Architecture

From 05693a538b488336af0a83e402b6d7eb584c7a73 Mon Sep 17 00:00:00 2001 From: Keiron Liddle Date: Mon, 11 Nov 2002 09:23:48 +0000 Subject: [PATCH] more to new docs git-svn-id: https://svn.apache.org/repos/asf/xmlgraphics/fop/trunk@195466 13f79535-47bb-0310-9956-ffa450edef68 --- docs/design/architecture.xml | 58 ----- docs/design/areas.xml | 186 --------------- docs/design/book.xml | 27 --- docs/design/breakpos.xml | 289 ----------------------- docs/design/embedding.xml | 114 --------- docs/design/extending.xml | 99 -------- docs/design/fotree.xml | 92 -------- docs/design/intro.xml | 51 ---- docs/design/layout.xml | 438 ----------------------------------- docs/design/optimise.xml | 58 ----- docs/design/properties.xml | 257 -------------------- docs/design/renderers.xml | 30 --- docs/design/status.xml | 73 ------ docs/design/useragent.xml | 176 -------------- 14 files changed, 1948 deletions(-) delete mode 100644 docs/design/architecture.xml delete mode 100644 docs/design/areas.xml delete mode 100644 docs/design/book.xml delete mode 100644 docs/design/breakpos.xml delete mode 100644 docs/design/embedding.xml delete mode 100644 docs/design/extending.xml delete mode 100644 docs/design/fotree.xml delete mode 100644 docs/design/intro.xml delete mode 100644 docs/design/layout.xml delete mode 100644 docs/design/optimise.xml delete mode 100644 docs/design/properties.xml delete mode 100644 docs/design/renderers.xml delete mode 100644 docs/design/status.xml delete mode 100644 docs/design/useragent.xml diff --git a/docs/design/architecture.xml b/docs/design/architecture.xml deleted file mode 100644 index a9a19e318..000000000 --- a/docs/design/architecture.xml +++ /dev/null @@ -1,58 +0,0 @@ - - - -

- Architecture - Architecture information for FOP - - - -

- - - - - - -

-The overall process is controlled by org.apache.fop.apps.Driver. -This class handles the FO Tree building, renderers, output and logging. -

-The process in general is that the FO document is sent to the tree -builder via SAX events. This creates an FO Tree. The FO Tree is then -handled by the layout processor which converts the FO Tree into an area -tree. This area tree is then given to the renderer and the renderer converts -the area tree into a stream of data containing the output document. -

- - - -

-The class org.apache.fop.fo.FOTreeBuilder is responsible for -actually constructing the FO tree. The key SAX events used are

startElement(),

endElement() and characters().

- -

All formatting objects derive from abstract class -org.apache.fop.fo.FONode. The other FO classes inherit from -FONode as follows:

- - - - -

-This is a separate process. The render() method in -Driver is invoked (say, -by CommandLine) with the laid-out AreaTree and a -PrintWriter as arguments. -This actually calls the render() method in a specific implementation of -the Renderer interface, typically PDFRenderer or -AWTRenderer. -

- - - - - - diff --git a/docs/design/areas.xml b/docs/design/areas.xml deleted file mode 100644 index d8b9c830f..000000000 --- a/docs/design/areas.xml +++ /dev/null @@ -1,186 +0,0 @@ - - - - - -

- Area Tree - Area Tree Design for FOP - - - -

- - - -

-The code to implement the area tree will attempt to match the areas -defined in the specification. A number of optimisations may be possible -for similar areas and groups of areas. -

-Since the area tree will be used during the layout by the layout managers -it will need to store information that affects the layout. The information -such as spacing and keeps will be held in such a way that it can be -discarded once the layout is finalised. -

- -

-The area tree is a root element that has a list of page-viewport-areas. -Each page viewport has a page-reference-area which holds the contents of -the page. To handle the processing better FOP does not maintain a list -at the root level but lets another class handle each page as it is added. -

- - -

-A page is made up of five area regions. These are before, start, body, -end and after. Each region has a viewport and contains the areas -produced from the children in the FO object heirarchy. -

-For the body area there are more subdivisions for before floats, -footnotes and the main reference area. The main reference area is -made from span areas which have normal flow reference areas as -children. The flow areas are then created inside these normal flow -reference areas. -

-Since the layout is done inside a page, the page is created from the -pagemaster with all the appropriate areas. The layout manager then -uses the page to add areas into the normal flow reference areas -and floats and footnotes. After the layout of the body region -is complete then the other regions can be done. -

- - -

-Block areas are created and/or returned by all top level elements -in the flow. These areas have keep and spacing information that -needs to be retained until the page is finalised. A block area -is stacked with other block areas in a particular direction, it -has a size and it contains either line areas made from a group -of inline areas or block areas. -

-A block area can also be split into two block areas by splitting -between two line areas or splitting between two block areas (or -groups) that are stacked in the block progression direction of -the page. The split may also be in a child block area. -

- - -

-A line areas is simply a collection of inline areas that are stacked -in the inline progression direction. A line area has a height and -width. It also contains information about floats and footnotes -that are associated with the inline areas. -

-A line area gets a set of inline areas added until complete then -it is justified and vertically aligned. If the line area contains -unresolved areas it will retain the justification information -until all areas are resolved. -

- - -

-There are a few different types of inline areas. All inline areas -have a height. Their width may be variable until the line is -finalised. -

-Unresolved areas can reserve some space to allow for possible -sizes once it is resolved. Then the line can be re-justified -and finalised. -

- - -

-Any subtree of the area tree should be cloneable so that for -areas that are repeated the area tree can simply be copied rather -than going through the layout again. This will only work if the -width is the same. -

-Resolveable areas may be converted into an unresolved form. -

- - -

-The following class structure will be used to represent the area -tree. -

- -

-The page area classes hold the top level layout of a page. The -areas are created by the page master and should be ready to have -flow areas added. -

- - -

-The block areas typically hold either a set of line areas or a set of -block areas. The child areas are usually stacked in a particular -direction. -

-Areas for tables and lists have their child block areas stacked -in different ways. Lists also can have spacing between the block -areas. -

- - -

-The inline areas are used to make up a line area. An inline area -typically has a height, width and some content. The alignment is -used for block progression direction displacement and to determine -the height of a line. -

