On Thu, May 13, 2021 at 12:24:02PM +0200, Tim Düsterhus wrote: > > So what I'm proposing is to simply change vars_get_by_name() to call > > register_name() with alloc=0 in order to fix this mess. We can then > > check during 2.5 how to refine this to also consider the scope with > > the variable's name. It's just this, and fixes Joao's test case to > > always return 403: > > > > diff --git a/src/vars.c b/src/vars.c > > index 996141f5d..15dcb3c3d 100644 > > --- a/src/vars.c > > +++ b/src/vars.c > > @@ -583,7 +583,7 @@ int vars_get_by_name(const char *name, size_t len, > > struct sample *smp) > > enum vars_scope scope; > > /* Resolve name and scope. */ > > - name = register_name(name, len, &scope, 1, NULL); > > + name = register_name(name, len, &scope, 0, NULL); > > if (!name) > > return 0; > > Tim, do you agree with this analysis ? > > > > Yes, that change makes sense to me.
Great, thanks for the fast response, I'm going to do that and mark it as a bugfix so that after some observation we can consider backporting it. > If you'd see my full use case then I > recommend taking a look at haproxy-auth-request. It's super simple and even > comes with VTest tests: > > https://github.com/TimWolla/haproxy-auth-request#usage > https://github.com/TimWolla/haproxy-auth-request/blob/main/auth-request.lua#L50 > https://github.com/TimWolla/haproxy-auth-request/tree/main/test Thanks. I vaguely remembered it was something simple but I really can't remember every use case that end up as a patch :-) Cheers, Willy
