Re: [sqlite] Lemon: Non-terminal destructors and cleanup

[Ben]

On 18 Oct 2012, at 20:07, Richard Hipp  wrote:

> On Thu, Oct 18, 2012 at 3:03 PM, Ben  wrote:
> 
>> Hi list,
>> 
>> I'm having a little trouble getting my head around memory management
>> within a Lemon-generated parser. Specifically the part of the docs stating
>> when a destructor will or will not be called.
>> 
>> For example, this is a portion of a grammar based on the SQLite parse.y
>> file:
>> 
>> 
>> columnName ::= nm(N). {
>>BSSQLiteColumn *col = [[BSSQLiteColumn alloc] init];
>>col.name = N.textValue;
>>[[parsedTable columns] addObject:col];
>> }
>> 
>> nm(A) ::= id(X). { A = X; }
>> nm(A) ::= STRING(X). { A = X; }
>> id(A) ::= ID(X). { A = X; }
>> 
>> Notes:
>> - The token type here is a struct containing an Objective-C string which
>> needs freeing when done with.
>> - Only a %token_destructor is defined, not any others
>> 
>> 
>> I know that the last three assignments are leaking memory, but I don't
>> know when I should be explicitly freeing my allocated memory within a token
>> and when I should be relying on the destructor defined by
>> %token_destructor{}. Or for that matter whether I should be declaring a
>> more specific symbol destructor.
>> 
>> Can anyone shed some light on how this should be done?
>> 
> 
> If the nonterminal payload is passed into an action (as in your example
> where N is processed because of nm(N)) then Lemon assume that your code
> will free the content, if needed.
> 
> If the rule had been:  columnName ::= nm {...}   (without the (N) argument
> to nm) then the destructor would have been called.

Got it. I've added two release calls for X at the end of the C code blocks for 
these two:
nm(A) ::= STRING(X). { A = X; }
id(A) ::= ID(X). { A = X; }

and now it's working leak-free.

Thank you.

Ben


> The destructor is also called if nm is popped from the stack for any reason
> other than the columnName ::= nm rule, such as when the stack is popped
> during error recovery.
> 
> 
>> 
>> Thanks,
>> 
>> Ben
>> 
>> 
>> ___
>> sqlite-users mailing list
>> sqlite-users@sqlite.org
>> http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
>> 
> 
> 
> 
> -- 
> D. Richard Hipp
> d...@sqlite.org
> ___
> sqlite-users mailing list
> sqlite-users@sqlite.org
> http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users

___
sqlite-users mailing list
sqlite-users@sqlite.org
http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users

Re: [sqlite] Lemon: Non-terminal destructors and cleanup

[Richard Hipp]

On Thu, Oct 18, 2012 at 3:03 PM, Ben  wrote:

> Hi list,
>
> I'm having a little trouble getting my head around memory management
> within a Lemon-generated parser. Specifically the part of the docs stating
> when a destructor will or will not be called.
>
> For example, this is a portion of a grammar based on the SQLite parse.y
> file:
>
>
> columnName ::= nm(N). {
> BSSQLiteColumn *col = [[BSSQLiteColumn alloc] init];
> col.name = N.textValue;
> [[parsedTable columns] addObject:col];
> }
>
> nm(A) ::= id(X). { A = X; }
> nm(A) ::= STRING(X). { A = X; }
> id(A) ::= ID(X). { A = X; }
>
> Notes:
> - The token type here is a struct containing an Objective-C string which
> needs freeing when done with.
> - Only a %token_destructor is defined, not any others
>
>
> I know that the last three assignments are leaking memory, but I don't
> know when I should be explicitly freeing my allocated memory within a token
> and when I should be relying on the destructor defined by
> %token_destructor{}. Or for that matter whether I should be declaring a
> more specific symbol destructor.
>
> Can anyone shed some light on how this should be done?
>

If the nonterminal payload is passed into an action (as in your example
where N is processed because of nm(N)) then Lemon assume that your code
will free the content, if needed.

If the rule had been:  columnName ::= nm {...}   (without the (N) argument
to nm) then the destructor would have been called.

The destructor is also called if nm is popped from the stack for any reason
other than the columnName ::= nm rule, such as when the stack is popped
during error recovery.


>
> Thanks,
>
> Ben
>
>
> ___
> sqlite-users mailing list
> sqlite-users@sqlite.org
> http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
>



-- 
D. Richard Hipp
d...@sqlite.org
___
sqlite-users mailing list
sqlite-users@sqlite.org
http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users

[sqlite] Lemon: Non-terminal destructors and cleanup

[Ben]

Hi list,

I'm having a little trouble getting my head around memory management within a 
Lemon-generated parser. Specifically the part of the docs stating when a 
destructor will or will not be called.

For example, this is a portion of a grammar based on the SQLite parse.y file:


columnName ::= nm(N). {
BSSQLiteColumn *col = [[BSSQLiteColumn alloc] init];
col.name = N.textValue;
[[parsedTable columns] addObject:col];
}

nm(A) ::= id(X). { A = X; }
nm(A) ::= STRING(X). { A = X; }
id(A) ::= ID(X). { A = X; }

Notes:
- The token type here is a struct containing an Objective-C string which needs 
freeing when done with.
- Only a %token_destructor is defined, not any others


I know that the last three assignments are leaking memory, but I don't know 
when I should be explicitly freeing my allocated memory within a token and when 
I should be relying on the destructor defined by %token_destructor{}. Or for 
that matter whether I should be declaring a more specific symbol destructor.

Can anyone shed some light on how this should be done?

Thanks,

Ben


___
sqlite-users mailing list
sqlite-users@sqlite.org
http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users