2002)

23 December 2002

The memory consumption seems almost stable now...
Tried to draw something with Virtual Pen. The missing functions such as duplicate, mirror, rotate, etc. would be very useful even for simple drawings.
Stumbled on a bug (not identified yet) with the undo/redo chain. When doing a PurgeUndo, VP crashes mysteriously. And when doing/undoing/doing/PurgeRedo, I get tons of asserts (because of the fork ?).

22 December 2002

Searching the reason of the wasted MB of RAM...
Found a first culprit : VP::Render::GdiFigRegion was not being cleaned up. It was therefore leaking GDI+ regions and paths every time an object was drawn.

21 December 2002

Fixed small alignment problem when dragging figures by their outline or by their surface when a magnetic grid is active.
The Bézier curve edition and creation classes now support the different constrain modes (symmetry, locked angle, free).
Simply moving the mouse around some time eats MB of memory. Why ?

20 December 2002

Fixed VP.FrontEnd.Controllers.DocView to avoid handling of ghost button release events. These events are generated when the user dismisses a menu by clicking outside of it. The button press event is consumed by the menu, but not the release event.
Fixed problem in Magic library, used for the pop-up menus... Method InternalTrackPopup of class VP.Magic.Menus.PopupMenu did not propagate the WM_MOUSELEAVE event, and this caused a lot of trouble with the MouseLeave/MouseEnter .NET events, especially for the document view, which thought it was active all the time...
After a pop-up is displayed and dismissed, the click is sent to the correct window, but since no mouse move event was sent before it, the position of the click is wrong. This required the generation of a dummy mouse move event.
These little annoyances took about 10 hours to be fixed.
Ready to add the sink commands into Bézier curve plug-in.

19 December 2002

Working on the Bézier creator.
Added commands to query the points while an object is being created.
Added commands TestObjectIsClosedFigure and QueryPointPos needed by the Bézier curve creator and editor front end classes.
Cleaned-up old references to A-ClosePath.
The Bézier curve is now functional. It is not yet possible to change the constrain mode of the primary grips, but otherwise, everything seems OK.
Started work on a contextual menu which would be displayed whenever the user right-clicks on a grip. What seemed simple is going to be tricky, since I have observed strange behaviours with the mouse cursor...

18 December 2002

Method VP::Render::Context::DrawHairLine accepts an alpha value (to display the tangent control segments of the Bezier curve).
Updated the document object multiple point creator in order to work with the Bézier curve too.
Updated the Bézier core plug-in and fixed a few details.
Fixed VP::Render::Context::TestPointInText which did not properly calculate the position of the cursor in ligatures. The current implementation is better, but still not perfect, since it assumes that the ligature is composed of n equally wide sub-glyphs.
Added information about raw cursor position in the cursor event class, which is required by the text tool.
Added a hint in the main tool cursor class, which specifies whether a ghost cursor is allowed or not. Having a ghost cursor with the text tool is rather annoying !
Fixed figure/grip detection when using grid.
The BelowCursorFinder class now properly handles out-of-view events.
Working on the Bézier creator.

17 December 2002

Implemented the settings dialog for the grid.
Worked on the Bezier curve implementation. This required work on other classes in the front end.

16 December 2002

Bugs never end... this could be last week's moto. While fixing the problem found late yesterday evening, I stumbled on several other problems. And I guess that the first beta testers will also find lots of bugs elsewhere. My, my...
Fixed VP::DocText::Meta::CollapseFontState which did not remove the correct number of bytes when a font reset tag cancelled another tag. Also cleanup any dangling format tags.
Made class VP::DocText::Meta methods ClearBreakTags, OptimiseBreakTags and GenerateBreaks independent of the cursor's order.
Added better debugging output in VP::DocText::Meta::DebugDump.
Switched to the front-end... and added support for a new kind of plug-ins, called a manager, which has no user interface, but needs to communicate with the document view.
In the document view controller, the cursor position used for detection is always the raw one, not the filtered one.
Added support for paint event registration and exported a few properties needed by the Grid plug-in in class VP.FrontEnd.Controllers.DocView.
Implemented a magnetic grid (no user interface yet, however).

15 December 2002 (Sunday)

More bug hunting...
Fixed bug in cursor backward movement (character by character). This was the same problem as the one fixed the day before for the forward movement.
Fixed VP::DocText::Meta::MoveBackwardsInText which did not move consistenly across the tags, depending on the position of the cursor in a skip tag.
Fixed small bug in VP::DocText::Meta::CreateTempCursorB, which caused VP to raise a fatal error in some cases (trying to skip past the terminal fence tag).
Improved method VP::TextLib::Scribe::GlyphLine::GetCaret so that the caller can provide its own version of the font record. This is needed, since only the text flow has enough information to know if the caret should be displayed slanted, if the cursor lies between an empty <I></I> sequence (no information is stored in the scribe for empty runs).
Fixed cursor maintainance done by VP::DocText::Meta::NotifyCollapse (if the cursor was pointing into the first skip tag of a collapsing tags collection, then its skip tag offset was not reset).
Fixed VP::DocText::Meta::CollapseFontState to work in multiple line paragraphs and if the document is completely empty.
When a selection gets erased, the active cursor is now systematically placed inside the selection before doing any text removal (see right example in image below, where after deletion of the word "how", the typed text should be formatted in italic too).
Improved ligature suppression : this is now only done in justified paragraphs.
The I-beam tool handles the ESC key : when it is pressed, the main window is activated first, then the selection is discarded.
Added support for font and paragraph attribute resetting : clicking on an already active button in the tool palette's text tab can be used to apply/reset the corresponding attribute. This works only when a selection is active.
Fixed a bug related to the use of VP::DocText::Meta::CollapseFontState, which did not properly update the text flow after a change in the tags stream, which could lead to line nodes pointing at the wrong place in the text.
Selections, even spanning several paragraphs, can now be changed with the ApplyFontState command. This had far reaching ramifications...
There is still a little omission in the CollapseFontState method, which does not remove the trailing formatting tags of a paragraph, but I am too tired to fix it this evening.

