OK, see my responses inline below. Michael Whapples On 14/10/2012 00:50, John J. Boyer wrote:
MW: Really, I thought BrailleBlaster was an application for producing high quality tactile documents. I thought that the word processing facility was just part of the application assisting in reaching that goal. If making a word processor with Braille facilities, then why not take an existing word processor, which probably will be far more advanced than we could hope to produce ourselves, and create an add-on. MS Word being the most well known, OpenOffice/LibreOffice being probably the best known open source alternative, but there are others like lotus symphony.Michael, Your view of BrailleBlaster is more restrictive than mine, and I think than most of the people who are working on it. My understanding of the original specification is that BrailleBlaster is to be a word processor with special Braille facilities.
MW: Yes and no. A sighted user even in WYSIWYG editors do see the structure of the document, the problem is that sometimes authors/authoring tools don't produce proper structures and fake the look of the structure. So My point was that the WYSIWYG editor of BrailleBlaster needs to highlight (not screen highlighting, but rather just drawing attention to) what is actual structure and what is not. Again its a representation of information.The print view is simply a standard WYSIWYG eord procesor. The user does not normally see the structure of the xml document any more than the user of MSWord sees the structure of a doc file.
MW: As an example, sometimes authors will fake the look of a heading by applying text attributes to a block of text rather than specifying an actual heading. In such a case, MS word's WYSIWYG editor will know the difference of this to where the author used a heading style, where a heading style is used a screen reader will actually announce heading information, where as for the case of altered text attributes it won't (also Word can use things marked with heading styles to create table of contents, etc). However, visually the two ways may look the same, and so the structure is less possible to see.
MW: Coming back to BrailleBlaster, the editor needs to draw the user's attention to these sorts of deviation from proper structuring. Continuing with the heading example, as far as I know font size has no bearing on what Braille will be produced, therefore it need not show the font size changes, if headings are shown larger (which is a fairly normal print convention) then the structure can be seen in quite an intuitive way. The user may not believe they are seeing "structure" but they are.
MW: Also, I must make it clear, they would not be seeing the XML structure, they would be seeing the document structure. The subtlety being that they would not be seeing DAISY, ePub, docbook, Word doc, etc structure, they would be seeing the abstract document (probably the document model) structure, or more accurately a representation of the structure.
MW: Well reducing duplication is normally not stated as a project goal, its just a good practice principal. It generally makes things much more maintainable, fewer places to fix discovered bugs, fewer places to add enhancements, greater chance things will exhibit consistent behaviour, etc. Having looked at things, I feel there is even unnecessary duplication inside liblouisutdml, configuration files just cover too much, I mean cover too many processes, linking stuff which probably should never be linked. Just because I desire using a different paper size, why do I need to redefine things like XML processing? Surely XML processing has nothing to do with my output paper size?Reducing duplication between BrailleBlaster and liblouisutdml is not a goal. The goal is clarity. That is the reason for separate files. In liblouisutdml there are cases of blocks of code that are similar but that have enough difference that combining them would reuire a lot of extra code and probably a much greater chance for bugs. The same is true to an even greater extent between liblouisutdml and BrailleBlaster.
MW: If clarity is what is being sought, may be its time to decide taking a different approach. I seem to remember when you first sent information about BrailleBlaster configuration and semantic action files there being confusion, now I have been confused, this doesn't seem to hold good indications for clarity.
MW: I just await those questions like: "I have changed my BrailleBlaster configuration yet the Braille output is not changed, why is this so?" Will people really grasp the idea that the one piece of software they downloaded has two sets of configuration files, will they be aware when only one needs changing and when both will?
MW: I just feel its unmaintainable, both technically and from user experience/support, therefore for this corner of BrailleBlaster I have to say, I have said my view and if you pursue it then I may be best to not be involved with this part so I don't have conflicts (how can I support something I don't believe in).
