Re: [RFC] gdb.base/float.exp and gdb.base/commands.exp patch

[Daniel Jacobowitz <drow@false.org>]

On Wed, Mar 02, 2005 at 11:17:15PM -0600, Manoj Iyer wrote:
> > What compiler are you testing with?
> 
> I am using GCC (gcc version 3.4.3 20041212 (Red Hat 3.4.3-9.EL4))
> 
> 
> > On what line is it reporting that
> > it has left the block?  I'd like to understand the difference before we
> > change this.
> 
> I believe the line number is 82 in my case, and the testcase had 57.
> 
> Ok here is the piece of output...
> 
> --------------- paste -------------
> Continuing.^M
> Watchpoint 11: local_var^M
> ^M
> Old value = 0^M
> New value = 1^M
> factorial (value=1) at ./gdb.base/run.c:81^M
> 81          return (value);^M
> $38 = 1^M
> ^M
> Watchpoint 11 deleted because the program has left the block in^M
> which its expression is valid.^M
> 0x0000000010000604 in factorial (value=511) at ./gdb.base/run.c:82^M
> 82      }^M
> $39 = 511^M
> 1^M
> ^M
> Program exited normally.^M
> (gdb) FAIL: gdb.base/commands.exp: continue with watch
> ------------- end paste -------------------------

Accepting any line number is not OK.  Accepting this particular line
number does seem plausible.  Can you do that instead?

> > What about PowerPC targets which don't have an FPU?  Hmm, it looks like
> > GDB more or less assumes the FP is present.  Not sure about SPE though.
> >
> 
> Yes I did think about ppc nofpu system, I believe they are mostly used in
> embedded devices, but I could be wrong. So what is the probability of
> coming accross one that someone will run this test on? But there should be

I've done it in the past.  It's not that hard to get ahold of one. 
But, as I said, I do not know what will be output.

Might as well do this for now and fix it up later.

> Also, fileio.exp does some permission checks, ie read/write to a file
> which has no read/write permissions, but if you run as root the test will
> fail. There should be some mechanism by which the testcases should check
> if user id is root, if it is temporarily become another user like "nobody"
> or something like that. Couple of tests failed because of this, but I dont
> know how to handle this.

Or just don't run the tests as root.  Seems like a good policy anyway.

-- 
Daniel Jacobowitz
CodeSourcery, LLC