Re: [RFA] testsuite/gdb.arch/i386-prologue.c: Fix compiling on Cygwin

[mec.gnu@mindspring.com (Michael Elizabeth Chastain)]

> the gdb.arch/i386-prologue.c testcase doesn't compile on Cygwin since
> on Cygwin the assenmbler labels need leading underscores to match the
> same names in C.  The below patch fixes that.

I acknowledge the problem, but I don't like to see #ifdef __CYGWIN__
in the middle of i386-prologue.c.  So I am declining this patch.

Can you add a layer to this?  Use a new symbol such as SYMBOL_PREFIX so
that i386-prologue.exp can tell i386-prologue.c whether to add a leading
underscore or not.

In i386-prologue.exp, something like this:

  # some targets have leading underscores on assembly symbols.
  # TODO: detect this automatically
  set additional_flags ""
  if { [istarget "i?86-*-cygwin*"] {
    set additional_flags "additional_flags=-DSYMBOL_PREFIX=_"
  }
  ...
  gdb_compile ... executable {debug $additional_flags} ...

It would be better to auto-detect the need for a leading underscore,
perhaps by compiling a test program.  But you don't need to do that
for this patch.

i386-prologue.exp would look like this.  First there is some
preprocessor gunk to add "..." around the value of SYMBOL_PREFIX:

  #define xstringify(arg) stringify(arg)
  #define stringify(arg) #arg

  /* add "..." around SYMBOL_PREFIX to make SP_STRING */
  #ifdef SYMBOL_PREFIX
  #define SP_STRING xstringify(SYMBOL_PREFIX)
  #else
  #define SP_STRING ""
  #endif

  ...

Then in the places where you want optional leading underscores, just add
SP_STRING to the big asm string.  The big asm string already depends on
string concatenation so this ought to work fine:

  "    .align 8\n"
  SP_STRING "gdb1253:\n"
  "    pushl %ebp\n"

I'm open to ideas if you have improvements or a different idea.  My goal
here is that i386-prologue.c does not have any target-specific tests in
it.

Michael C

===

	* gdb.arch/i386-prologue.c: Conditionalize label names to build
	on Cygwin.