Re: rhl9 gdb tests -- June 24 2003

[Michael Elizabeth Chastain <mec at shout dot net>]

Hmmm, I think store.exp is running into some kind of UTF-8 style
problem, and it's a problem with the dejagnu version.

Here is what my log looks like (red hat linux 8):

  (gdb) PASS: gdb.base/store.exp: next char
  print l^M
  $1 = -1 'ÿ'^
  (gdb) PASS: gdb.base/store.exp: print old l - char

And here is your log (red hat linux 9):

  (gdb) PASS: gdb.base/store.exp: next char
  print l^M
  $1 = -1 'ERROR: Process no longer exists
  UNRESOLVED: gdb.base/store.exp: print old l - char
  ERROR: Couldn't send print r to GDB.
  UNRESOLVED: gdb.base/store.exp: print old r - char
  ERROR: Couldn't send set variable l = 4 to GDB.
  UNRESOLVED: gdb.base/store.exp: set variable l = 4
  ERROR: Couldn't send print l to GDB.

Obviously something died when it tried to print that character.

I don't think gdb died, because when gdb dies, the test framework
goes into timeout spasms, and this test completed in 5 seconds.
I think that Expect got hosed.

What happens with the stock FSF tcl+expect+dejagnu (the default case in
TestGdb.pl)?  Can you just do RUNTESTFLAGS=store.exp TestGdb.pl ?

Or you could back off and use the vendor tcl+expect+dejagnu
installed from RPM's.  Just cut down the path in TestGdb.pl.

I screwed up with BuildDejagnu.pl.  There are really two Dejagnu
projects in the world: the FSF cvs repository, and the sourceware cvs
repository (which includes tcl+expect+dejagnu in one package).
BuildDejagnu.pl actually is meant for the sourceware cvs version.
That's why it worked for me, but the paths inside it never worked for
you.

I will have to refactor this as BuildFsfDejagnu.pl and
BuildSwareDejagnu.pl, and always qualify the name 'Dejagnu', to avoid
this confusion.  I tried using plain 'Dejagnu' to mean "FSF Dejagnu' and
look where it got me.

Michael C