14 December 2002 (Saturday)

Removed the Cursor::xxxAutoSync mechanisms.
Bug hunting.
Added the CollapseTextState command, to explicitely force the removal of useless formatting tags.
Fixed bug in cursor advance (character by character), when reaching the end of a line.
There can be trouble when calling some methods of VP::DocText::Meta, which rely on a correctly fitted text (tags with the TAG_SKIP_BREAK_ACTIVE bit set) to detect the line boundaries. By swapping two method calls in command InsertText, this problem could be avoided. But there might be other cases !
Improved VP::DocText::State::Serialiser::SerialiseState in order to replace tags which would simply return to the style's default state with their FORMAT_FontResetTag tags. This is used by VP::DocText::Meta::InsertFontState.
Fixed logic of VP::DocText::Meta::CollapseFontState.
Fixed logic of VP::DocText::Meta::RemoveText, in order to keep the font formatting tags. This avoids any problems in following cases (deletion of italic style begin; inconsistent deletion of italic style begin or end) :

Other small fixes.

13 December 2002

Improving the font format tag handling.
When using text insertion commands or cursor movement commands, the affected paragraph will be analysed and all duplicate/useless formatting tags will be removed.
Modified VP::DocText::Meta::MoveToLineBeginInText to provide an easy means of finding the real beginning of a line in the tags stream too.
Added VP::DocText::Meta::NotifySkipTagChange for use by VP::DocText::Meta::OptimiseSkipTags and SplitSkipTag, which allows to work on the text and tags streams without ever desynchronising the cursor records. This paves the way to the removal of the Cursor::xxxAutoSync mechanisms.
Implemented VP::DocText::Meta::CollapseFontState to remove the duplicate/useless formatting tags.

12 December 2002

Fixed baseline alignment for the characters. The call to ExtTextOut uses the mode specified by SetTextAlign to position its output. TA_BASELINE does what we need.
Fixed highlighting gaps in the text. Because of some rounding errors, white hairlines could show up between adjacent highlighted characters (in VP::TextLib::Scribe::LineStore::PaintLine).
Fixed H1/H2 adjustment happening when a line has not enough vertical room on the first fitter iteration.
The caller to VP::TextLib::Fitter::Context::ExtractWord can now specify if he wants the method to consider the state changes, or not. The state changes must be ignored if the caller is just interested in the words, but not the geometry.
Added command QueryTextSelectionLength.
The I-beam cursor tool recognises if the activated flow had a selection by executing the QueryTextSelectionLength command.
Spent a lot of time trying to debug the UxTheme wrapper DLL on Windows 98.
Fixed the fitter for the cases where the line fit got restarted. Several local variables were not reinitialised, which caused the fitter to misbehave with respect to end of lines.

11 December 2002

Continuing work on text style edition.
When a paragraph break is inserted in the middle of a paragraph, VP creates a new paragraph with exactly the same settings as the previous paragraph. This copies both the paragraph style and all the modified paragraph formatting tags.
The ApplyParaState command is now applied to the whole selection.
Added a new FormatDef tag, FORMAT_FontResetTag, which can be used to reset a tag's value to its default value.
Fixed VP::DocText::State::StreamParser::ExtractFormatText, which did not properly extract the text if used in the non embedded mode.
Added command ApplyFontState.
Implemented VP::DocText::Meta::InsertFontState, which changes the font formatting locally. This caused a few display problems to become visible (regular and bold fonts, for instance, do not share the same baseline).
Fixed the text fitter in order to properly handle font format changes in words (the font change was considered as a text end in the word extraction code).
Some work for Crésus Gestion PE today, too.

10 December 2002

Working for Crésus Gestion PE today.

9 December 2002

Continuing work on text style edition.
In class VP.Forms.Cmd.Sink.MessageFilter, added an event which is fired whenever the number of keyboard consumers changes, so we can disable the blinking text caret when the user is typing into a text box or combo box...
When clicking in a text flow, update the associated text styles.
Fixed icon update on font style changes.
First draft implementation of command ApplyParaState.
Fixed bug in VP::DocText::Accessor::Synchronise which moved to the wrong place.
Added command QueryStateInText which is used to determine the exact state at the specified cursor position in the text stream.
Class VP.FrontEnd.GUI.PaletteTabText now uses the real paragraph state, not just the state defined by the active style (it queries the exact state with the QueryStateInText command).
Fixed the forward/backward navigation in the text stream. Don't ever get stuck between the paragraph style format tag (FORMAT_ParaStyle) and other paragraph formatting tags (FORMAT_... belonging to the FORMAT_CAT_PARAGRAPH category).
The reflection in .NET is great to build methods such as VP.FrontEnd.Logic.TextState.Delta. It does all its calculations without knowning anything about the fields represented in the text state, just by peeking and poking through the class public methods.

7-8 December 2002

Week-end. Some work for Cresus PE. No work on VP.

6 December 2002

Implemented and tested command CreateTextFlow. The current implementation of the text fitter does not support multiple flows. This implies that I have to rework the fitter... and maybe the scribe too.
After surprisingly few changes, VP now supports several text flows without any trouble.
Fixed a hyphenation problem in VP::DocText::Meta::AnalyseTextBreaks with the last word in a text stream.
Improved VP.FrontEnd.Logic.MainToolCursorIBeam to create the text flow as required when attaching to an object which has no text yet.