- - - - -

-The rendering of an area tree is done by rendering each page -to a suitable output. The regions are rendered in order and each -region is contained by a viewport. -

-The relevent structures that will need to be rendered are: -Page -Viewport -Region -Span -Block -Line -Inline -

-The renderer will need to be able to: - - -render each individual page - - -clip and align child areas to a viewport - - -handle all types of inline area, text, image etc. - - -draw various lines and rectangles - - -

-An abstract renderer will be able to handle the generic positioning -of child areas, iterating through areas that have child areas. -

- - - - - - - diff --git a/docs/design/book.xml b/docs/design/book.xml deleted file mode 100644 index 315ed9cac..000000000 --- a/docs/design/book.xml +++ /dev/null @@ -1,27 +0,0 @@ - - - - - - - - - - - - - - - - - - - - - - - - - - - diff --git a/docs/design/breakpos.xml b/docs/design/breakpos.xml deleted file mode 100644 index 34df68d9f..000000000 --- a/docs/design/breakpos.xml +++ /dev/null @@ -1,289 +0,0 @@ - - - - - -

- Layout Managers - Break Possibility Proposal - - - -

- - - -

-As explained in Layout, -the hierarchy of Layout Managers is responsible for building and placing -areas. Each Layout Manager is responsible for creating and filling -areas of a particular type, either inline or block. This document -explains one potential algorithm for this process. It is based on the -the generation of break possibilities (BP for short). The -Layout Managers (LM for short), will generate one or more BP and -choose the best one. The BP is then used to generate the corresponding -areas. -

- -

A break possibility is represented by the BreakPoss class. A -BreakPoss contains size information in the stacking direction and in -the -non-stacking direction (at least for inline areas, it must have both). Flags -indicating various conditions (ISFIRST, ISLAST, CAN_BREAK_AFTER, -FORCE_BREAK_AFTER, ANCHORS etc). A BreakPoss contains a reference to -the top-level LayoutManager which generated it. -

A BreakPoss contains an object implementing -the BreakPoss.Position interface. This object is specific to the layout -manager which created the BreakPoss. It should indicate where the -break occurs and allow the LM to -create an area corresponding to the BP. A higher level LM Position -must somehow reference or wrap the Position returned by its child LM in its -BreakPoss object. The layout manager modifies the flags and dimension -information in the BP to reflect its own requirements. For example an -inline FO layout manager might add space-start, space-end, border and -padding values to the stacking or non-stacking dimensions. It might also -modify the flags based its on keep properties.

- - -

Once break possibilities have been generated, the galley-level -layout manager selects the best one -and passes it back to the LayoutManager which generated it to create -the area. A LayoutManager is responsible for -storing enough information in its Position objects to be able to -create the corresponding areas.

- - -

Layout Managers are created from the top down. First the -page sequence creates a PageLM and a FlowLM. The PageLM will manage -finding the right page model (with help from the PageSequenceMaster) -and managing the balancing act between before-floats, footnotes and -the normal text flow. The FlowLM will -manage the normal content in the main flow. We can think of it as a -galley manager. -

In general, each LM asks its child LMs to return sucessive -break possibilities. It passes some -information to the child in a flags object and it gets back -a break possibility which contains the size in -the stacking direction as well as information about such things as -anchors, break conditions and span conditions which can change the -reference area environment. This process continues down to the lowest -level of the layout manager hierarchy which corresponds to atomic -inline-level FOs such as characters or graphics. -

-Each layout manager will repeatedly call getNextBreakPoss on its current -child LM until the child returns a BP with the ISLAST -flag set. Then the layout manager moves on to its next child LM (ie, -it asks the next child FO to generate a layout manager.) Galley level -layout managers which are Line and Flow will return to their parent -layout managers either when they have finished their content or when -they encounter a a BP which will fill one of their areas. -

The break possibilities are generated from the bottom up. -All inline content must first be broken into -lines which are then stacked into block areas. This is done by the -LineLayoutManager, which creates line areas. -The LineLM asks its child LM to generate a break possibility, which -represents a place where the line can end. This -initially means each potential line-end (primarily spaces or forced -linefeeds and a few other potential line-end characters such as hard -hyphens.) The text LM returns an object which stores the size in the -stacking direction as a MinOptMax triplet -and a cost, which is based on how well this break -would satisfy the constraints. The Text LM keeps track of its position in -the text content and returns the total size of the text area it would -create if it were to break at a given point. The returned BP -object also contains information about whether the break is forced -(linefeed) or whether this is the last area which can be generated by -the LM (ISLAST flag). If a textFO ends on a non-break character, the -ISLAST flag is set, but the CAN_BREAK_AFTER flag isn't, since we don't -know if there is any following text in another inline object for -example. -

Variable size content is taken into account from -the bottom up. Each LM returns a range of sizes in the stacking -direction, based on property values. For text, this comes from -variable word-space values or letter-space values. For other inline -objects, it may include variable space-start and space-end values -(after calculation of the entire sequence of space specifiers at a -particular break possibility.)

The main constraint for laying out -lines is the available inline-progression-dimension (IPD) for the line -area to be created. This -depends on the IPD of the reference area ancestor, on the indents of the -containing fo:block, and on any side-floats which may be intruding on -this line.

-See below Getting the Reference -IPD -for discussion of how the reference area IPD is -transmitted to the Line LM. -

For now, let's assume that only the LineLM knows about the IPD -available to it. Therefore only it can make a decision about which BP -is the best one; the lower level inline layout managers can only -return potential break points.

