[kelbt-users] [feature proposal] autocommit mode

[Elijah Epifanov]

That issue stumped me..
Seems autocommit with hold was only possible when commit didn't flush
the whole stack.
BTW, old 'commit' was much more elegant and pure, but I understand
that it was an unnecessary difficult aspect of kelbt and no usages so
far.

2009/2/22 Adrian Thurston <thurston at complang.org>:
> What I mean is, how does the user decide on which productions to place a
> 'hold'?
>
> -Adrian
>
> Elijah Epifanov wrote:
>> I think the rule of least surprise wins here:
>>
>> In autocommit mode:
>> append `commit` to every production except those marked with `hold`.
>>
>> It's that simple... both for user and for kelbt.
>>
>> 2009/2/15 Adrian Thurston <thurston at complang.org>:
>>> But how do you choose the production(s) to place the hold on? That's the
>>> problem I'm seeing. Every choice point in a backtracking LR is covered
>>> by many productions above it and I think a hold would need to be placed
>>> on every one of those.
>>>
>>> Maybe we have different views of 'commit' at this point ... In kelbt a
>>> commit clears everything below it in the parse tree. Perhaps in the
>>> autocommit mode this should not be the case?
>>>
>>> -Adrian
>>>
>>> Elijah Epifanov wrote:
>>>> Hey, sorry for the long delay.
>>>>
>>>> I think it should be just a flag at the end of a rule (where commit is
>>>> implied in autocommit mode).
>>>> e.g.:
>>>>
>>>> autocommit;
>>>>
>>>> # Second level supports multiplication.
>>>> F: tok_number tok_mult F hold try { ... };
>>>>
>>>> I think hold is a correct antonym to commit in this case.
>>>>
>>>> 2009/1/25 Adrian Thurston <thurston at complang.org>:
>>>>> I'm curious if it would be hard to tell where to put the 'noncommit'
>>>>> flags. What do you think?
>>>>>
>>>>> -Adrian
>>>>>
>>>>> Elijah Epifanov wrote:
>>>>>> 2009/1/23 Adrian Thurston <thurston at complang.org>:
>>>>>>> So let me see if I understand ... autocommit would cause all alternative
>>>>>>> parses to be discarded on every reduction, with the option to disable it
>>>>>>> on specific reductions? This would make Kelbt behave like standard Yacc
>>>>>>> by default, with the kelbt behaviour turned on only where needed?
>>>>>> exactly, like any ordinary lalr parser
>>>>>>
>>>>>>> -Adrian
>>>>>>>
>>>>>>> Elijah Epifanov wrote:
>>>>>>>> There's no need to delay commit when parsing an unambiguous language,
>>>>>>>> therefore every rule will need (if efficiency is of concern) to commit
>>>>>>>> after reduce.
>>>>>>>>
>>>>>>>> It would be convenient to set an auto-commit mode on on per-grammar
>>>>>>>> basis and (rarely, if at all) disable it for specific portions when
>>>>>>>> needed.
>>>>>>>>
>>>>>>>> %%
>>>>>>>> # Second level supports multiplication.
>>>>>>>> F: tok_number tok_mult F commit try { ... };
>>>>>>>> F: tok_number commit try { ... );
>>>>>>>> };
>>>>>>>> %%
>>>>>>>> ->
>>>>>>>> %%
>>>>>>>> autocommit;
>>>>>>>>
>>>>>>>> # Second level supports multiplication.
>>>>>>>> F: tok_number tok_mult F try { ... };
>>>>>>>> F: tok_number try { ... };
>>>>>>>> %%
>>>>>>>>
>>>>>>>> _______________________________________________
>>>>>>>> kelbt-users mailing list
>>>>>>>> kelbt-users at complang.org
>>>>>>>> http://www.complang.org/mailman/listinfo/kelbt-users
>>>>>>> _______________________________________________
>>>>>>> kelbt-users mailing list
>>>>>>> kelbt-users at complang.org
>>>>>>> http://www.complang.org/mailman/listinfo/kelbt-users
>>>>>>>
>>>>>> _______________________________________________
>>>>>> kelbt-users mailing list
>>>>>> kelbt-users at complang.org
>>>>>> http://www.complang.org/mailman/listinfo/kelbt-users
>>>>> _______________________________________________
>>>>> kelbt-users mailing list
>>>>> kelbt-users at complang.org
>>>>> http://www.complang.org/mailman/listinfo/kelbt-users
>>>>>
>>>> _______________________________________________
>>>> kelbt-users mailing list
>>>> kelbt-users at complang.org
>>>> http://www.complang.org/mailman/listinfo/kelbt-users
>>> _______________________________________________
>>> kelbt-users mailing list
>>> kelbt-users at complang.org
>>> http://www.complang.org/mailman/listinfo/kelbt-users
>>>
>>
>> _______________________________________________
>> kelbt-users mailing list
>> kelbt-users at complang.org
>> http://www.complang.org/mailman/listinfo/kelbt-users
>
> _______________________________________________
> kelbt-users mailing list
> kelbt-users at complang.org
> http://www.complang.org/mailman/listinfo/kelbt-users
>