Updates about Enhanced Database Ranges.

Markus Mohrhard markus.mohrhard at googlemail.com
Mon Jun 3 12:45:18 PDT 2013

Hey Akash,

2013/6/2 Akash Shetye <shetyeakash at gmail.com>:
> Hey devs,
> Thank you moggi and kohei for helping me out on the IRC constantly.
> With the design I am describing now, the feature becomes an n-ternating (not
> alternating) row formatting (not just colouring) feature. Please help with
> your advice and discussions.
> The feature design can be compartmentalised as follows:
> 1. The 'Row Formatting definition' part, where used selects the targeted
> database range and the styles to be applied to it.
> 2. The 'Row Formatting core' where DB Range code and this feature interact.
> 3. The rendering end.

After reading this I think we need to split the task into smaller and
easier to handle steps. Hopefully it helps us to focus on the
important tasks for now. I'll explain some of the problems with your
proposal below but the general point is that we should work on the
model right now. Don't think about anything redering related, updating
references or similar stuff until we have implemented the model and
maybe have the import and export implemented. The visual
representation of the feature is something that we can later add but
is of less importance than you think right now.

> Row Formatting definition:
> [Data Structures]
> About the storage of this formatting information and how alternating
> formatting info will be defined:
> User defines each alternating format as a band, and as such 'n' number of
> bands constitute particular n-ternating format. The user select the
> formatting options and the size of the band i.e, how many consecutive rows
> it will cover/applied to. When the user defines another band, the new band
> will follow the previously defined band. Thus we can have more than just
> alternating colours for our DB ranges. Eg. define first band of size 1 and
> bold text, band 2 of size 3 italic text and band 3 of size 1 underlined
> text; for a range of 10 rows, the first will be bold followed by 3 italic
> rows followed by 1 underlined row and so on..
> Every defined band for an Alternating row format will be stored as a
> ScDBDataRowFormat that will have the attributes: 1. index number which helps
> define the sequence of these bands 2. band size 3. Formatting information
> (ScStyleSheet instance??). It has its method GetStyle() that returns the
> style to be applied

Let us limit ourself for now to 2 alternating styles. Extending this
later is easier than trying to implement everything at once.

> Many such ScDBDataRowFormat instances are added to a container which is a
> part of ScDBDataRowFormatting object. This object represents the total style
> you have created and also contains the defined style name you give to it.
> All such defined styles are added to a pool of ScDBDataRowFormatting
> objects; lets say this pool is represented by
> ScDBDataRowFormattingCollection (the names are getting scarily long), so
> styles once custom defined can be used again and again.
> Associating every ScDBData object with it's ScDBDataRowFormatting object:
> Every ScDBData object has an attribute of type ScDBDataRowFormatting which
> is assigned to it by picking it up from the ScDBDataRowFormattingCollection
> pool.
> Thus many DB Ranges can use the same style without having to have any
> multiple copies to themselves.

That design is overcomplicated and lacks some of the important
features. We need at least the following features: two column styles,
two row styles, information about header row, total row, column
labels, ... Please have a look at Excel to see some of the information
we need to store in data structures. A good starting point for the
design is to list all the information we need to store for each table.
Starting with the list we will then decide for each information how to
represent it and how they interact and influence each other.

Just a short note: When you think about styles in Calc's core
implementation please always think only about the style name.
Everything that is hidden behind ScDocument should at least for you
only use the style name and never the ScStyleSheet object. If there is
ever a case where we need access to the full style we will need to
look closer into it.

> Row Formatting Core:
> The row formatting core is composed of the modifications needed in ScDBData
> itself to support the new alternating colour functionality.
> ScDBData will have a new method GetRowFormatting() just like the one in
> conditio.cxx (GetData()); however this will return a ScDBDataRowFormatting
> object.
> The rendering half will have to iterate over each row of the the range while
> iterating over the ScDBRowFormat items in the container in ScDBRowFormatting
> instance taking care of the band size of each and thus applying it to the
> row cells accordingly. This makes the rendering code a little more
> complicated as multiple styles have to be related to different rows.
> Events (not actually events) in ScDBData such as:
> UpdateMoveTab - changes the table
> UpdateReference()
> ExtendDataArea - extends/shrinks the area under ScDBData

Don't worry about them right now. We will come back to them when we
have implemented all the data structures and the import/export.

> make changes to the structure and coverage of the defined database range. We
> need to take care of such events that may wipe out or add newer rows to the
> range. Such updates set the bModified bool in ScDBData, this needs to be
> known and be handled by the rendering half of this feature.
> The rendering end:
> This rendering end is the something I am discovering with each passing day.
> Not much to write here.
> Some design issues:
> I see one big problem with the way this are laid out now. The formatting
> information and ScDBData's bounds (i.e its rows) are very decoupled. It
> becomes the job of the rendering end to apply the correct formatting by
> iterating over the ScDBRowFormatting container and getting the ScDBRowFormat
> objects taking care of their band sizes as it moves over each row adding in
> the formatting information.
> This makes the rendering end very complicated (maybe). Since rendering is
> not just taking the concerned rows and style and shoving in the attributes.
> One way out can be having a member function in ScDBData that returns an
> array of  size[No.OfRows], each entry containing the applicable stylesheet
> for that row. This shifts the 'making sense of' part, out of the rendering
> part.

Don't think about the rendering right now. This will be much simpler
than you expect and all most of the logic that you think belongs into
the rendering part actually is part of the mode and we will need it
also for other parts of the mode implementation. In the end we will
just have a method that returns the style name for a cell and not much
more. But please don't worry about that right now because in the end
you'll see that this will look just natural after all other parts are

> [UI Stuff]
> Speaking of what to call this functionality and where to place it in the
> menu hierarchy:
> Either we can:

That is even one step past the implementation of the rendering. We'll
work with the UX team when we are ready to come up with a mock-up for
the UI parts.

> Greet the user with the same dialog as seen in Calc's Data->Select Range,
> from where the user selects a range to be formatted as table i.e, coloured
> alternately.
> Or, since clicking format as table option in MS Excel marks a bunch of cells
> as a table with a name, which 'optionally' can be coloured; we will need to
> incorporate the alternate colouring option in the Data -> Define Range
> dialog in its 'more' section.
> In any case once this option is clicked, the user must define a custom
> design or choose from a few existing ones, just like in Excel.

So let us split the task into several smaller steps that we tackle and
where you should not worry about the next step.

1.) Model data structures
2.) Import/export (OOXML and ODF)
3.) Rendering and user interaction
4.) UI

So for now we concentrate on how to implement all the model data
structures. Mainly how we can represent all the information we need in
calc's core implementation and let them work together. As first step
please list all the information we need to store in the model and we
will discuss on IRC how to represent them.


More information about the LibreOffice mailing list