Sighted users do like to get hard copies of documents. They will expect a reasonable format, though not, of course, all the formatting of MSWord. I would like to see others join in this discussion. John On Sat, Oct 13, 2012 at 08:03:02PM +0100, Michael Whapples wrote:This is just puzzling me further. I think there must be some fundamental parts of BrailleBlaster which are not properly defined. While documentation can seem burdensome, it sometimes is necessary to ensure people are talking about the same things. May be BrailleBlaster needs more documentation and may be a glossary for what certain terms mean. In this spirit, I have quoted certain phrases in the following message and given definitions for what I understand these terms to mean (see end of my message for my glossary). First thing is, may be semantic action files are doing too much, in this I mean both liblouisutdml and BrailleBlaster semantic action files. When one looks at BrailleBlaster and liblouisutdml there are certain things which are common to both (eg. loading a document into a "document model" from the "original source"). In trying to minimise duplication then these common actions should be made common. While the ideal case might be the code and the rules for how to do this would be shared, it may not be possible to have common code, but if the rules are defined in external files then these should be reusable. I would argue that the code could even be common, while liblouisutdml may not want to become specific to BrailleBlaster, BrailleBlaster is certainly highly linked to liblouisutdml, so why couldn't that offer a function to applications for loading a "document model". Obviously this would need the Java bindings for liblouisutdml to know how to convert the C "document model" into a Java "document model" and probably back. So what I am saying is, may be semantic action files need to be changed as to what they work on, rather than working on "original source" they should work on a "document model". Then the files defining the rules for mapping from "original source" to "document model" could be common and so no duplication there. Now to the fundamentals which aren't making sense to me. Well I will only mention one, as its relevant and I don't want to spend too much time writing this email. What is the "print view"? What is the "print view" for? How does the "print view" add to BrailleBlaster goals? What is the "print view" not? My answers to these say that the suggested configuration files contain stuff they should not need to contain, hence why I am asking these questions in case the definition has moved on. Here are some answers from my understanding: 1. The print view is a representation of the document using "print text characters". The print view will allow the user to easily see the structure of the document according to the actual structural definition. 2. The print view will be used by the user to identify where the "original source" did not give specific structural information and so will allow the user to correct this so that when the document is translated the Braille will contain the correct layout. An example of where a document might not contain correct structural information is where the author specified fonts and other text attributes for a heading rather than using the correct mark up for a heading. 3. By allowing the user to identify where the document does not contain sufficient structural information to allow a good quality Braille translation, it will assist the user in ensuring that the document has correct structural information and so hopefully lead to better quality Braille documents being produced. 4. The "print view" is not designed for "pretty print documents". If a user desires a pretty print document to read there document in print form then they may be advised to find alternative software designed for that purpose. Equally the "print view" would not be for creating perfectly formatted hard copy pretty print documents, again other software may be better designed for that task. To implement any of this answer in BrailleBlaster's print view would not contribute to the goals of BrailleBlaster as I understand them, hence why they are what the print view is not. Taking my answers, particularly to the question of what the "print view" is not, I have questions over why BrailleBlaster configuration files need page information. Why would someone be printing the "print view", if the view is purely on screen then why does it need splitting into pages? The only case I can see for when a page break needs to be seen in the "print view" is when the document contains an explicit page break. On screen need not have paper size, margins, etc, these are all just unnecessary. Again, may be configuration files are covering too much, may be they need to focus on specific parts of the process. As an example, when would it ever be needed that BrailleBlaster would need internet access to load the document but liblouisutdml would not? Surely this information could be shared? In fact, is there anything in the xml section of a liblouisutdml configuration file which would not be relevant to BrailleBlaster? So in seeking maximum reuse, why cannot BrailleBlaster use a liblouisutdml configuration file? May be correct uses of sections in the configuration files would be needed, eg. a [BrailleBlaster] section which liblouisutdml will happily ignore. These configuration files could be provided by the BrailleBlaster project if you don't want liblouisutdml to ship with this extra stuff, also correct uses of include statements could help here in keeping the separation you want. Also the liblouisutdml specific stuff is actually relevant to BrailleBlaster here, BrailleBlaster will need to provide a user friendly way of configuring