-There are certainly optimizations to this model which can be -examined later. -

So the Line LM will ask its child LM(s) for break possibilities until -it gets back a BP whose stacking dimension could fill the -line. This means that the BP.stackdim.max >= LineIPD.min. It can look -for further BP, perhaps one whose stackdim.opt is closer to the -LineIPD.opt. If it isn't happy with the choice of break possibilities, -it can go past the end of the line to the next one, and then try to -find a hyphenation point between the last one which fits and the first -one which doesn't. If no possibility is found whose min/max values -enclose the available IPD, some constraint will be violated (and -reported in the log.) The actual strategy is up to the Line LM and -should be able to be easily replaced without changing the architecture -(Strategy pattern). -

The definition of a good break possibility depends on the -properties at the block and inline level which govern things such as -wrapping behavior and justification mode. For example, if lines are -not to be wrapped, only an explicit linefeed can serve as a BP. If -lines are wrapped but not justified then there is no requirement to -completely fill the IPD on each line, but a sophisticated layout -manager will try to achieve "aesthetic rag". -

Note that no areas have actually been created yet. Once the LineLM -has found a potential break point for the inline content, it can -calculate the total size of the line area which would be created. The -size in the IPD is determined by the Line LM based on the chosen BP. -The size of the line area in the the block-progression-dimension -depends on the size of the text (or other inline content). These -values are set by the inline-level LM -in their returned BP (in terms of ascender and descender heights with -respect to the baseline). The LineLM adds spacing implied by the -current line-stacking strategy and line-height property values. It -stores a reference to the chosen inline BP and "wraps" that in its own -Position object which it stores in the BP it returns to its parent LM -(the block layout manager). -

The block LM now has a potential break position after its -first line. It assigns that possibility a cost, based on widow, orphan -and keep properties. It can also calculate the total size of the block -area it would create, were it to end the area after this line. It does -this by adding any padding and border (taking into account -conditionality). It also calculates space-before and space-after -values, or contributes to building up a sequence of such values. -With this information, the block LM creates a new BP (or -updates the existing one). It stores a Position object in this -BP which wraps the returned BP from its child Line LM. -It returns the new BP to its parent and so on, back up to the -FlowLM.

Obviously there is more complicated logic involved when dealing -with lists and tables. These cases need to be walked through in detail.

The FlowLM sees if the returned stacking dimension will still -fit in its available block-progression-dimension (BPD). It repeatedly calls -getNextBreakPoss on its -child LMs until it reaches the maximum BPD for the flow reference area -or until there is no more content to lay out. If one child LM is -finished, it moves on to the next until the last child LM has returned -a BP with the ISLAST flag set. If any child LM returns a -BP with a FORCE_BREAK_BEFORE or SPAN flag set, the FlowLM will -force layout of any pending break possibilities and return to its -parent (the PageLM) in order to handle the break or span condition.

If the returned BP has any new before-float or footnote anchors in -it (ANCHOR flag in the -BP), the FlowLM will also return to the PageLM. The PageLM must then -try to find space to place the floats, possibly asking the FlowLM for -help if the body contains multiple columns.

- - -

Following are a few remarks on specific issues.

- -

If the first child FO in a block FO is an inline-level FO -such as text, the block LM creates an intermediate level LineLM -to layout the -sequence of inline content into Lines. Note that the whole sequence of -inline FOs is managed by a single instance of LineLM. The LineLM -becomes the parent to the various inline-level LM created by each -individual inline FO. -Since an fo:block can have both block and inline content, its LM -may create a sequence of intermixed BlockLM and LineLM.

- - -

When the layout process starts, with the FlowLM asking its first -child LM for a break possibility, the IPD isn't known, since we don't -know whether -the first FO might be spanning, or on which page it might start. (Of -course, if all page masters in the sequence have the same region-body IPD -and all have only a single column, the IPD will never change -and could already be calculated before starting layout.) -The FlowLM gets its -first child LM and calls its getNextBreakPoss method. That is a child LM for -some block-level FO. For now, suppose it's an fo:block. The BlockLM -will create its first child LM, which may be another block-level LM in -the case of nested blocks or a LineLM as explained above. (Question: -do we need a START flag for layout status?) -

We keep calling getNextBreakPoss on lower level layout managers until we -get down to the inline level or to a level which cannot have break-before -properties, such as a list-item-label. At that point, we assume we are -going to have to layout some actual content. But we can't do that yet -since we don't know the inline-progression-dimension. So we return a -BP object which has 0 size in the stacking dimension, but which -has flags set to signal to -higher-level layout managers what needs to be done. If it has a break-before -property or a span property, it stores these in the BP. If -no reference IPD is yet defined, it sets a flag to get that. It then -returns to its parent. The parent LM will inspect the BP object -returned. In general, it "wraps" it with information about its own -needs. If the returned BP is not actually returning any potential -areas, the LM can still add information about its own break or span -requirements. This return path continues back up to the PageLM. It -will then check break and span requirements and create a new page -if necessary using the appropriate page-master. At that point, the -reference IPD for the main -flow is known and is set in the flags object used for -the next getNextBreakPoss call to the lower level LM. -

Using this information, the BlockLM parent can now calculate -the available IPD for its LineLM child, based on its indents. -(If there are any -side-floats information about the intrusion must be passed down by the -FlowLM to lower level managers.) The LineLM can now generate a series -of BreakPoss objects, which it passes back to its parent LM. -

- - -

-The LineLM is responsible for initiating hyphenation if it is allowed -by the properties and if no satisfactory BP can be found without -hyphenating. The hyphenation manager is passed two break -possibilities, one whose IPD is less than the desired line area IPD -and one whose IPD is greater. These break possibilities might have -been generated by different inline-level layout managers (text + a -wrapper with a color change for example), though -frequently they represent two positions in a single text run. -If hyphenation is successful, a new BP is -returned. The LineLM may look for several intermediate BP -based on the "cost" of the returned possibilities. If no intermediate -BP is found, the line will be "short", the white-space stretch will be -exceeded, or perhaps the content will be overflowed or clipped, -depending on various property settings.

- - -

It obviously seems inefficient to go down to the lowest level -LM and back up to the FlowLM for every possible line-break -decision. It seems like it would be possible to optimize by letting -the lower level layout managers run until they had exceeded the -current limit in -the stacking direction. They would then return control to the "galley" -level (LineLM or FlowLM) which would fine-tune the break decision by -asking the lower level LM to find a previous BP which would fit. At -the inline level, this means hyphenation as described above.

Another interesting question is at what point pending break -possibilities can be turned into areas.The idea is to wait until we -are sure we won't have to redo the breaking. This depends on the -sophistication of the layout strategy. For example, if a -linebreak can be considered final if the line is full and there are no -anchors on the line, we could create the LineArea at that point. But -if we are willing to change a previous line-end decision to get a -better overall composition of a whole group of lines (to prevent multiple -hyphens for example), we might wait until the LineLM had finished -laying out all its material and then make all the Lines at once.

