[Scilab-Dev] SEP #40: Cell Arrays
Eric Dubois
grocer.toolbox at gmail.com
Thu Feb 18 20:41:40 CET 2016
Hello
I am inclined to share Samuel point of view: this is a compliocation than
could be avoided.
But I cannot resist noting that the annoucement is 6 years older than the
announcement of the weapon of mass destruction that consist in changing the
behaviour of the addition of a matrix with a null matrix.
Sorry for insisting, but I will again call for the removal from the final
Scilab 6.0 release of this planned change.
First, I am not convinced at all by the argument put forward that it will
make Scilab more consistent with other language such as Matlab, Octave,
Julia.. : after all, every language has its indiosycrasies; a Matlab user
will yet have to adapt herslef to this change, bu along many other ones;
and I,do not think that changing this behavour will convice any Matlab user
to switch to Scilab nor prevent anyone thinking about switching to give up
because of this beahviour.
Second The argument that it enhances Scilab internal consistency is a
little bit more compelling, but not much: after all, addition and
subtraction are different operations from multiplication and division, such
one can justify different behaviour. And there are cases when it make tho
code more compact.
Adnd lastly both arguments are anyway swept away by the simple fact that
the change should make all previous code unreliable: you cannot be sure in
advance that the working of your program has not been affected by the
change (and the warning that is designed to alert to the user is not
sufficient: a warning can easily be missed, especially for second hand
users not so famaliar with Scilab and if it is hidden among many other
warnings).
I hope that the Scilab will come to its senses and give up making this
change.
Regards.
Éric.
2016-02-18 18:45 GMT+01:00 Samuel Gougeon <sgougeon at free.fr>:
> Hello,
>
> Le 29/03/2010 11:40, Bruno JOFRET a écrit :
>
> Hi,
>
> Please find attached SEP #40 talking about Cell Arrays.
> This SEP is targeting Scilab 6.0 as we will introduce a new syntax for
> cells.
>
> Any feedbacks are welcome.
>
>
> 6 years after, it is very (very) late to comment or make any suggestion.
> Yet, Scilab 6.0 is not yet released, so i hope that this won't be too late
> for the essential:
> Indexation (extraction and insertion) with {} should really not be
> introduced.
>
> This message comes after first comments and rational recently posted there:
>
> http://mailinglists.scilab.org/Scilab-users-a-i-j-specific-extraction-syntax-with-should-not-be-introduced-tp4033484.html
>
> As another way to explain why {} extractors and inserters must not be
> introduced,
> we may follow the pages of the current version of the SEP available here:
> http://gitweb.scilab.org/?p=scilab.git;a=blob;f=SEP/SEP_040_cell_arrays.odt
>
> - After a cell array *c = { %pi %i %t ; %z "abc" list(1,%s)}* has been
> (very friendly) built with the new {} heterogeneous concatenator:
>
> - the *extraction of the content of any single cell component* should
> be direct:
> - with Scilab < 6 : c(5).entries or c(1,3).entries
> - with Scilab >=6: c(5) or c(1,3)
>
> - with Scilab < 6 : c(5) or c(1,3)
> - with Scilab >=6: {c(5)} or {c(1,3)}
> Whenever any wrapped-in-cell answer would be needed, it still
> can be obtain by packaging the answer, instead of implementing a
> "dereferencing" way to address a content through a very specific {}
> extractors and inserters when unwrapped values are required.
>
> - As well, *the insertion into a single component must be
> direct, and any type of data should be accepted*:
> - Scilab < 6: c(5).entries = ["Hi" "Hello"] or c(1,3).entries =
> ["Hi" "Hello"]
> - Scilab >=6: c(5) = ["Hi" "Hello"] or c(1,3) = ["Hi" "Hello"]
> So, page 4:
> - c = cell([4,3,2]); for i = 1:24, c*{i}* = i, end
> becomes
> - c = cell([4,3,2]); for i = 1:24, c*(i)* = i, end
>
> - Is there any reason to not address cells components simply
> as we do for matrix components, directly with (i,j,..)? I do not find any.
> The .entries addressing was needed due to the encoding of cells as
> mlists. But what could motivate keeping any intermediate level to access to
> the values of data, for extraction as well as for insertion? I do not see a
> single reason.
