db-derby-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Rick Hillegas <Richard.Hille...@Sun.COM>
Subject Re: R: Using derby to parse an SQL statement
Date Fri, 14 Nov 2008 14:15:29 GMT
Hi Flavio,

I don't think that we have a good primer on the AST nodes. However, you 
can get some sense of what the nodes mean by building the Derby javadoc 
and looking at the javadoc for the package 

Now that you've gotten over the hard hurdle of building the Derby 
classes, building the javadoc is easy:

  ant -quiet javadoc

This will build the javadoc into the directory javadoc/engine. Once you 
have built the engine javadoc, browse to the 
org.apache.derby.impl.sql.compile package. The header comments on the 
classes are usually pretty helpful. I also recommend taking a look at 
the tree view of that package. You will see that the AST nodes are all 
of the classes indented under QueryTreeNode.

Understanding the AST graph takes some patience. There are two useful 
techniques for figuring out how the nodes snap together into a graph:

1) The nodes themselves implement 
org.apache.derby.iapi.sql.compile.Visitable. That interface has one 
method, accept(). Not all AST nodes directly implement accept() so you 
may have to inspect their superclasses. The accept() method shows you 
what each node thinks its subnodes are. You can write your own tool to 
explore the graph by coding your implementation of the graph walker, 

2) The nodes also implement a pretty-printing method, treePrint(). The 
treePrint() method is another example of what a node thinks its subnodes 
are. This is the method which ASTParser calls on the top node.

That should get you started. The devil is in the details, but keep 
posting questions and I think you'll get to the bottom of it.