- - - - - diff --git a/docs/design/embedding.xml b/docs/design/embedding.xml deleted file mode 100644 index a429dec59..000000000 --- a/docs/design/embedding.xml +++ /dev/null @@ -1,114 +0,0 @@ - - - - - -

- FOP Design - Design Approach to FOP - - - -

- - - -

-This is the design for the external interface when FOP is to be embedded -inside another java application. -

-Common places where FOP is embedded is in a report production application -of a server side application such as Cocoon. -

- - - -

-The user agent is responsible for supplying user or context -specific information. The list of user agent values can be found on the -User Agent page. -

- - -

logging level
logging messages of various levels
error handling
Logging setup (LogKit, Log4J, JDK14Logging)

- - -

various ways to supply FOP with the xsl:fo file, fo, xml+xsl
sax handler

- - -

base directory
uri resolvers
which implementation of a particular
LayoutManager to use

- - -

embedding fonts
compression in pdf
image embedding

-for the PS renderer (eventually): -

PostScript Level
PPD to use
binary/ascii switch

- - -

-Output from FOP: -- Generation statistics: Number of pages total, Number of pages of each - page-sequence, page-master used for each page (could be used to - control the paper bin to get paper from, important for me in - conjunction with PS Renderer). -

- - -

-The Driver handles the XML input. -The user agent information is through the FOUserAgent. -We could handle logging through the user agent. -Options could also be handled through the user agent, using mime type -selection for renderer options. -

- - -

-render to more than one renderer at once (maybe not from the command line). -For example you could generate a PDF for the archive -and the PS for the printer in one run. It would probably be faster than -converting the PDF to PostScript afterwards. -

-Several code pieces for resolving URLs and/or -file locations are scattered all over FOP and Batik. These should -be replaced with an URIResolver invocation to unify behaviour and -remove redundancies. -

- - - - - - diff --git a/docs/design/extending.xml b/docs/design/extending.xml deleted file mode 100644 index a0b45f371..000000000 --- a/docs/design/extending.xml +++ /dev/null @@ -1,99 +0,0 @@ - - - - - -

- FOP Extensions - Adding extensions to FOP - - - -

- - - -

-FOP provides an extension mechanism to add extra functionality. There -are a number of different types of extensions that apply to different -steps when converting FO into the rendered output. -

- -

-SVG Graphic - This applies to svg and any other xml document that -can be converted into svg in the output. All that is required is -the element mapping for the xml and a converter that changes the -document into svg. This conversion is done in the FO Tree. The -conversion is done by the top level element of the namespace -or in the case of an external image a Converter. -

-XML Document - Instead of converting the document into svg it -can be passed directly to the renderer. The renderer will need -to have a handler for the xml document. This handler can add -information directly to the output document. -

-Output Document - This is used to add document level information -to the output result. Such an extension will set information that -is passed to the output document. There needs to be a handler for -the output information which creates a document level result. -

-FO Area - This is where an extension creates an normal area in -the Area Tree. This is useful when the normal FO objects -cannot create the area in the way that is needed. -

-Resolveable - In some cases it may require information to be -resolved for information such as page numbers. This can apply -to the XML Document, FO Area or output document extensions. -

-- Add a string ['(Continued)'] to a table header if the table spans -multiple pages. These tables are part of the content and can start -anywhere in the page. -

-- Separate page number display for a subsection. ie. - master document -is page 4 of 7, but subsection is page 2 of 3. - -

- - -

-Plan - The plan extension is a simple SVG graphic extension. -Given a plan document either inside an InstreamForeignObject -or as an external graphic, it converts the plan document into -an svg graphic. The svg graphic is then passed through the -Area Tree to the Renderer. The Renderer then renders the svg -graphic as normal. -

-PDF Outline - This is output document extension. If rendering to -pdf and this extensionis used then the bookmark information is -passed to the pdf document. This information is then set on the -document. -

-PDF Additions - This can be done with an XML Document extension. -A simple xml document is defined that provides the appropriate -information. When the document is rendered a handler converts the -document into PDF markup. -

-eg. -Search]]> - -to result in a text box referencing the following PDF action: ->]]> - -

- - - - - - - - diff --git a/docs/design/fotree.xml b/docs/design/fotree.xml deleted file mode 100644 index 1f8ed6a3b..000000000 --- a/docs/design/fotree.xml +++ /dev/null @@ -1,92 +0,0 @@ - - - - - -

- FO Tree - Design of FO Tree Structure - - - -

- - - -

-The FO Tree is an internal representation of the input FO document. -The tree is created by building the elements and attributes from -the SAX events. -

-The FO Tree is used as an intermediatory structure which is converted -into the area tree. The complete FO tree should not be held in memory -since FOP should be able to handle FO documents of any size. -

- - -

-The class inheritance described above only describes the nature of the -content. Every FO in FOP also has a parent, and a Vector of children. The -parent attribute (in the Java sense), in particular, is used to enforce -constraints required by the FO hierarchy. -

- -

-FONode, among other things, ensures that FO's have a parent, that they -have children, that they maintain a marker of where the layout was up to -(for FObj's it is the child number, and for FOText's it is the character -number), and that they have a layout() method. -

- - - -

-An FO maker is read from a hashmap lookup using the namespace and -element name. This maker is then used to create a new class that -represents an FO element. This is then added to the FO tree as a child -of the current parent. -

- -

-Properties (recall that FO's have properties, areas have traits, and XML -nodes have attributes) are also a concern of FOTreeBuilder. It -accomplishes this by using a PropertyListBuilder. There is a -separate PropertyListBuilder for each namespace encountered -while building the FO tree. Each Builder object contains a hash of -property names and their respective makers. It may also -contain element-specific property maker hashes; these are based on the -local name of the flow object, ie. table-row, not -fo:table-row. If an element-specific property mapping exists, -it is preferred to the generic mapping.

