RE: [office-formula] Formula Processing Model

[robert_weir@us.ibm.com]

I think you need to address some of these issues, but in Part 1.  As far 
as Part 2 goes, I'm not sure you need to say anything about dependencies 
or calculation chains.  As far as Part 2 is concerned a reference like 
[.A1] is merely an address can dereference to get a value.  The fact that 
deferencing that value may trigger all sorts of recalculations elsewhere 
in the spreadsheet is the nature of the spreadsheet (Part 1), but it is 
not intrinsic to the evaluation of a formula (Part 2);.

-Rob

"Dennis E. Hamilton" <dennis.hamilton@acm.org> wrote on 01/22/2010 
01:21:12 PM:

> 
> RE: [office-formula] Formula Processing Model
> 
> I happen to believe that we do need a processing model for a number of
> reasons.  One of them has to do with interdependent formulas in a
> (re-)calculation and the ways that data not provided directly in the 
formula
> itself is acquired.  One might not be able to answer all of the 
questions,
> but a processing model of some sort is needed so one can tell at least 
what
> the questions are and which ones are not being answered (or, preferably. 
are
> explicitly and identifiably implementation-defined/-dependent).
> 
> The interesting question is, what is the least one must constrain
> OpenFormula with normative language in order to have any prospect of
> interoperability without compelling out-of-band agreements among
> implementers (including use of a common code base) because the 
specification
> is inadequate on the subject.
> 
>  - Dennis
> 
> MORE ANALYSIS AND THOUGHTS
> 
> The interdependence with recalculation will arise, in the evaluator, 
when a
> cell or other value source referenced in the first formula is accessed 
and
> that source is determined by a formula.  Clearly, if there is no known
> value, there will be a (re-) calculation.  If there is a value that has 
been
> previously calculated, it is tricky whether recalculation should occur 
or
> not.  Clearly recalculation is not required if the result would not be
> impacted.  Presumably, recalculation will be done if the "recalculate" 
flag,
> the extra "=", is present.  Or maybe not: If that same formula is 
indirectly
> referenced more than once in evaluation of another formula, how many 
times
> does the recalculation happen?
> 
> Unfortunately, we don't have a way of saying "don't recalculate this." 
We
> only have a way of saying "must recalculate."  There is probably more 
that
> could be said on how even "must recalculate" is handled (as I just
> suggested) and what its value(s) is/(are) in these situations. 
> 
> There are complex interdependencies.  For example, one can, in expanding 
a
> set of calculations and the calculated results they depend on, arrive at 
a
> dependency on a formula whose recalculation is in progress.  Evaluators 
have
> ways of detecting such things and failing.  It helps that formulas tend 
to
> be free of side effects although there are issues when a formula depends 
on
> the same formula-determined value in more than one way, and when there 
are
> dependencies on external dynamic variables, as when using random 
numbers,
> NOW(), some external data access or user input, etc.
> 
> I recall evaluators that will iterate some number of times seeking a 
stable
> result.  (I haven't checked lately to see if current products have such
> features and whether they are user controllable.)
> 
> There is also interaction with processor setting on whether 
recalculation is
> done every time a formula is entered or a value that a formula depends 
on is
> altered.  Delayed recalculation can be more useful when recalculation is
> costly, but trouble-shooting and failure detection becomes far more 
complex,
> as is the evaluator determining an order to perform the recalculations 
in so
> everything works out and unbearable repetition of the same calculations 
is
> avoided. 
> 
> I'm not sure many cyclic-dependency failures ever show up in produced
> formulas these days.  When there is immediate calculation on entry or
> alteration of a formula in my simple-spreadsheet work, a cyclic 
dependency
> usually causes an interruption that an user must act on and it is not 
clear
> the offending formula (that is, the one that reveals the problem) is 
even
> allowed into the spreadsheet if the user fails to act and if failing to 
act
> is even an option.  Note that if formula evaluations are not immediately
> attempted or their dependencies assessed when entered by an user, this
> detection can be delayed until a recalculate point and be awful for an 
user
> to then figure out. 
> 
> There are more cases, and I am not sure it is always possible to prevent
> cycles and other inter-formula dependency cases (for example, if a cell
> reference is determined dynamically via functions that effectively 
calculate
> references or indexes).
> 
> Probably the worst case of delayed recalculation is on "save," 
especially
> when the user or the operating system is attempting to save and close 
the
> spreadsheet in a dangerous situation (imminent power failure is my 
nightmare
> case, something that plagues both laptop and desktop workers).  I'm not 
sure
> whether the persisted document (assuming work is not lost) can be 
assured to
> be in a clean, "conformant" state under such conditions.
> 
> 
> 
> -----Original Message-----
> From: Patrick Durusau [mailto:patrick@durusau.net] 
> http://lists.oasis-open.org/archives/office-formula/201001/msg00056.html
> Sent: Friday, January 22, 2010 04:39
> To: office-formula@lists.oasis-open.org
> Subject: Re: [office-formula] Formula Processing Model
> 
> Eike,
> 
> Eike Rathke wrote:
> http://lists.oasis-open.org/archives/office-formula/201001/msg00055.html
> > Hi Patrick,
> >
> >
> > 
> <snip>
> http://lists.oasis-open.org/archives/office-formula/201001/msg00054.html
> >> So, the lead in paragraph under General:
> >>
> >> "OpenFormula defines the semantics of expressions, when 
recalcalculation
> 
> >> occurs, and limits on formulas. It does not specify any particular 
> >> process for evaluations consistent with those definitions."
> >> 
> >
> > Actually we do not define when recalculation occurs, with the one
> > exception of formulas tagged for forced recalculation. It is up to the
> > implementation to decide when a formula or a part thereof is to be
> > recalculated.
> >
> > 
> I was re-casting:
> 
> "This section describes the basic formula processing model: how 
> expressions are calculated, when recalculation occurs, and limits on 
> formulas."
> 
> But after re-reading the section on recalculation, it is just advisory 
> and not normative. Yes?
> 
> Do we need to reduce and keep as a note, perhaps with formulas tagged 
> for forced recalculation?
> 
> The opening line then becomes:
> 
> "OpenFormula defines the semantics of expressions and limits on those 
> expressions. It does not specify any particular process for evaluations 
> consistent with those definitions."
> 
> Thinking that saying "formulas" isn't any different from "expression." 
Yes?
> 
> Hope you are looking forward to a great weekend!
> 
> Patrick
> 
> -- 
> Patrick Durusau
> patrick@durusau.net
> Chair, V1 - US TAG to JTC 1/SC 34
> Convener, JTC 1/SC 34/WG 3 (Topic Maps)
> Editor, OpenDocument Format TC (OASIS), Project Editor ISO/IEC 26300
> Co-Editor, ISO/IEC 13250-1, 13250-5 (Topic Maps)
> 
> 
> ---------------------------------------------------------------------
> To unsubscribe from this mail list, you must leave the OASIS TC that
> generates this mail.  Follow this link to all your TCs in OASIS at:
> https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php 
> 
> 
> ---------------------------------------------------------------------
> To unsubscribe from this mail list, you must leave the OASIS TC that
> generates this mail.  Follow this link to all your TCs in OASIS at:
> https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php 
>