On Wed, 2023-03-01 at 12:16 +0100, Shengyu Huang wrote:
> Hi Dave,
>
> > On 1 Mar 2023, at 00:59, David Malcolm <dmalc...@redhat.com> wrote:
> >
> > Did you get it to output your messages?
> >
>
>
> Yes, I chose to emit the warning before the supergraph or exploded
> graph is created (I guess this is enough, right?). I checked out from
> the trunk a week ago, and I checked out from the latest trunk just
> now and built from modified source again, by adding a line in the
> following code in analyzer/engine.cc:
>
> FOR_EACH_FUNCTION_WITH_GIMPLE_BODY (node) {
> node->get_untransformed_body ();
> warning_at (DECL_SOURCE_LOCATION (node->decl), 0, "hello world, I’m
> compiling %qE", node->decl); // ADDED
> }
>
> Compiling my own test script without optimizations, I got the output
> (surprisingly no warning from -Wanalyzer-shift-count-negative
> anymore):
>
> test.c: In function 'main':
> test.c:42:9: warning: left shift count is negative [-Wshift-count-
> negative]
> 42 | b = b << -1;
> | ^~
> test.c: At top level:
> test.c:36:5: warning: hello world, I'm compiling 'main'
> 36 | int main()
> | ^~~~
> test.c:27:6: warning: hello world, I'm compiling 're'
> 27 | void re (int c)
> | ^~
> test.c:12:6: warning: hello world, I'm compiling 'f'
> 12 | void f (unsigned long *p, int r, int i)
> | ^
> test.c:9:5: warning: hello world, I'm compiling 'fun2'
> 9 | int fun2()
> | ^~~~
> test.c:4:5: warning: hello world, I'm compiling 'fun1'
> 4 | int fun1()
> | ^~~~
Looks great.
[...snip...]
>
>
>
> >
> > The next thing to do might be to try stepping through the code in
> > the
> > debugger; that's often a good way to learn about a new codebase.
> > See:
> > https://gcc-newbies-guide.readthedocs.io/en/latest/debugging.html
> > and maybe have a look at the support scripts mentioned on that
> > page.
> >
>
> I did try to use gdb more to inspect the internals, but one thing I
> noticed when using it is that I got `??()` in the backtrace, which
> I’ve never seen before. Some online sources say it happened due to
> “corrupted stack”, but I don’t know how that can happen
> either…However, after pulling changes from the trunk and rebuilding
> from the source, “??()” disappeared and now I can step through the
> execution without any problem (previously `step` and `continue` did
> not work as expected…). Do you have any clues what happened so that I
> can fix it myself later if that happens again?
I've noticed that if I invoke "make" in the top-level build directory,
when it recurses into the "gcc" subdirectory it builds with -O2,
whereas if I invoke "make" directly in the build tree's gcc
subdirectory it builds without optimization, and I get a much better
debugging experience. Maybe that's what happened? Because of this, I
sometimes find I have to do "rm analyzer/*.o" in the "gcc" subdirectory
of the build, and rerun "make" in there. Though even with -O2 my gdb
still gives me function names in the backtrace, albeit on x86_64, and
probably a different version of gdb...
>
> Best,
> Shengyu
>
> > BTW, are you building trunk, or GCC 12? I've made a *lot* of
> > changes
> > to the analyzer in trunk, so it would be good for you to be working
> > with something that's reasonably up-to-date.
Sounds like you're pulling from trunk; good.
Thanks
Dave