The base class for all -properties is Property, and all the property makers extend -Property.Maker. A more complete discussion of the property -architecture may be found in Properties. -

- - - -

-FOP supports the handlingof foreign XML. -The XML is converted internally into a DOM, this is then available to -the FO tree to convert the DOM into another format which can be rendered. -In the case of SVG the DOM needs to be created with Batik, so an element -mapping is used to read all elements in the SVG namespace and pass them -into the Batik DOM. -

- - - -

-It is possible to add extensions to FOP so that you can extend the ability of -FOP with respect to render output, document specific information or extended -layout functionality. -

- - - - - - - diff --git a/docs/design/intro.xml b/docs/design/intro.xml deleted file mode 100644 index 101512ede..000000000 --- a/docs/design/intro.xml +++ /dev/null @@ -1,51 +0,0 @@ - - - - - -

- FOP Design - Design Approach to FOP - - - -

- - - -

-The information here describes the design and architecture details for FOP. -Currently this is part of a redesign process for some of the core parts of -FOP. -

-The redesign is mainly focusing on some particular process involved -with the layout process when converting the FO tree into the Area Tree. -

- -

-The main aim for FOP is to comply with the spec and to be able to -process files of any size. -

-In achieving this aim we need to understand the problem and break it -into small problems that can be solved. -

use SAX as input
process FO elements ASAP
dispose of unused memory, keep memory minimal
layout handles floats, footnotes and keeps in a simple straight forward way
id references are kept simple
pages are rendered ASAP, can be cached until resolved
renderers are totally responsible for their output format
output is sent to a stream

- - - - - - diff --git a/docs/design/layout.xml b/docs/design/layout.xml deleted file mode 100644 index 8ee748b6b..000000000 --- a/docs/design/layout.xml +++ /dev/null @@ -1,438 +0,0 @@ - - - - - -

- Layout - Layout Process in FOP - - - -

- - - -

-The aim of the layout system is to be self contained and allow for -easy changes or extensions for future development. For example the -line breaking should be decided at a particular point in the process -that makes it easier to handle other languages. -

-The layout begins once the hierarchy of FO objects has been constructed. -Note: it may be possible to start immediately after a block formatting -object has been added to the flow but this is not currently in the scope -of the layout. It is also possible to layout all pages in a page sequence -after each page sequence has been added from the xml. -

-The layout process is handled by a set of layout managers. The block -level layout managers are used to create the block areas which are -added to the region area of a page. -

- -

-The layout managers are set up from the hierarchy of the formatting -object tree. A manager represents a hierachy of area producing objects. -A manager is able to handle the block area(s) that it creates and -organise or split areas for page breaks. -

-Normally any object that creates a block area will have an associated -layout manager. Other cases are tables and lists, these objects will -also have layout managers that will manager the group of layout managers -that make up the object. -

-A layout manager is also able to determine height (min/max/optimum) -and keep status. This will be used when organising the layout on -a page. The manager will be able to determine the next place a break -can be made and then be able to organise the height. -

-A layout manager is essentially a bridge between the formatting objects -and the area tree. It will keep a list of line areas inside block areas. -Each line area will contain a list of inline areas that is able to be -adjusted if the need arises. -

-The objects in the area tree that are organised by the manager will mostly -contain the information about there layout such as spacing and keeps, this -information will be thrown away once the layout for a page is finalised. -

- - -

-The managers are created by the page sequence. The top level manager -is the Page manager. This asks the flow to add all managers in this -page sequence. -

-For block level objects they have a layout manager. Neutral objects -don't represent any areas but are used to contain a block level -area and as such these objects will ask the appropriate child to -create a layout manager. -

-Any nested block areas or inline areas may be handled by the layout -manager at a later stage. -

- - -

-Block area layout managers are used to create a block area, other block -level managers may ask their child layout managers to create block areas -which are then added to the area tree (subset). -

-A manager is used to add areas to a page until the page is full, -then the manages contain all the information necessary to make -the decision about page break and spacing. A manager can split an -area that it has created will keep a status about what has been -added to the current area tree. -

- - -

-Once the Page layout manager, belonging to the page sequence, is ready -then we can start laying out each page. The page sequence will create -the current page to put the page data, the next page and if it exists -a last page. -

-The current page will have the areas added to it from the block layout -managers. The next page will be used when splitting a block that goes -over the page break. Note: any page break overrides the layout decided -here. The last page will be necessary if the last block area is added -to this page. The size of the last page will be considered and the -areas will be added to the last page instead. -

-The first step is to add areas to the current page until the area is full -and the lines of the last block area contain at least n(orphans) and at least -n(orphans) + n(widows) in total. This will only be relevant for areas at -the start or end of a particular reference area. -

- -

-The spacing between the areas (including spacing in block areas inside -an inline-container) will be set to the minimum values. This will allow -the page to have at least all the information it needs to organise the -page properly. -

-This should handle the situation where there are keeps on some -block areas that go over the end of the page better. It is possible that -fitting the blocks on the page using a spacing between min and optimum -would give a closer value to the optimum than putting the blocks on the -next page and the spacing being between optimum and max. So if the objects -are placed first at optimum then you will need to keep going to see if -there is a lower keep further on that has a spacing that is closer to the -optimum. -

-The spacing and keep information is stored so that the area positions -and sizes can be adjusted. -

- - -

-The page is vertically justified so that it distributes the areas -on the page for the best result when considering keeps and spacing. -

- - -

-First the keeps are checked. The available space on the page may have -changed due to the presence of before floats or footnotes. The page break -will need to be at a height <= the available space on the page. -

-A page break should be made at the first available position that -has the lowest keep value when searching from the bottom. Once the first -possible break is found then the next possible break, with equally low -keep value, is considered. If the height of the page is closer to the -optimal spacing then this break will be used instead. -

-Keep values include implicit and explicit values when trying to -split a block area into more than one area. Implicit keeps may -be such things as widows/orphans. -

-If the page contains before floats or footnotes then as each area or line -area is removed the float/footnote should also be removed. This will -change the available space and is a one way operation. The footnote -should be removed first as a footnote may be placed on the next page. -The lowest keep value may need to be reassessed as each conditional -area is removed. -

-The before float and footnote regions are managed so that the separator -regions will be present if it contains at least one area. -

- - -

-Once the areas for the page are finalised then the spacing will -need to be adjusted. The available height on the page is compared -with the min and max spacing. All of the spacing in all the areas -on the page is then adjusted by the appropriate percentage value. -

- - -

-In the case of multi-column pages the column breaks and eventually -the page break must be found in a slightly different way. -

-The columns need to be layed out completely from first to last but -this can only be done after a rough estimate of all the elements -on the page in case of before floats or footnotes. -

-So first the complete page is layed out with all columns filled -with areas and the spacing at a minimum. Then if there are any -before floats or footnotes then the availabe space is adjusted. -Then each the best break is found for each column starting from -the first column. If any before floats or footnotes are removed -as a result of the new breaks and optimised spacing then all the -columns should still be layed out for the same column height. -

