Re: [PATCH] testsuite: Extend TLS core file testing with an OS-generated dump

["Maciej W. Rozycki" <macro@mips.com>]

On Thu, 24 May 2018, Pedro Alves wrote:

> >  I'd keep that:
> > 
> > rename tls_core_test ""
> > 
> > command that I added at the end though, so as not to clutter the procedure 
> > space.  Otherwise it'll stay there thoughout the rest of a test suite run 
> > (we have some leftover clutter in the testsuite already, which sometimes 
> > makes different .exp scripts interact with each other).
> 
> Ah, forgot to comment that.  I think that would be a losing battle.
> We already have too many tests defining procedures and global variables
> and we never took that care.  And frankly, it looks a bit too onerous
> to me.  Particularly since to do it throughout would imply undefining
> procedures at every early-bail-out return path too.
> 
> I think a better approach is to isolate the testcases somehow.
> 
> One way is to put each testcase in its own tcl namespace.  It still
> wouldn't isolate properly if you test with more than one board at the
> same time, like "--target_board=foo,bar".

 Given your observations I've thought about namespaces too, especially as 
they can be deleted once they're done with.  So why wouldn't they work 
with `--target_board=foo,bar', how is the test suite executed in such an 
arrangement?

> Another way is to make sure that we run each testcase in its own
> separate 'expect' process.  We actually can already do the latter,
> if you run the testsuite with "make check-parallel", like e.g.,:
> 
>  $ make check-parallel TESTS="gdb.*/tls-core.exp */break.exp"
> 
> so we'd need to make "make check" do that by default.  I think
> this might be the most promising approach.

 Are globals available to those separate processes though?  What if one is 
updated as a side-effect of a library procedure called from a test case, 
(like to achieve a caching effect)?

> > WARNING: can't generate a core file - core tests suppressed - check ulimit -c
> > UNSUPPORTED: gdb.threads/tls-core.exp: native: load core file
> > UNSUPPORTED: gdb.threads/tls-core.exp: native: print thread-local storage variable
> > 
> > here.  Will you handle all this or shall I?
> 
> I've handled this now, and pushed the patch, as below.

 Great, thanks!

> >  Many thanks for taking care of this.  Your updated test script actually 
> > helped me greatly with a test case for the next change I am going to push 
> > (another MIPS/Linux core file mishandling -- we live in a reality separate 
> > from the kernel's as it turns out).
> Nice, great to hear that.

 Pushed as commit d8dab6c3bbe6 ("MIPS/Linux: Correct o32 core file FGR 
interpretation") now.  I hope I didn't make the test case outdated from 
the beginning.

  Maciej