On Thu, Apr 12, 2018 at 3:06 PM, Willy Tarreau <w...@1wt.eu> wrote: > Hi Ben, > > On Thu, Apr 12, 2018 at 02:25:58PM -0600, Ben Draut wrote: > > This changes the parser to run section postparsers once per section > > instance, rather than only when the section type changes. > > > > This is motivated by the work discussed in > > https://www.mail-archive.com/haproxy@formilux.org/msg29527.html. It > should > > make it easy to produce the warning mentioned in the summary at most once > > per section. > > Hmmm I don't much like this principle, because it's not a post-section > parser anymore in this case, it's a post config parser. Note that it is > very possible that you need a different type of post-processing to help > with the resolvers stuff, but I really think this approach is mistaken > because it removes the ability to perform post-section processing. From > what I'm seeing by quickly grepping in the code, at least the cache > relies on this. >
Ah, I missed the cache usage, thanks for pointing that out. > I *think* we do have a way to register a post-config call, but to be > honest I really don't remember, as we've added a few such hooks over > time. Maybe you'd need to call this thing before entering > check_config_validity() which resolves every cross-dependency between > the config elements and which detect inconsistencies. We may imagine > having a pre_final_check and a post_final_check callback for this maybe. Great, I'll dig some more and give it some more thought. Thanks!