- - -

-After the region body has been finished the static areas can be -layed out. The width of the static area is set and the height is -inifinite, that is all block areas should be placed in the area -and their visibility is controlled be other factors. -

-The area tree for the region body will contain the information -about markers that may be necessary for the retrieve marker. -

-The ordering of the area tree must be adjusted so that the areas are -before, start, body, end and after in that order. The body region -should be in the order before float, main then footnote. -

- - -

-Creating a line areas uses a similair concept. Each inline area -is placed across the available space until there is no room left. -The line is then split by considering all keeps and spacing. -

-Each word (group of adjacent character inline areas) will have keeps -based on hyphenation. The line break is at the lowest keep value -starting from the end of the line. -

-Once a line has been layed out for a particular width -then that line is fixed for the page (except for unresolved -page references). -

- - -

-The before float region and footnote region are handled by the page -layoutmanger. These regions will handle the addition and removal -of the separator regions when before floats/footnotes area added -and removed. -

- - -

-If a float anchor is present in a particular line area then the available -space for that line (and other in the block) will be reduced. The side float -adds to the height of the block area and this height also depends -on the clear value of subsequent blocks. The keep status of the block is -also effected as there must be enough space on the page to fit the -side float. -

- -

- - -

-Once the layout of the page is complete there may be unresolved areas. -

-Page number citations and links may require following pages to be -layed out before they can be resolved. These will remain in the -area tree as unresolved areas. -

-As each page is completed the list of unresolved id's will be checked -and if the id can be resolved it will be. Once all id's are resolved -then the page can be rendered. -

-Each page contains a map of all unresolved id's and the corresponding -areas. -

-In the case of page number citations. The areas reserves the equivalent -of 3 number nines in the current font. When the area is resolved -then the area is adjusted to its proper size and the line area is -re-aligned to accomodate the change. -

- - -

-Any formatting object that has an ID or any inline link defines an area -that will be required when rendering and resolving id references. -

-This area is stored in the parent area and may be a shape that exists -in more than one page, for example over a page break. This shape consists -of the boundary of all inline (or block) areas that the shape is defined -for. -

- - -

-This is the definition of all inline areas that will exist in the -area. -

- - -

-instream-foreign-object, external-graphic, inline-container -

-These areas have a fixed width and height. They also have a viewport. -

- - -

-leader, inline space -

-These areas have a fixed height but the width may vary. -

- - -

-character -

-This is an simple character that has fixed properties according to -the current font. There are implicit keeps with adjacent characters. -

- - -

-float anchor, footnote anchor -

-This area has no size. It keeps the position for footnotes and floats -and has a keep with the associated inline area. -

- - -

-page-number-citation -

-A page number area that needs resolving, behaves as a character and -has the space of 3 normal characters reserved. The size will adjust -when the value is resolved. -

- - -

-The block area has info about the following: - - -all anchors including which lines they are on - - -unresolved page references with line info - - -id and link areas - - -height (min/max/optimum) or area including floats - - -holds space before/after and keep information - - -widows and orphans - - -

-Once the layout has been finalised then this information can be -discarded. -

- - -

-Contains inforamtion about all the block areas in the body, -before area and footer area. -

-Has a list of the unresolved page references and a list of id refences -that can be used to obtain the area associated with that id. -

- - -

-Here a few layout possibilities areas explored to determine how the -layout process will handle these situations. -

- -

-All blocks (including nested) are placed on the page with minimum spacing -and the last block has the minimum number of lines past the page end. -The lowest keep value is then found within the body area limits. Then the next -equally low keep is found to determine if the spacing will be closer to -the optimum values. -

- - -

-After filling the page with the block areas then the new body height -is used to find the best position to break. Before each line area or block -area is remove any associated before floats and footnotes are removed. -This will then adjust the available space on the page and may allow -for a different breaking point. Areas are removed towards the new -breaking point until the areas fit on the page. When finding the -optimum spacing the removal of before floats and footnotes must also -be considered. -

- - -

-First the page is filled with all columns for the intial page area. -Then each column is adjusted for the new height starting from the -first column. The best break for the column is found then the next -column is considered, any left over areas a pre-pended to the next -column. Once all the columns are finished then all the columns are -adjusted to fit in the same height columns. This handles the situation -where before floats or footnotes may have been removed. -

- - -

-If in the process of adding areas to a page it is found that there -are no more areas in the flow then this page will need to be changed to -the last page (if applicable). The areas are then placed on a last -page. -

- - - - - - - - diff --git a/docs/design/optimise.xml b/docs/design/optimise.xml deleted file mode 100644 index 0b8fcd19e..000000000 --- a/docs/design/optimise.xml +++ /dev/null @@ -1,58 +0,0 @@ - - - - - -

- FOP Optimisations - Notes for Optimising FOP - - - -

- - - -

-FOP should be able to handle very large documents. A document can be -supplied using SAX and the information should be passed entirely through -the system, from fo elements to rendered output as soon as possible. -

-A top level block area, immediately below the flow, can be added to the -page layout as soon as the element is complete. -

-The fo elements used to construct a page can be discarded as soon as the -layout for the page is complete. Some information may be stored in the -area tree of the page in order to handle unresolved page references -and links. -

-Once the layout of a page has been completed, all elements are fully -resolved, then the page can be rendered. Some renderers may support -out of order rendering of pages. -

-The main problem that will remain is that any page with forward -references will need to be stored until the refence is resolved. -This means that the information contained in the page should be -as minimal as possible. -

-Line areas can be optimised once the layout for the line has -been finalised. Consecutive characters with the same properties -can be combined into a "word" to hold the information with -limited overhead. -

-If there are a large number of pages where forward references -cannot be resolved the a method of writing a page onto disk -could be used to save memory. The easiest way to achieve this -is to make the page and all children serializable. -

- - - - - diff --git a/docs/design/properties.xml b/docs/design/properties.xml deleted file mode 100644 index 1d0744194..000000000 --- a/docs/design/properties.xml +++ /dev/null @@ -1,257 +0,0 @@ - - - - - -

- Properties - Properties overview - - - -

- - - - - -

The property datatypes are defined in the -org.apache.fop.datatypes package, except Number and String which are java -primitives. The FOP datatypes are:

Number
String
ColorType
Length (has several subclasses)
CondLength (compound)
LengthRange (compound)
Space (compound)
Keep (compound)

The org.apache.fop.fo.Property class is the superclass for all -Property subclasses. There is a subclass for each kind of property -datatype. These are named using the datatype name plus the word -Property, resulting in NumberProperty, StringProperty, and so -on. There is also a class EnumProperty which uses an int -primitive to hold enumerated values. There is no corresponding Enum -datatype class.