5 December 2002

Added new text manipulation commands : QueryTextLength, CreateTextFlow, DeleteTextFlow, AttachFrameTextFlow, DetachFrameTextFlow, QueryFrameTextFlow.
Renamed DeleteText into RemoveTextBackwards.
The bug hunting was successful :

Added tests to the text style manipulation commands to avoid having them crash VP if there is no active style.
The caret won't be displayed if it is not active.
The selected object will be actively displayed even if the text tab is shown in the palette.
Fixed hyphenated line width calculation : the hyphenated line length did not take in account the hyphenated word, which caused VP to choose quite bad hyphenation points.
Fixed identifcal node optimisation : two nodes which point at the same place in the text stream are considered to be identical only if they will graphically appear at the same position.
The break points displayed by command DebugBreakTags will be properly displayed even if the window has been scrolled.

4 December 2002

Continuing work on text formatting GUI logic.
Fixed several minor bugs caused by abusive cyclic events in the text formatting GUI.
Added support for the optical names in the text formatting GUI (this is fully transparent to the user).
Improved blinking caret : when text is being displayed, the VP wastes much less CPU cycles.
Fixed cursor movement bug when a line ended with spaces : moving forwards would detect the end-of-line when reaching the first space, which is correct, but the MoveTextCursor command expected that by skipping a single character, the cursor would reach the next line. This is not the case.
EPSITEC meeting this afternoon.
Bug hunting.

3 December 2002

Working on text formatting. The PaletteTabText class is approaching an almost usable status.
The font name analysis stumbled on some fonts where the base name included the "Light" suffix, just as it would normally include an optical suffix (such as "Display"). I added a list of exceptions. Maybe I should do it the other way : define the optical suffixes and only recognise those.
Added a mode field to the text style, which allows to mark some styles as non delete-able.
The text state serialiser accepts a prefix of the fields to be serialised.
Fixed the QueryFamilyStyles and FindFamilyStylesByName commands.
Made the tool panel by default transparent, which produces much nicer looking icon bars in the palette.

2 December 2002

Added command for font enumeration (doc.QueryFontList). This command required a few changes in TextLib's font information classes, the most important being the support for the optical name of a font.
Added a font information class in the front end : VP.FrontEnd.Logic.FontInfo. This class can extract all pertinent information from the font information provided by the core and organise it into family name, style name and optical name (for instance "Warnock Pro", "Regular", "Display").
Improved class VP.Forms.ComboBox in order to be able to generate events when the highlighted line in the dropped down list changes. This could be useful to build a preview of the selected font.
Improved VP.Unit.Value parsing and pretty printing.
Class VP.FrontEnd.Logic.TextState can serialise its state to a string builder.
Changed the font leading into a paragraph leading, which is not as intuitive (the leading applies to the font height, but in fact it is a property set for a whole paragraph).

1 December 2002

Improved core event notification in the text related commands.
Added preliminary support for the text state representation in the front end (class VP.FrontEnd.Logic.TextState).
Added a lot of widgets into the text tab. They do not do anything yet.

30 November 2002

Implemented text style usage counter and text style naming. The usage counter is maintained automatically by class VP::DocText::Meta when insertions and deletions are done in the text and tags streams.
Implemented a set of commands used to manipulate styles and text styles, which are needed by the front end in order to build the text GUI. Also added new commands to list named styles in given families.
Improved the text support in class VP::DocObj::Label. Maybe we should switch to UTF-8 encoding ?
Introduced new core events (Text_TextChanged and Text_StyleChanged).
Working on text GUI.

29 November 2002

No real work on Virtual Pen today.

28 November 2002

Improved and finished serialisation and deserialisation code for the textual representation of the state record in VP::DocText::State.
EPSITEC meeting.

26-27 November 2002

Microsoft DevDays 2002 at Zurich Oerlikon. I did some refactoring while travelling and during the most annoying conferences :

Command parser accepts quoted strings.
Added FORMAT_TextStyle and appropriate handling in VP::DocText::State.
Improved the serialisation and deserialisation code in VP::DocText::State (it supports both the tags stream and a textual representation).

25 November 2002

Fixed several small repainting bugs caused by the tool palette refactoring.
Added a splash screen.
Started refactoring of the VP::DocText::State class, in order to generalise the state serialisation and state deserialisation code; we need to be able to exchange the internal state representation with the higher level .NET front-end...

23-24 November 2002

Week-end.

22 November 2002

Implementing the text user interface.
Refactored the VP.FrontEnd.GUI.SamplePalette class into a VP.FrontEnd.GUI.Palette class. Extracted the attribute specific code to class VP.FrontEnd.GUI.PaletteTabFigure.
Started implementation of the VP.FrontEnd.GUI.PaletteTabText class.

21 Novembre 2002

Design for the text user interface.

20 November 2002

Still not feeling 100% fit.
Added the doc.DebugTextBreaks command, which shows all possible text breaks in a text flow. This is very useful to understand how the paragraph fitter decides to break its lines. Should have added this much sooner...
Fixed left aligned lines which did not get compressed when displayed.
Fixed glyph ligature stretching : when a line gets stretched, the fitter no longer replaces glyphs with their matching ligature.

19 November 2002

Sick today...
SQL and SQL meeting at David's with Pascal Zweilin, Alex Maillefer, Daniel Roux, Denis Dumoulin, Michael Walz, Yves Raboud and David Besuchet.

18 November 2002

Sick today...
Tested InstallShield software, which could be used to build an installer for Virtual Pen. However, this product seems to be too rich for my needs (and too expensive, too).
Tested the PreInstaller (which is freeware), but it is only available in German.
Finally, worked on Microsoft's sample bootstrapper and got a working installer, which sets up .NET before starting the installation, if needed.

