cayenne-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Aristedes Maniatis <>
Subject Re: split expressions
Date Thu, 14 Jan 2010 09:40:56 GMT
On 14/01/10 8:10 PM, Andrus Adamchik wrote:
> On Jan 13, 2010, at 2:38 PM, Aristedes Maniatis wrote:
>> I've tried to explain the current usage better in the docs, given what
>> I now understand. Please let me know if I misunderstood something. I
>> had originally thought they were there as shortcuts to a path string.
> Actually we may start using them that purpose as well in 3.1 when and if
> we merge SelectQuery and EJBQLQuery into a new SuperSelectQuery. Now I
> think I remember that a possibility of unifying expressions with EJBQL
> was considered when designing the API for splits. Hence aliases were
> suggested.
>> A way think of aliases is that they are markers within part of an
>> expression which give Cayenne hints about merging. Like a query cache
>> key, they are important only if you use the same key twice. But
>> because aliases are created when you create the Expression path
>> string, and then expanded when you construct the query, they can be
>> awkward to use. It would be nicer if the Expression carried not only
>> the idea of the key, but also how it expands and passed that onto the
>> query.
>> Expression e1 ="path1|",
>> "foo");
>> Expression e2 ="path2|",
>> "bar");
>> This puts the definition of the alias and the expansion in the same
>> logical place. It would also allow
>> Expression e2 ="path2|", "bar");
>> to use the alias previously defined. Or to have Cayenne construct an
>> arbitrary random, non-reusable alias:
>> Expression e1 ="|", "foo");
>> Then you could even define an expression
>> Expression e1 = Expression.fromString("| = foo
>> and | = bar")
> So the name before the pipe is an alias, right? Then how do we create
> splits in the middle of the path?


You could even this which uses an existing alias:


or this which makes a named alias:


But the syntax is prone to confusion because it is hard to read. On the plus side though,
I can't imagine why anyone would want to use named aliases in this way, so they are likely
to be rare.

>  I guess we can have a more complex
> notation:

This is a long way from obvious for the most common use of the syntax: combining AND queries
across a to-many join. Sure named aliases might be useful in themselves, maybe, for someone.
I can't personally think of a reason. But the following seems clearer to me:

   Artist.PAINTINGS.join(Painting.EXHIBITS, Expression.SPLIT_PATH).join(Exhibit.GALLERIES).like("Y")

My overall preference is to make the expression functions slightly more SQL-like, with words
like join and like. Avoiding abstract ideas like 'alias' where they aren't relevant to the
user is part of that same goal. In fact 'alias' has a specific SQL common meaning (table aliases)
which is only going to confuse new users.

If you really want to name an alias:

   Artist.PAINTINGS.join(Painting.EXHIBITS, "aliasX").join(Exhibit.GALLERIES).like("Y")

> or setting the path, specifying the number of segments:
>"X", 2);

Maybe as an option, but I'd never use it. Consider the method:

performAQuery(Expression e) {
   e1 = e.and(somequalifier);
   performMyQuery(e1.alias("X", 2); // what did that just do? How long was the path expression
passed to the method?

> (s/dot/join/ or whatever)
> or the old way:
> // this will probably require defining a special Expression subclass -
> KeyValueCondition or something
> // that ensures there's only a single path inside
>"", "foo").split("X", 2);

Just throw a runtime exception...

Or if you really wanted to, promote 'path expression' from being just a String to having its
own class. Then alias is just a special case of Path. Come to think of it, what is the difference
between your new Key typesafe class and Path, other than a few dots...

> So yeah, we are definitely making progress. One remaining case that is
> still not accommodated is when subexpressions are created "in the
> vacuum" with no knowledge of the overall query context. So we need to
> take an existing (possibly nested) conditional expression and setup
> correct splits for one or more of its paths. I guess some expression
> transformer is needed here.

My guess is that 99% of use cases would be met with a parameter to pass into the query which
splits ALL paths. It would actually solve every instance of why we started this conversation
in the first splace.


Aristedes Maniatis
GPG fingerprint CBFB 84B4 738D 4E87 5E5C  5EFA EF6A 7D2E 3E49 102A

View raw message