The Property class provides a "wrapper" around any possible -property value. Code manipulating property values (in layout for -example) usually knows what kind (or kinds) of datatypes are -acceptable for a given property and will use the appropriate accessor.

The base Property class defines accessor methods for all FO property -datatypes, such as getNumber(), getColorType(), getSpace(), getEnum(), -etc. It doesn't define -accessors for SVG types, since these are handled separately (at least -for now.) In the base Property class, all of these methods return -null, except getEnum which returns 0. Individual subclasses return a value of the appropriate type, -such as Length or ColorType. A subclass may also choose to return a -reasonable value for other accessor types. For example, a -SpaceProperty will return the optimum value if asked for a Length.

- - - -

The Property class contains a nested class called -Maker. This is the base class for all other property Makers. It -provides basic framework functionality which is overridden by the -code generated by properties.xsl from the *properties.xml files. In -particular it provides basic expression evaluation, using -PropertyParser class in the org.apache.fop.fo.expr package.

Other Property subclasses such as LengthProperty define their own -nested Maker classes (subclasses of Property.Maker). These handle -conversion from the Property subclass returned from expression -evaluation into the appropriate subclass for the property.

For each generic or specific property definition in the -properties.xml files, a new subclass of one of the Maker classes is -created. Note that no new Property subclasses are created, only new -PropertyMaker subclasses. Once the property value has been parsed and -stored, it has no specific functionality. Only the Maker code is -specific. Maker subclasses define such aspects as keyword -substitutions, whether the property can be inherited or not, which -enumerated values are legal, default values, corresponding properties -and specific datatype conversions.

- - - -

In the properties xml files, one can define generic property -definitions which can serve as a basis for individual property -definitions. There are currently several generic properties defined in -foproperties.xml. An example is GenericColor, which defines basic properties -for all ColorType properties. Since the generic specification doesn't include -the inherited or default elements, these should be set in each property -which is based on GenericColor. Here is an example:

-- <property type='generic'> - <name>background-color</name> - <use-generic>GenericColor</use-generic> - <inherited>false</inherited> - <default>transparent</default> - </property> -

A generic property specification can include all of the elements -defined for the property element in the DTD, including the description -of components for compound properties, and the specification of -keyword shorthands.

- -

Generic property specifications can be based on other generic -specifications. -An example is GenericCondPadding template which is based on the -GenericCondLength definition but which extends it by adding an inherited -element and a default value for the length component.

-Generic properties can specify enumerated values, as in the -GenericBorderStyle template. This means that the list of values, which -is used by 8 properties (the "absolute" and "writing-mode-relative" -variants for each BorderStyle property) is only specified one time.

-When a property includes a "use-generic" element and includes no other -elements (except the "name" element), then no class is generated for the -property. Instead the generated mapping will associate this -property directly with an instance of the generic Maker.

-A generic class may also be hand-coded, rather than generated from the -properties file. -Properties based on such a generic class are indicated by the -attribute ispropclass='true' on the -use-generic element.

This is illustrated by the SVG properties, most of -which use one of the Property subclasses defined in the -org.apache.fop.svg -package. Although all of these properties are now declared in -svgproperties.xml, no specific classes are generated. Classes are only -generated for those SVG properties which are not based on generic -classes defined in svg.

- - -

Properties may be defined for all flow objects or only for -particular flow objects. A PropertyListBuilder object will always look -first for a Property.Maker for the flow object before looking in the -general list. These are specified in the -element-property-list section of the properties.xml -files. The localname element children of this element specify for -which flow-object elements the property should be registered.

NOTE: All the properties for an object or set of objects -must be specified in a single element-property-list element. If the -same localname appears in several element lists, the later set of -properties will hide the earlier ones! Use the ref -functionality if the same property is to be used in different sets of -element-specific mappings. -

- - -

A property element may have a type attribute with the value - ref. The - content of the name child element is the name of the referenced - property (not its class-name!). This indicates that the property - specification has - already been given, either in this same specification file or in a - different one (indicated by the family attribute). The - value of the family attribute is XX where the file - XXproperties.xml defines the referenced property. For - example, some SVG objects may have properties defined for FO. Rather - than defining them again with a new name, the SVG properties simply - reference the defined FO properties. The generating mapping for the - SVG properties will use the FO Maker classes.

- - -

Some properties have both absolute and -writing-mode-relative forms. In general, the absolute forms -are equivalent to CSS properties, and the writing-mode-relative forms -are based on DSSSL. FO files may use either or both forms. In -FOP code, a request for an absolute form will retrieve that value if it -was specified on the FO; otherwise the corresponding relative property -will be used if it was specified. However, a request for a relative -form will only use the specified relative value if the corresponding -absolute value was not specified for that FO. -

Corresponding properties are specified in the properties.xml files -using the element corresponding, which has at least one -propval child and may have a propexpr child, -if the corresponding -value is calculated based on several other properties, as for -start-indent. -

NOTE: most current FOP code accesses the absolute variants -of these properties, notably for padding, border, height and width -attributes. However it does use start-indent and end-indent, rather -than the "absolute" margin properties. -

- - - - -

The XSL script propmap.xsl is used to generate -property mappings based on -both foproperties.xml and svgproperties.xml. The mapping classes -in the main fop packages simply load these automatically generated -mappings. The mapping code still uses the static -"maker" function of the generated object to obtain a Maker -object. However, for all generated classes, this method returns an -instance of the class itself (which is a subclass of Property.Maker) -and not an instance of a separate nested Maker class.

For most SVG properties which use the SVG Property classes directly, -the generated mapper code calls the "maker" method of the SVG Property -class, which returns an instance of its nested Maker class.

The property generation also handles element-specific property -mappings as specified in the properties XML files.