9 -17 November 2002

No work. I had a break thanks to the Swiss Army...

8 November 2002

Today is a special day : "Fête des 25 ans d'EPSITEC" ...
Added text deselection through Escape key.
Added zoom in and zoom out buttons and menu commands.
Added new attribute : VP::DocAttr::ImageFrame is the equivalent to TextFrame for images.
Added new tool to add images into arbitrary figures (VP.FrontEnd.Logic.MainToolCursorImage); this tool creates an a.imageframe attribute block which describes the image (the block contains a Label object for the file name).
Implemented a cache for the image frame, in order to speed up the display. We should probably cache the stretched image to further improve the performance.

7 November 2002

Cleaning up the internal setup of the text frame attribute, in order to make it a real attribute object. For now, this was just a collection of hacks used to test VP's text rendering.
Found a strange behaviour in the font rendering. The blackness of the characters is not constant. Some text frames have very dark output, some medium and some still lighter. This must be related to anti-aliasing, but I do not know where to search. Here is a sample (from dark to light) :

Associated more information to a text flow frame, in order to quickly find the flow for a given object.
Fixed bug in VP::TextLib::Fitter::ParaStore::FitLine which prevented text to flow into multiple text frames.
The FlowManager is now associated with the document rather than the rendering context.
The hit testing for text frames now restricts its search to the matching frame rather than stopping at the first possible hit.
Finally found the bug with the non constant character blackness. This was the result of painting several times the same text at the same place... Now, VP::TextLib::Scribe::LineStore::PaintLines only paints the lines which belong to a specified flow/frame.
Fixed text and caret/selection update details.
Avoid painting the caret multiple times (same problem as for the text output).
Improved slanted caret positioning.
Fixed small problem in VP::TextLib::Scribe::LineStore::CacheLine which caused lines with ligatures to lose their trailing punctuation marks.
Written tons of OpenType structures and definitions in order to read the ligature information in the fonts. This requires walking through the GSUB (glyph substitution) table, looking for the 'liga', 'dlig' and 'rlig' features, which in turn give access to lookup tables, which contain the real information...
I have working ligatures (23h55) !

6 November 2002

Fixed font height calculation. What I called H1 in my terminology is represented by the font ascent and the font line gap (taken from OUTLINETEXTMETRIC). Using just the ascent produced text lines which were too condensed, the caret was too short and the highlighting did not reach down to the bottom of the chacters. Windows seems to be adding the gap before the line.
Added support for displaying selected text. This is all handled by classes in VP::TextLib::Scribe and class VP::TextLib::Flow.
Implemented text selection with the cursor keys. Erasing the selected text works, too.
Implemented text selection with the mouse.
Fixed a lot of small problems related to the caret and to its refreshing.
Implemented double-click for word selection, triple-click for line selection, quadruple-click for paragraph selection and ...yes... quintuple-click for full text selection.
Modified class VP::DocText::Style to make it compatible with the other VP attributes. This was not the case, which meant that a style containing a T.style object crashed the rendering engine when it tried to analyse the style's attributes.
Fixed document setup code.

5 November 2002

Fixed yesterday's fix on keyboard filtering : the I-beam cursor wasn't consuming any more events.
Working on keyboard navigation in text flow. Implemented following cursor movements :

Move to begin/end of text flow.
Move to next/previous paragraph end/begin.
Move to next/previous line, moving vertically in the text. This is the most complicated of all moves, since it requires geometric knowledge about the text.
Move to next/previous word. This was not as easy as expected because of a lot of special conditions which must be taken in account.

Fixed stream data insertion, which generated empty objects when inserting/removing a pair of cursors.

4 November 2002

Checking in yesterday evening fixes.
UNICODE defines a lot of accented characters to be AI_AmbiguousAlphabeticOrIdeographic, which does not make sense to me. These must be mapped internally to AL_OrdinaryAlphabeticAndSymbol, or else the line breaking algorithm will behave badly.
Added support for spying WM_CHAR, WM_DEADCHAR and others in VP.Forms.Cmd.Sink.
Fixed cursor record access in the document store (split cursor records were not skipped properly).
Fixed offset bug in caret position calculation, caused by the possible clipping region.
Added support for DELETE and BACKSPACE keys when editing text.
Fixed bug in VP::DocText::Stream::InsertData, when overflowing into next object.
Fixed bug in VP::DocText::Meta::GenerateBreaks which caused the temporary cursor to fall within UTF-8 sequences.
Fixed UnicodeHelper::AnalyseBreaks to accept line breaks at the very beginning of lines. And also modified the handling of trailing spaces in runs. This is no longer 100% Unicode standard conformant when considering the extremities of a run.
Fixed cursor handling in empty lines (line with just a line break in it) in VP::TextLib::Scribe.
Fixed badness calculation for zero-length lines.
Naïve floating point comparisons caused VP::TextLib::Scribe to store lots of runs instead of merging them.
Improved cursor position detection.
Added support for the HOME and END keys when editing text. And again a few bugs surfaced in the text stream management. It was not as simple as expected to locate a begin or an end of line in the text flow, because nobody was setting the TAG_SKIP_BREAK_ACTIVE flag after fitting a paragraph... This has been addressed.
Fixed caret display if multiple spaces hang outside of the right margin.
Fixed the debugging tool so that its text box gets the keystrokes which are intended for it.

