Re: [patch, testsuite] Clean up gdb.trace results

[Simon Marchi <simon dot marchi at polymtl dot ca>]

On 2018-10-06 21:09, Sandra Loosemore wrote:
> Continuing through my pile of testsuite problems found while working
> on cleaning up nios2 test results....
>
> I found that the results for the gdb.trace tests were really messy on
> targets that don't support the trace features.  The trouble affected
> various other embedded targets I tried, not just nios2.  Specific
> problems included:
>
> * The header file trace-common.h that is used by most of the test
> programs in this directory causes multiple compilation errors (not
> just the obvious preprocessor #error) on architectures it doesn't know
> about, and the checks in the .exp files for successful compilation of
> the test case didn't distinguish between failures for missing arch
> support (should have "unsupported" status) and failures due to unknown
> causes (should be "untested").  My understanding is that adding arch
> support to trace-common.h won't do anything useful without
> implementing the underlying fast tracepoint support too.
>
> * Several test cases depend on linking with/loading a shared library
> for in-process trace without checking for shared library support on
> the target.
>
> * The signal.exp test case didn't test for signal support.
>
> * The tspeed.c test case was failing to compile due to a missing
> #include, and its .exp file did not test for runtime target support in
> gdb.
>
> Anyway....  here's an attempt to fix all this.  I added an early test
> for arch support to the .exp files so that most tests report
> "unsupported" immediately when they cannot possibly work, and fixed up
> the other problems as needed.
>
> OK to check in?
>
> -Sandra

Hi Sandra,

Thanks for this, fixes to make the testsuite run more cleanly is always 
appreciated.

I just did a quick smoke check, and noticed some errors when running on 
x86:

  Running 
/home/emaisin/src/binutils-gdb/gdb/testsuite/gdb.trace/signal.exp ...
  ERROR: no fileid for elxacz23q12

(elxacz23q12 is my hostname)

This looks like a check done a bit too early.  Can you make sure that 
these tests:

$ make check TESTS="gdb.trace/*.exp"
$ make check TESTS="gdb.trace/*.exp" 
RUNTESTFLAGS="--target_board=native-gdbserver"
$ make check TESTS="gdb.trace/*.exp" 
RUNTESTFLAGS="--target_board=native-extended-gdbserver"

Don't have significant differences before/after your patch (on 
x86/Linux)?  If you don't see any difference, maybe the problem is on my 
side, then I'll look into it.

Simon