> As noted here-above, from the fact that the LHS object is a cell
> array, any data type can be accepted and inserted, without any prior
> packaging of the RHS as a cell array. The wrapping in cell must be done
> internally by the insertion process.
> - Then, Scilab 6: c(5) = {"abcd"} will insert a true elementary
> cell as the c(1,3) c's component, not the string "abcd". This is a
> straightforward and very clear syntax. What is in RHS parameter is just
> values that are inserted in the array, *as is*.
>
> - *The SEP does not present **insertion and extraction of multiple
> components in a once*.
> After still *c = { %pi %i %t ; %z "abc" list(1,%s)}, *the current
> implementation is the following:
>
> - *multiple insertion*: *a) of corresponding multiple components*:
> - c(:,1) = { %e "zz" }
> assigns %e to c(1,1) and "zz" to c(2,1) *in a distributive way*!
> 1. This kind of distributive assignment is a very great new
> feature!!
> 2. The assignment is transparently done using *linearized
> indices*. unmatching sizes/formats of the recipient and of
> the source is smoothly handled. Here, a row of cells feeds a column of
> cells. This is nice as is! This behavior could also be implemented with
> other types of RHS containers, at least for a list. So
>
> - c(:,1) = list(%e, "zz")
> should do the same. But it does not:
> --> c(:,1) = list(%e, "zz")
> Wrong insertion: A Cell expected: use {...} instead of (...).
> This feature might be implemented later. This is not so urgent
> as removing the {} addressing.
>
> If this feature is implemented, how will it be possible to
> insert a list in a single component?
> - c(3) = list(list(%t)) // will do it. Or if the size of the
> list is not 1, even
> - c(3) = list(%t, %z) // mismatch could be handled softly
> in a comprehensive way
>
> - c([1 2]) = { %e "zz" } does the same using a vector of
> linearized indices. Great!
>
> - *multiple insertion*: *b) of a single component to be
> replicated*:
> - c([1 4]) = {"abc"} inserts the same "abc" string at the 1st and
> 4th positions in c. This is great! The only thing is that the syntax should
> become simply
> c([1 4]) = "abc"
> In the final implementation that we suggest and hope, c([1 4]) =
> {"abc"} will be as well possible but will insert the cell {"abc"} instead
> of the string "abc" at the desired positions.
>
> - *multiple extraction:*
> - c(1,:) returns {%pi %i %t}: this is great! By default, this
> can't be anything else than a cell array.
> No c{1,:} syntax is required
> - c(:,3) returns {%t ; list(1,%s)}: still great and expected!
> No c{3,:} syntax is required
> - c(1:2,[1 3]) returns {%pi %t ; %z list(1,%s)} as expected.
> No c{1:2,[1 3]} syntax is required
> - with a linearized index: c([2 5 3]) returns {%z ; %t ; %i}
> column cell, as with matrices addressed with a linearized index a column is
> returned: Great!
> No c{[2 5 3]} syntax is required
>
> - Finally, *what about conversions between a cell array and a
> list? *We think that this kind of conversion between these 2 types of
> containers should be available in Scilab.
> - *list => cell* :
> - If it becomes possible to feed a cell array (or subarray) with a
> list as discussed above, then this kind of conversion won't need anything
> else. We will just have to do:
> c = cell(2,3); c(:) = list(%pi, %z, %i, "abc", %t, list(1,%s))
>
> - Otherwise: *makecell() *should be kept and renamed
> *list2cell()*, instead of being removed. It already works as
> expected :
> L = list(%pi, %z, %i, "abc", %t, list(1,%s));
> --> makecell([2 3], L(:))
> ans =
> [1x1 constant] [1x1 polynomial] [1x1 constant]
> [1x1 string ] [1x1 boolean ] [ list ]
>
> - *cell => list:*
> - The present special extraction with {:} does it, but this syntax
> must be removed. Keeping it only for that is meaningless:
> --> typeof(c{:})
> ans =
> list
> - A new *cell2list()* converter should rather be implemented.
>
> Hoping that this will convince you to remove the {} complicated addressing
> and the related data wrapping,
>
> Best regards
> Samuel Gougeon
>
>
> _______________________________________________
> dev mailing list
> dev at lists.scilab.org
> http://lists.scilab.org/mailman/listinfo/dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.scilab.org/pipermail/dev/attachments/20160218/31b74d87/attachment.htm>
More information about the dev
mailing list