3 November 2002

Continuing work on text interaction.
The shortcut message filter was consuming characters which were not meant for it (for instance, pressing the DEL key whould delete the selected object, even if the user was typing text in a text box). Class VP.Forms.Cmd.Sink now has the basic mechanisms needed to disable part of this filtering when a text consuming control has the focus.
The active visual style of the main window and the palette was not behaving as it did in other applications (such as Word, Photoshop and other prominent applications). Clicking into a palette would deactivate the main window, which was bad. And clicking in the main window would deactivate the palette, which was bad too. I had to design a new GUI.TopLevelForm class from which both the main window and the palettes derive, which does some tricky event juggling in its WndProc method.
Without the help of an article adressing a similar problem in VB, I would still be trying to guess how to get things right.
Hunted bugs while trying to type text in a text flow...

2 Novembre 2002

Continuing work on text interaction.
Fixed clicking after a paragraph end : the position is now correct (Unicode::BK_MandatoryBreak character is now set to STRETCH_Invisible). This is also the case when clicking before the line begin.
Added support for distinction between two same positions in the text flow, which are displayed at different positions on the screen (end of line before hyphen, start of line after hyphen).
Implemented support for slanted cursor when italic fonts are used.
Implemented centered and right-aligned layout.
Implemented the paragraph hyphenation settings in VP::DocText::StateRecord. This required VP to be able to determine if a letter is a CAPITAL LETTER or not (to avoid hyphenating capitalised words). This information is now provided by the UnicodeHelper class.
Implemented text "compression" which packs the glyphs as tight as possible to fit a line into less space than theoretically required.
Removed class VP::TextLib::Scribe::Renderer (all its methods could be moved to LineStore) and finished the cleanup of class VP::TextLib::Scribe::LineStore.

1 November 2002

Working on text interaction.
Finally got a blinking cursor...
Added polygon hit detection (method VP::TextLib::PolygonTools::Contains).
Added support for text detection : given a pair of [x,y] coordinates, find which flow/line/character is affected.
Fixed a nasty bug which was introduced on 28 August, as a side effect of the thin line hack implemented by scaling the pen width by 100 (and scaling by 0.01 when rendering the path). This caused the bounding box computation to assume that the outline of the figures was very, very thick. Therefore, most figures became unselectable by the rectangular selection tool.
Spaces and unbreakable spaces are both flexible; this makes much more sense than Microsoft's choice in Word.
Finished hit detection : automatically update the cursor position on successful hit detection. But there are still a lot of small details which must be taken care of :

VP does not distinguish the position after the last character of a hyphenated line and the first character of the next line; the cursor always shows up at the beginning of the next line.
VP does not properly handle the paragraph end control character embedded in the text : clicking after the end of a paragraph's last line moves the cursor to the next line, not to the end of the paragraph.

31 October 2002

Integrating the low-level work done on the text management with the front-end.
Added an AttributeBlock.Find method to search for a specific attribute in an object (the attribute is specified by its UCID).
Added support for named image definitions (the front-end can send bits to the back-end, which remembers them as a GDI+ bitmap, for use by some objects or by the view to display icons).
The I-beam tool detects when it is over a non-textual object and provides a visual feed-back.
Delegated more decision to the tool plug-ins :

The tools are notified of their future activation (OnAboutToActivate).
The tools can specify a custom cursor to be used as the standard cursor (GetCustomCursor).
The tools can specify a special candidate hilite mode and filter the candidates which should not be hilited (CandidateFilterDiscard).

30 Octobre 2002

Continuing work on text layout.
Added Freeze operation to the NodeStore and ParaStore, which release the huge virtual memory range they acquired through the turbo heap (see OPaC::Tools::TurboHeap::Freeze).
Cleaned up the text frame attribute : consolidated the painting code.
Fixed an old bug which caused VP to crash on document loading. The problem was that nobody ever called the LoadPostProcessing method of the management classes after a Load.

29 October 2002

Partially cleaned up the sources written hastily yesterday.
Fixed VP::TextLib::Fitter::ParaStore::Fit and FitLine to properly handle truncated paragraphs (if there are no more frames available).
Improved text flowing when the containing polygon geometry is not ideal (not wide enough at certain places to display the shortest word element).
Extracted the experimental text rendering code into a dedicated GlyphRunManager class.
Implemented the real justification algorithm. For now, we do not cache the results of the justification, but when the document will become larger, our current, naïve, approach won't be fast enough.
Enabled anti-aliasing of the displayed fonts.

28 October 2002

Fixed small problem with active node deactivation.
Replaced sub-optimal code in VP::TextLib::Fitter::ParaStore::CreateNode.
Refactored the fitter classes; class ParaStore is now much simpler to understand and the dedicated Scribe namespace has been added to contain the justification code.
The node now remembers the hyphenation point.
Method VP::TextLib::Fitter::Context::ExtractWord recognizes a text fence and rewinds the text and tags streams accordingly when the fence is reached.
Added Scribe::StretchDef which replaces the simple boolean for the space/non-space character identification (this will allow future support for kashidas and stretchable dashes).
Cleaned up TextFlow.cpp source file. Removed the outdated version of the fitter algorithm.
Reorganised the internal font representation.
The VP::DocText::State class now provides direct access to the active font instance.
First sample text displayed this evening... and printed too, after a few adjustments. GDI+ and GDI do not change their coordinate systems the same way, which caused a bit of head scratching.

27 October 2002

Working on the line justification algorithm. I made a first attempt at a justification algorithm which would be executed while the fitter is doing its job, in order to avoid having to reparse the text twice, but this appears to be a poor choice; so I ended the day with no progress at all...

26 October 2002

No work this Saturday.

25 October 2002

Fixed a memory deallocation problem in the TurboHeap code (the call to VirtualFree requires a zero length when releasing memory).
Method VP::DocText::Fitter::NodeStore::CreateNode now recycles previous nodes and thus dramatically recudes the number of active nodes in the defavorable cases.
Reorganised the source file Fitter.cpp into three distinct files : Fitter, FitterNodeStore and FitterParaStore. Cleaned up and added comments.
Fixed a bug in VP::TextLib::Fitter::ParaStore::RewindToCount.
Working on the paragraph/line justification algorithm.