- - - -

For any property whose datatype is Enum or which -contains possible enumerated values, FOP code may need to access -enumeration constants. These are defined in the interfaces whose name -is the same as the generated class name for the property, -for example BorderBeforeStyle.NONE. These interface classes -are generated by the XSL script enumgen.xsl. A separate -interface defining the enumeration constants is always generated for -every property which uses the constants, even if the constants -themselves are defined in a generic class, as in BorderStyle.

If a subproperty or component of a compound property has enumerated -values, the constants are defined in a nested interface whose name is -the name of the subproperty (using appropriate capitalization -rules). For example, -the keep properties may have values of AUTO or FORCE or an integer -value. These are defined for each kind of keep property. For example, -the keep-together property is a compound property with the components -within-line, within-column and within-page. Since each component may -have the values AUTO or FORCE, the KeepTogether interface defines -three nested interfaces, one for each component, and each defines -these two constants. An example of a reference in code to the constant -is KeepTogether.WithinPage.AUTO.

- - - - -

Some XSL FO properties are specified by compound datatypes. In the FO file, -these are defined by a group of attributes, each having a name of the -form property.component, for example -space-before.minimum. These are several compound -datatypes:

LengthConditional, with components length and conditionality
LengthRange, with components minimum, optimum, and maximum
Space, with components minimum, optimum, maximum, precedence and -conditionality
Keep, with components within-line, within-column and within-page

These are described in the properties.xml files using the element -compound which has subproperty children. A subproperty element is much -like a property element, although it may not have an inherited child -element, as only a complete property object may be inherited. -

Specific datatype classes exist for each compound property. Each -component of a compound datatype is itself stored as a Property -object. Individual components may be accessed either by directly -performing a get operation on the name, using the "dot" notation, -eg. get("space-before.optimum"); or by using an accessor on the compound -property, eg. get("space-before").getOptimum(). -In either case, -the result is a Property object, and the actual value may be accessed -(in this example) by using the "getLength()" accessor. -

- - - - - diff --git a/docs/design/renderers.xml b/docs/design/renderers.xml deleted file mode 100644 index 1fe18c32a..000000000 --- a/docs/design/renderers.xml +++ /dev/null @@ -1,30 +0,0 @@ - - - - - -

- Renderers - Design of Renderers - - - -

- - - -

-A render is primarily design to convert a given area tree into the output -document format. It should be able to produce pages and fill the pages -with the text and graphical content. Usually the output is sent to -an output stream. -

-Some output formats may support extra information that is not available -from the area tree or depends on the destination of the document. -

- - - - - diff --git a/docs/design/status.xml b/docs/design/status.xml deleted file mode 100644 index 93769ecc5..000000000 --- a/docs/design/status.xml +++ /dev/null @@ -1,73 +0,0 @@ - - - - - -

- Design Status - Current Status of FOP and Design - - - -

- - - -

-Currently some of FOP is being re-written so that the layout can be handled -properly without the problems that have been encountered and to make -it possible to handle keeps/breaks and spacing better. -

-A number of issues have been dealt with in the redesigned code. -

- - -

avalon integration - logging, configuration, component management, caching, uri resolver
improved interfaces -
- no threading/static problems
- simpler to use
-
better commandline handling

- - -

-Many of these changes will make FOP much better to use and develop. -So if you help out with the development you can get to use these -things sooner. -

better layout design - that can handle keeps and -spacing better
better area tree handling - means it can run in -less memory
better image handling - redone so it can use a -cache and synchronizes properly only on the current image while loading
svg now in an xml handler, FOP can be used without batik
bookmark extension improved a bit - changed bookmark extension, -now requires a wrapping element bookmark
new interface for structured documents, rtf and mif
better handling of unknown xml and xml from an unknown namespace
Changed extensions to allow for external xml
Can have a default element mapping for extensions
mathml extension
another xml -> svg extension
svg text normal text if that can be handled otherwise stroked -this is done automatically
new area tree model
new renderer model
added handlers for xml in renderer
changed area tree xml format to match the area tree hierarchy

- -

- - - - - diff --git a/docs/design/useragent.xml b/docs/design/useragent.xml deleted file mode 100644 index cf3bea97c..000000000 --- a/docs/design/useragent.xml +++ /dev/null @@ -1,176 +0,0 @@ - - - -

- FO User Agent - Design of FO User Agent - - - -

- - - -

-Technically the user agent is FOP in the role of determining the -output format and when resolving various attributes. The user -agent is represented by a class that is available to others to -specify how FOP should behave. -

-The user agent is used by the formatting process to determine -certain user definable values. -

-It will enable the customisation of values for generating and -rendering the document. -

-The user agent must be available to the layout processor and -the renderer. Users can supply their own user agent or use -the default one for a particular renderer. -

-The user agent needs to be made available to the property -resolution layout process and the renderer. -

- -

-Standard Features: -

-error handling, what to do if fo markup is invalid -
-auto overflow value and handling error-if-overflow -
-adjusting length values (eg. for borders) to renderable values -
-available fonts -
-converting cm/in to pt (dpi) -
-active state for multi properties -
-title, used to identify a set of pages (in a page sequence) -
-the width (in inline-progression-dimension) of a character with -treat-word-as-space true -
-maximum space used by conditional areas from region-reference-area -
-if there should be "hot links" to before floats or footnotes -
-when to clear side floats if space in inline-progression-dimension -is not enough -
-placement of left over footnotes on a page with a region-body -
-using color property as border colour -
-interpretting all border styles (except outset) as solid -
-thin, medium and thick values for border width -
-initial font-family value -
-absolute font sizes (eg, xx-small, x-small etc.) -
-relative font sizes (eg. larger, smaller) -
-small caps simulation -
-font weight mapping -
-baseline info for a font if not available -
-determining sub/superscript when another baseline is dominant -
-scaling method for external-graphic and instream-foreign-object -
-the width of a replaced element -
-"normal" line height value -
-text alignment (stretching the line with letter and word spacing) -
-text transform -
-initial color -
-rendering intent of auto -
-dot character for leader -
-line breaking with leaders, use optimum length when breaking the line -
-page height/width of auto -
-left and right caption widths -
-glyph orientation vertical of auto -
-rendering processor of content-type (mime type) -

- -

-Interactive Features: -

-inline and block scroll amount -
-dynamic effects, links and property sets -
-initial "pause-after", "pause-before" and "voice-family" value -
-treating fixed as scroll on background attachement -
-media usage of auto -

- - - - - - -- 2.39.5