 
skip the test if only stabs format is available?

 The test just add "-g3" option to the compilation command line
and that does not switch over to dwarf-2 debugging format.

Using 
make check RUNTESTFLAGS="gdb.base/macscp.exp CC_FOR_TARGET=\"gcc -gstabs+\""
I get 67 PASS, 87 KFAIL and 157 FAIL.

The strange thing is that even with
make check RUNTESTFLAGS="gdb.base/macscp.exp CC_FOR_TARGET=\"gcc
-gdwarf-2\""
(gcc version 3.4.4 (cygming special, gdc 0.12, using dmd 0.125))
I still get the same results :(

Only if I force the use of gcc version 4 (gcc version 4.3.2 20080827
(alpha-testing) 1 (GCC))
make check RUNTESTFLAGS="gdb.base/macscp.exp CC_FOR_TARGET=\"gcc-4
-gdwarf-2\""  
do I get good results:
PASSES: 286 FAIL:6 KFAIL:19
FAIL: gdb.base/macscp.exp: c99 varargs expansion
FAIL: gdb.base/macscp.exp: c99 varargs expansion without an argument
FAIL: gdb.base/macscp.exp: c99 varargs expansion, multiple formal arguments
FAIL: gdb.base/macscp.exp: gnu varargs expansion
FAIL: gdb.base/macscp.exp: gnu varargs expansion without an argument
FAIL: gdb.base/macscp.exp: gnu varargs expansion special splicing without an
argument
These failures are all due to the fact that stderr
is itself a macro in cygwin that is expanded into
"((__getreent())->_stderr)".

I also checked that gcc-4 with stabs, and I also get the 87 KFAIL and 157
FAIL.
 
  So all these tests seem to require gcc version 4, and explicit 
dwarf-2 selection.

  Is gcc version 3 still being developed?
If not, we could probably just fail generally once 
and skip the whole rest of the test if gcc version 3 is used.

  Should we add something to force use of 'dwarf-2' debug format?
Or at least also fail only once if stabs is used and skip the rest?


Pierre Muller
Pascal language support maintainer for GDB