24 October 2002

The paragraph fitter did its first 100% bug-free fit at 16:00 (local time ;-) !
Method VP::DocText::State::Setup still contained a few problems. This is now fixed.
Fixed a lot of details in the fitter code itself :

VP::TextLib::Fitter::NodeStore::FreeDiscardedNodes : don't get stuck in a loop.
VP::TextLib::Fitter::ParaStore::DeactivateAndMoveToNextActiveNode : accept active nodes, but also accept output nodes.
VP::TextLib::Fitter::ParaStore::SelectBestOutputNode : update the output node index after discarding those which are no longer needed.
VP::TextLib::Fitter::ParaStore::ExtractWord and others now return the offset in the text with respect to the context's text/tags stream positions.

Added support for following features in the fitter :

Quality setting for the fit (progressive quality degradation).
EndOfText and ErrorNeedMoreFrames event handling.

Implemented the missing VP::TextLib::Fitter::ParaStore::CreateNode method.

23 October 2002

Testing and debugging the paragraph fitter.
While testing the paragraph fitter code, I stumbled over a few bugs hidden in already tested code :

Fixed VP::DocText::Meta::SynthesizeBreakTags : when UTF-8 characters used more than one byte, then the resulting break tags would become desynchronised.
Fixed VP::DocText::Accessor::Synchronise : moving backwards could cause accesses outside of the stream, since the recorded original offset was sometimes wrong.
Fixed VP::DocText::State::Setup which did not properly reinitialise the internal parsing parameters and caused re-runs of the same text to become desynchronised with respect to the tags stream.

22 October 2002

Continuing work on the paragraph fitter. Adding real OpenType measurements.

21 October 2002

Worked on OpenType integration.

17 October 2002

Continuing work on the paragraph fitter. 90% of the fitter is in place, but there are still a lot of places where functions still need to be fully implemented.

16 October 2002

Resuming work on the paragraph fitter.

12-15 October 2002

Week-end and two days off duty, reading a very interesting article on what typographers expected from Adobe's K2 project (which turned out to be InDesign).

11 October 2002

Continuing implementation of the new paragraph fitter.

10 October 2002

End of design of the paragraph fitter. I am happy with the result.
Started implementation of the new paragraph fitter. This will take a few days !
Implemented a TurboHeap class which can be used to allocate memory as a contiguous array very quickly, without committing memory before it is really needed. This is based on virtual memory allocation (the default is to reserve a chunk of 16 MB in virtual memory and grow it as needed), which guarantees that the logical addresses of the allocations will be contiguous.

7-9 October 2002

I stopped working on the paragraph fitter. More design is needed before I can start coding something useful and extensible. I hope to have all ideas put down on paper tomorrow (10 Oct.).

5-6 October 2002

Continuing work on paragraph fitter.
Cleaned up code and structures used by the fitter.
Added a FitGlyphRange structure used to store the glyphs and the associated width information, which will be used to cache the text to speed up the calls to ExtTextOut.

4 October 2002

Working on the paragraph fitter (added a few text fitter related structures).
Fixed a few problems with VP::DocText::Accessor.
Implemented a DebugDump method in VP::DocText::Meta, which helped me find several bugs :

The file Document.cpp contained invalid testing code which broke the text stream and tags stream integrity, causing some strange bugs in the fitter...
The file Document.cpp initialised the text formatting style too late, which caused some paragraphs to have invalid paragraph style IDs, breaking once more the fitter.
The VP::DocText::Meta::RemoveText method was off-by-one in the tags stream (in the cases where the starting position pointed into a skip tag range), which caused an incorrect number of tags to be removed and brought the streams out of sync.

3 October 2002

Hacked a bit in the code to make things progress faster, and help me find some additional bugs. The a.textframe uses a manually created text flow and meta text block to allow for some simple experimentation.

2 October 2002

Added support for text and tags generation IDs in VP::DocText::Meta, which can be used to find out whether a specific pointer into the text is still valid or not.
Added flow management classes in VP::TextLib.
Added support for a special kind of unit (i24) which encodes a raw 24-bit unsigned integer, used to store indexes into objects, such as a.textframe.
Updated the flags defined by VP::Render::Context, and their use in the different document management classes. This cleanup was the consequence of trying to understand how they worked... and failing.
Working on the Text Frame attribute (a.textframe), which calculates the geometry of a text frame, based on a path definition provided by any kind of object. Added interaction with the Text Flow.

1 October 2002

Rewritten VP::DocText::Meta::InsertText to handle the paragraph style format tags.
Read the Guide du typographe romand.

30 September 2002

Continuing work on state and format processing.
Rewritten the text deletion method (VP::DocText::Meta::RemoveText) to properly handle format tags. The old implementation relied on the user to provide a number of characters to remove, which would not take in account the complex logic behind cursor movement (which ensures that the cursor never points to an invalid position, e.g. between a paragraph end character and a paragraph style tag).
Fixed VP::DocText::Meta::MoveBackwards, which could get stuck in an infinite loop.
Updated VP::DocText::Meta navigation methods to handle the FORMAT_ParaStyle formatting tag.
Added a first version of VP::DocText::State::SerialiseState.
Working on VP::DocText::Meta::InsertText to properly handle the paragraph separation marks, which require the insertion of special format tags (FORMAT_ParaStyle).
Added paragraph style accessing code to VP::DocText::Meta and VP::DocText::Format.

28-29 September 2002

Week-end.

