As a note to myself, pasting long pieces of a convo with Steve Ebersole:

(03:27:56 PM) sebersole: some.joined.path.separated.by.dots
(03:28:01 PM) sebersole: thats a implcit join
(03:28:17 PM) sebersole: as opposed to the explicit form
(03:28:31 PM) sebersole: from ... join ... join ...
(03:29:05 PM) sebersole: this is where the parser is not very smart
(03:29:13 PM) sebersole: the new parser is much better at this
(03:29:40 PM) sebersole: really the rule is that it is ok to reuse non-leaf parts of implicit join paths always
(03:30:04 PM) sebersole: for leaf parts it depends on a few things
(03:30:20 PM) sebersole: the old parser does not understand all that
(03:30:59 PM) sebersole: it does not understand that it has seen part of prod.supplier.supplierName before
(03:31:57 PM) sebersole: in general it just elects to be safe when it does not know
(04:09:11 PM) sebersole: select prod.supplier from Products prod order by prod.supplier.supplierName
(04:09:17 PM) sebersole: so thats your query
(04:09:37 PM) sebersole: but imagine...
(04:11:01 PM) sebersole: well first let me back up...
(04:11:17 PM) sebersole: the concern is filtering results that should not be filtered
(04:12:41 PM) sebersole: its hard to make up these models on the fly to explain by way of an example
(04:13:26 PM) sebersole: product -> supplier -> industry lets say
(04:15:06 PM) sebersole: so first look at:
(04:16:25 PM) sebersole: ... from Product p where p.supplier.industry.code = 'R&D' or p.supplier.hq.city = 'Austin'
(04:17:23 PM) sebersole: so the issue is that inner joins have a semantic of removing missing correlations
(04:18:03 PM) sebersole: here, if a product's supplier does not have a hq-address or does not have an industry... should they still be returned?
(04:18:31 PM) sebersole: say for example, a product from the supplier Acme that has no industry, but is in Austin
(04:18:39 PM) sebersole: should they get returned?
(04:19:36 PM) sebersole: as far as the alias part...
(04:19:42 PM) sebersole: there is another forms here..
(04:20:48 PM) sebersole: from Product p join p.supplier.industry i join p.supplier.hq h where i.code = 'R&D' or h.city = 'Austin'
(04:21:17 PM) sebersole: but really where the alias part in my above commments come into play is in cases like:
(04:22:32 PM) sebersole: select p.supplier.supplierName from Product p join p.supplier.industry i1 join p.supplier.industry i2 ...
(04:23:19 PM) sebersole: its complicated, but the idea is that the 'p.supplier' join is probably safe to reuse
(04:23:42 PM) sebersole: any dereference to 'p.supplier.industry; is not
(04:23:51 PM) sebersole: any dereference to 'p.supplier.industry' is likely not
(04:24:11 PM) sebersole: and even if it were safe... which woould you use?
(04:24:27 PM) sebersole: so "safe" is releative to the path
(04:24:44 PM) sebersole: aka, these aliases implicit joins are named
(04:24:52 PM) sebersole: wow
(04:24:56 PM) sebersole: aka, these aliased implicit joins are named
(04:25:18 PM) sebersole: and so should not be reused
(04:25:22 PM) sebersole: thats the safest
(04:25:59 PM) sebersole: actually the new parser looks at each "path part" separate and decides
(04:26:21 PM) sebersole: starting a new Join branch whenever it deems it needs to
(04:26:27 PM) sebersole: so part could be reused
(04:26:38 PM) sebersole: i think the current parser is not that smart
(04:26:43 PM) sebersole: i think its all/none to it
(04:27:12 PM) sebersole: aka, either the full p.supplier.industry path
(04:27:28 PM) sebersole: aka, either the full p.supplier.industry path can be reused, or none of it (no part of it) can