Hope this helps,

Flavio Palumbo wrote:
> Hi guys,
> I followed this post finding it very exciting, cause I post a similar
> question a while ago.
> Now I'd like to test the work submitted by Rick but I'm not so inside Derby
> to catch he whole job I have to do.
> I understood that I have to :
> - download Derby sources (from where ?)
> - apply the patch suggested by Rich (seen the classes but not catch the
> lines of code)
> - recompile Derby (what libraries/jars I need to)
> - use ASTParser as suggested
> Can somebody give me a hint 
> Thanks a lot
> Flavio
>> -----Messaggio originale-----
>> Da: news [mailto:news@ger.gmane.org]Per conto di Christian Riedel
>> Inviato: giovedì 13 novembre 2008 6.37
>> A: derby-user@db.apache.org
>> Oggetto: Re: Using derby to parse an SQL statement
>> Hi Rick,
>> at first thank you very much for your efforts so far. At a first glance
>> your changes to the code seem to be exactly what we want. I will try if
>> it works asap.
>> To test it with the derby libs I'd have to work on the current trunk and
>> apply the patch an then compile derby manually, right?
>> I'll keep you updated
>> Thanks for your help
>> Christian
>> Rick Hillegas schrieb:
>>> Hi Christian,
>>> I have created a JIRA to track this issue:
>>> https://issues.apache.org/jira/browse/DERBY-3946
>>> I have attached to the JIRA a small patch which exposes the AST
>> produced
>>> by the parser. I have also attached a simple program, ASTParser, which
>>> shows how to retrieve the AST from Derby. I am inclined to check this
>>> patch in to the trunk. Please let me know if you find this
>> useful and if
>>> you would like me to port this patch to another Derby branch.
>>> Hope this helps,
>>> -Rick
>>> Christian Riedel wrote:
>>>> Hi Rick,
>>>> first of all thanks for your answer ... now the relations have become
>>>> a lot clearer ...
>>>> Your are right, there is a lot of things to be done that we probably
>>>> don't want to go through. You asked why we cannot take the whole derby
>>>> engine and use it ... well there is nor real reason not to do so. The
>>>> only "problem" I see is, that derby is a dbms - if I am not mistaken -
>>>> and we only have an SQL statement that we extract from a text file and
>>>> want tot analyze it to extract some metadata from it.
>>>> So if we take the derby engine as it is, how can I prevent that we
>>>> have to set up a "dummy" DB in order to be able to actually use thje
>>>> parsing feature ....
>>>> I hope you see my point.
>>>> We could live with setting up a dummy DB ... and I do think that the
>>>> derby AST offers all information we need. It's just that I don't see
>>>> how we can set this thing up. So having a dummy DB is necessary to be
>>>> able to intercept the parsing process to get hold of the AST? Can we
>>>> actually access the AST if we choose to set up a dummy DB? I think
>>>> that would be something we could live with ;-)
>>>> Thanks for your support
>>>> Christian
>>>> Rick Hillegas schrieb:
>>>>> Hi Christian,
>>>>> I think you will have difficulty isolating the Parser from the rest
>>>>> of the SQL interpreter. In theory, you should be able to isolate the
>>>>> compiler from the execution engine and the storage layer--but that is
>>>>> an untested theory.
>>>>> The Parser wants to turn out abstract syntax trees (AST). Ideally,
>>>>> the Parser would just need to ask a NodeFactory for AST nodes and you
>>>>> could supply your own NodeFactory. But I think that there is a fair
>>>>> amount of coupling between the Parser and Derby's concrete
>>>>> implementation of NodeFactory. I think that you could uncouple the
>>>>> two, but you may not want to spend your time on that.
>>>>> So the Parser is going to force you to pull in the AST nodes. Once
>>>>> you do that, you will end up with the whole compiler. In particular,
>>>>> the AST nodes (and the Parser itself) expect that you will supply an
>>>>> implementation of LanguageConnectionContext, the master state
>>>>> variable for the whole SQL interpreter. Untangling that requirement
>>>>> is another chunk of work you may not want to do.
>>>>> Then there is the Monitor. It has been a while since I was in that
>>>>> code but I seem to recall that fairly early on the Monitor wants to
>>>>> fault in a storage layer. In theory you ought to be able to supply
>>>>> the Monitor a list of modules that doesn't include a storage layer.
>>>>> But since no-one runs in this configuration, there are probably a lot
>>>>> of undocumented surprises that you may not want to fix either.
>>>>> Can I ask you what breaks if you just pull in the whole Derby engine?
>>>>> Are you concerned that you will fault in too much code that you
>>>>> barely use? Are you concerned that you'll end up with a dummy
>>>>> database that you don't need? Are Derby's AST nodes not a usable
>>>>> representation of statement syntax?
>>>>> Thanks,
>>>>> -Rick
>>>>> Christian Riedel wrote:
>>>>>> Hi there,
>>>>>> we are working on a small project where we need to analyze an SQL
>>>>>> statement that can be of any kind: very simple, with inner selects,
>>>>>> complex join etc.
>>>>>> We figured it inappropriate to start to write our own parser when
>>>>>> there are other projects, like derby, out there that can do it much
>>>>>> better than we would possibly do ... so this was our idea:
>>>>>> Can we use derby to create an instance of Parser
>>>>>> (org.apache.derby.iapi.sql.compile.Parser.class) and let our SQL
>>>>>> statement be parsed by calling the parse() method on this instance?
>>>>>> What we want to have is a syntax tree of the statement that allows
>>>>>> us to see which tables and which fields are accessed / included in
>>>>>> the statement (including any possibly done "renames" á la SELECT
>>>>>> street AS "ADDRESS" FROM USER_DATA ).
>>>>>> The problem is, that we are stuck ... we spent several days now to
>>>>>> try to find the proper way to create an instance of the Parser. Is
>>>>>> it possible at all without having to set up a running derby system?
>>>>>> Is the Monitor class the right entry point? How can we create a
>>>>>> CompilerContext so that a Parser instance can be created?
>>>>>> This sure is off-topic but we don't see any way through all this.
>>>>>> Can you help us?
>>>>>> Thanks in advance
>>>>>> Christian
>> --
>> To reply to this posting directly use the following address and
>> remove the 'NO-SPAM' part: Riedel.Christian.NO-SPAM@gmx.net
> -----------------------------------------------------------
> Il presente messaggio non costituisce un impegno contrattuale tra SILMA S.r.l. ed il
> Le opinioni ivi espresse sono quelle dell'autore.
> SILMA S.r.l. non assume alcuna responsabilita riguardo al contenuto del presente messaggio.
> Il messaggio è destinato esclusivamente al destinatario.
> Il contenuto e gli allegati sono da considerarsi di natura confidenziale
> Nel caso abbiate ricevuto il presente messaggio per errore siete pregati di comunicarlo
> alla casella segreteria@silmasoftware.com.

View raw message