27 September 2002

Working on the text state and format tag processing :

Added VP::DocText::State class to manage state computation based on a position in the tags/text streams.
Added VP::DocText::StateRecord structure to store the information about the active state.
Added VP::DocText::Style to store the text style information (this is in fact a tags stream, but it is guaranteed to contain contiguous data; there will never be a linked list of text style objects).

This is a day with a lot of new code :-)

26 September 2002

Refining the format description for paragraphs, characters and text in general. Originally, I planned to use a format stream to store additional information on the formatting, but this is not a good idea. Put all the formatting information directly into the tags stream and only longer arguments (such as font names) into the format stream.
Implemented access to format data in VP::DocText::Meta, based on a cache. This can be used to find, allocate, store and remove pieces of format data identified by a 16-bit format ID. It is also possible to make sure that two pieces of data with the same content won't get duplicated by searching first for a matching format data record.
Re-refactored the font listing mechanism used by VP, in order to classify the fonts in families (based on their preferred family as specified by OpenType, not as provided by brain-dead Windows API). Now fonts such as Warnock Pro Light and Warnock Pro Semibold belong to the same family as Warnock Pro Regular.

25 September 2002

Started implementation of paragraph splitting.
Nice Adobe fonts with typographic finesse :

Adobe Jenson Pro.
Minion Pro Opticals.
Warnock Pro Opticals. Bought this family to experiment with the different weights (light, regular, bold, semi-bold) and with the additional glyphs.

I had to redesign the font management code in Virtual Pen to avoid GDI+ completely, since the OpenType fonts from Adobe did not show up in the list. This took me about 5 hours to try to figure out where I had done something wrong, and 1 hour to rewrite my code to use plain old GDI.
Defined a format record, which contains cache-able information (amongst which a pointer to a VP::TextLib::FontRecord object).

24 September 2002

Working on paragraph splitting. I am going to use a modified version of Knuth's algorithm, which is the best to date, to my knowledge...
Found an article on direction and language specification related to HTML : http://www.w3.org/TR/REC-html40/struct/dirlang.html.
The UTF-8 specification is part of RFC2279.
Unicode characters in the range U+E000 up to U+F8FF included are defined as private area. This is split into an end user zone and a corporate user zone (U+F000-U+F8FF). Some vendors such as Apple (U+F800-U+F8FF, Apple Logo U+F8FF), Microsoft (U+F000-U+F0FF), Adobe (U+F600-U+F7FF), Linux (still evolving in the CUS, Klingon starting at U+F8D0) use this for characters not found in the standard Unicode range (e.g. partial glyphs from the Symbol font). And some would like to standardise the use of the private are for fictional character sets. Some Commodore fans would like to use part of the CUS for PETSCII.
How should these characters be handled by Virtual Pen ?
What character codes may I select for Virtual Pen's internal use ?
It is better not to use any private area codes, if possible. For now, I will stick to characters 02 (ASCII SOT) and 03 (ASCII EOT) as markers for begin and end of text inclusions (such as foot notes or URL targets).
Lots of thinking, no coding...

23 September 2002

Integrated Daniel's hyphenator with VP::DocText::Meta.
Split the large VP::TextLib::Meta::GenerateBreaks method into a main method and four sub-methods, which implement following funcitonality :

Find start of zone and create temporary cursor A.
Find end of zone and create temporary cursor B.
Break according to Unicode rules and hyphenate the alphabetic text.
Update the skip tags and insert break tags as needed.

Added a text distance measurement method to VP::TextLib::Cursor class.
Debugged thoroughly (at least I hope so) the break tag generation and management.

22 September 2002

Working on line splitting at the text tags stream level.
Fixed a bug in VP::DocText::Accessor::Read/Write, which caused navigation problems.
Fixed a bug in the hyphenator initialisation, which caused the the default hyphenator to be deleted without notice (and caused mysterious crashes in method VP::TextLib::Meta::GenerateBreaks).
Fixed a bug in VP::DocText::Cursor::Find, which caused cursor deletions to remove the wrong cursor record.
The line breaking algorithm is now working with a byte provider rather than a string, which allows a better implementation of the break analysis (the source must no longer be copied into a temporary buffer before it can be used).
The tags used to mark breaking positions in the text were not sufficient (formerly, I had just the break here and break option flags). The following information needs to be encoded into the break tags :

No break can happen here.
A break must always happen here.
A break might happen here if needed (word boundary).
A hyphenation and a break might happen here if needed (hyphenation within a word).
It would be nice to also be able to record the quality of the hyphenation (poor, good, best).
With the current layout, a break is done here (active break).

This information needs more than just 2 bits to be properly encoded. The first 4 states (6, counting the hyphenation quality) fit into a 2-bit (3-bit) field. The last information is a flag (break or no break). I finally selected following layout :

TAG_SKIP                    0b xxx nnn

TAG_SKIP_BREAK_ACTIVE       -b --- ---
TAG_SKIP_BREAK_STATUS       -- xxx ---
TAG_SKIP_VALUE              -- --- nnn

This reduces the range which can be encoded with TAG_SKIP to 1..7. It would be possible to encode a wider range with the 3 bits (e.g. 1, 2, 4, 8, 16, 32, 64), building the intermediate values by summing up skip tags. This does not make sense, however, since most words are less than 8 characters, or can be hyphenated into short character sequences.

21 September 2002

Continuing work on font management and line geometry measurement. Mostly worked on OpenType interface.

20 September 2002

Added break handling for embedded objects (Unicode character 00FFFC), through interface IObjectBreakInfoProvider.
Implemented Unicode-character-class based line breaking algorithm. This is based on the Unicode Standard Annex #14 and on file LineBreak.txt provided by the Unicode Consortium.
Here are a few related articles :

A criticism of the Unicode rules : http://www.cs.tut.fi/~jkorpela/unicode/linebr.html
About typographical units : http://css.nu/articles/typograph1-en.html
Dashes and hyphend : http://www.cs.tut.fi/~jkorpela/dashes.html
Word division (in IE, but of general interest) : http://www.cs.tut.fi/~jkorpela/html/nobr.html
HTML delimiters : http://www.cs.tut.fi/~jkorpela/rfc/2396/full.html#E

Working on the hyphenator, which meant restructuring part of the Unicode related code in TextLib.
Working on the font management.

19 September 2002

Tried to find an elegant work-around for the cursor update bug, but getting the updates right is very difficult and quite inefficient. To simplify the design, I decided to add support for a temporarily incoherent cursor state (VP::DocText::Cursor::StopAutoSync and ResumeAutoSync).
This is now implemented and works...
Worked a bit on MSDE for Daniel's project. Nice thing, this light SQL Server.

18 September 2002

Fixed coherence problem with VP::DocText::Meta::RemoveText (the bug was found thanks to the use of the new VP::DocText::Cursor::IsCoherent function).
Added VP::DocText::Meta::ClearBreakTags and VP::DocText::Meta::OptimiseBreakTags.
Added support for virtual cursor IDs, for both special cases : begin and end of stream.
Added word break detection, as specified by the UNICODE Consortium : this requires that VP parses a LineBreak description file in order to extract the break class of every Unicode character.
Found a bug with the cursor updates, when optimising a skip tag stream.

17 September 2002

Continuing work on VP::DocText namespace.
Fixed VP::DocText::Stream::MoveBackwards and VP::DocText::Meta::MoveForwardsInText.
Finished, tested and fixed VP::DocText::Meta::MoveBackwardsInText. Text navigation is therefore fully implemented now.
Added more functions to the accessor (in order to check a cursor's coherence, that is, make sure that the text position is the same as the tags position, as specified by the skip tags).

14-16 September 2002

Week-end (Jeûne Fédéral).

13 September 2002

Continuing work on VP::DocText::Meta : backwards navigation.

12 September 2002

No work for EPSITEC today... but days have evenings.
Debugged the VP::DocText::Meta::MoveForwardsInText method, which crashed in case of a stream overrun.
Debugged the VP::DocText::Cursor class, which had a few problems :
- Finding a cursor could get stuck in an infinite loop. ü
- Accessor synchronisation now returns false if there is no more data available, as expected. ü

11 September 2002

Continuing work on VP::DocText::Meta management class, to add support for higher level text manipulation and navigation. While doing this, I realised that there were a few flaws in the design : some tags are irrelevant and difficult to handle properly (the kerning pair adjustment, specifically). So here we go : re-engineering of the tags is in progress.
Added simplified failure management with the VP::Failure function; used it for some stream navigation methods.
Implemented forward navigation in the text/tags stream.
Daniel Roux has implemented a word breaking algorithm for the French language (the basic algorithm was originally developed for the Smaky computer and implemented in the COUPE library).

10 September 2002

The following low-level functions are now implemented and partially tested :
- UTF-8 text insertion, staying in sync with the tags. ü
- UTF-8 text removal, partial tag management. ü
Implemented in-place manipulation of data in the VP::DocText::Stream, properly handling the fact that the data can be distributed over several objects. The data access is coupled with an optional relative distance, which can be used to peek at data before the current position (e.g. the previous tag) without having to explicitely move the stream index/offset.
The cursor record needed additional information to properly handle skip tags (it is possible that, while moving through text, the cursor points into a skip tag; a relative skip offset is needed here).

9 September 2002

Added dumping support for the VP::DocText::Cursor object.
Defined the command tags used by VP::DocText::Tags.
Added support for the UTF-8, UTF-16 and UTF-32 Unicode representations. Internally, the text is stored as UTF-8, which is simple to manipulate.

7-8 September 2002

Week-end.
Defined the hyphenation engine, so that it can get implemented by others.

6 September 2002

EPSITEC meeting today.
Handle properly cursors pointing into deleted stuff thanks to the DocText::Meta::NotifyCollapse method.

5 September 2002

A compiler bug caused the polygon tools to fail compiling : somehow, a call to VP::Point::operator = in the VP::TemplateSABase<VP::Point>::Remove was the origin of the mess.
Implemented data removal from the text stream and tested both insertion and removal. Everything seems to be working properly.
Implemented basic text cursor management : when an insertion or deletion affects a cursor position, it gets updated automatically.

4 September 2002

Continuing work on text stream.

3 September 2002

Implementing the basic text objects used for internal data representation :
- VP::DocText::Story, block containing the text story (the text meta information, the text itself, the text tags, the text formatting and the text cursors).
- VP::DocText::Meta, meta information about a text story.
- VP::DocText::Text, the text itself, stored as a UTF-8 stream.
- VP::DocText::Tags, the text tags, stored as a stream of bytes encoding skip marks, kerning deltas, typographic variants, state changes, etc
- VP::DocText::Format, the formatting information associted with the text tags.
- VP::DocText::Cursor, position information pointing into the text stream and maintaining the current state.
Added a base class for Text, Tags, Format and Cursor, which is VP::DocText::Stream. It contains all the common insertion/removal/navigation logic.
Added support for cacheable data to the document store : an object tagged with the OBJFLAG_CACHE flag will be handled specially by the object store, on deletion.

2 September 2002

Fixed bug in polygonal path validation.
Implementing font management code.

History of previous months.

To do soon :
- Further improve the line style edition dialog (width, cap, join, miter); a job for Daniel ?
- Finish the polygon plug-in; a job for Daniel ?
- Allow reordering of attributes in the attribute panels (e.g. color first, transparency last).
- Change the way layers are represented in pages.
- Restore Bézier curve functionality.
- Adding support for grouping objects.