liblouisutdml, so why not have it that BrailleBlaster works on one file? Glossary: Document model: An abstract representation of a document. It should be a common model regardless of where the document came from or what the "original source" format is. There may be different implementations of a document model for different programming languages, but it might be conceivable that there may be ways of passing thenm between programming languages if such interoperability was desired. Original source: This is the format of a particular document, examples might be epub, docbook, daisy, XHTML, and so on. Pretty print documents: This is a print document as one would desire if producing it for a sighted audience, IE. how a publisher would intend to make a book look, making an article presentable for publishing, etc. Print view: The visual view of the document in BrailleBlaster, where the text will use print text characters and the view will allow identification of the document's structure. The print view will allow editing of the document, this is mainly to allow the user to correct minor text issues (eg. spelling mistakes) and correct formatting (eg. if the original source presented a heading by increasing the font, etc rather than specifying the correct structural element for a heading). Print text characters: These are the print characters of the alphabet. Most commonly the characters within the latin character set, but could also be Chinese characters, etc. Michael Whapples On 13/10/2012 17:03, John J. Boyer wrote:well, there is a lot of misunderstanding. I misunderstood your presentation in your first message, thinkikng it was not well-thought-out. Here is more explanation. BrailleBlaster semanntic-action files are used to construch a document model. This model is then used for displaying and editing both the print and Braille views. Braille rules vary wildly from one country to another. It is less confusing to have a fairly standard print display of the document. The user can always see how a block of characters is rendered in Braille by focusing on the Braille window. This will show the part of the print text corresponding to each line of the Braille in the print window. There is a great deal in liblouisutdml semantic-action files that is meaningless to BrailleBlaster. Similarly, the structure of semantic-action files that is best for BrailleBlaster would not work in liblouisutdml. Trying to combine the two to keep down the number of files would lead to a great deal of confusion. There is already a lot of confusion between what pertains to print and what pertains to Braille. liblouisutdml and BrailleBlaster development should remain separate, because liblouisutdml and liblouis are intended to form a transcription engine that can be used in almost any application. Of course, liblouisutdml can be enhanced with functionality that we discover is needed while developing BrailleBlaster. An example would be a function to report how much of a transcription is finished as a percentage, so that it can be used in a progress bar. John On Sat, Oct 13, 2012 at 04:11:31PM +0100, Michael Whapples wrote:I did put thought into the reply, to say it wasn't well thought out to me feels insulting. Whether I understood everything fully may be a question, if there are misunderstandings then we would do better in trying to explain it better. I am still unclear to really what the difference of semantic action files for BrailleBlaster and liblouisutdml are. Yes print is not Braille and vice versa, but both are generated from documents. So map from the document model to the view, why have the view go right back to the original XML? Documents may come from many sources, so why not map from the file format to the document model? This surely would reduce duplication of how to process a particular flavour of XML as there would only be one definition of how to process it. With what I have suggested, there would be files to map from file formats into the document model and files to map from the document model to the user's representation of the document (print or Braille). With what I am suggesting, add a new supported document format, you only need to add one file to map from the format to the document model, rather than one for each user's representation (at current that would be two files). Want to add another user's representation, my idea would mean add one mapping, under what has been proposed I understand it to require as many files as formats supported. May be the current design of BraileBlaster and liblouisutdml where they are fairly separated would not allow exactly what I suggest, but still why do they need separate files to use? Isn't there a way to allow them to share semantic action files? This would at least reduce the work in supporting various formats. There may be features of the files not used by both (eg. the files may specify editing information which liblouisutdml does not need), but any uneedede information could be ignored by that tool (liblouisutdml could ignore editing information). Further to the above, I also reject the idea that BrailleBlaster's GUI is producing print documents. This is quite a subtle point, the rejection is that we are not producing software which sighted people would choose to use to print documents because the visual output is so pretty, rather the visual aspect is more to show the document structure so that a sighted person will understand how the document will be translated (IE. a heading need not look like a print heading as in a print book, they just need a way that they can easily distinguish there is a heading which will appear in Braille). My point here is that the visual display needs to represent what should happen in the translation process, so may be the two are fairly linked in what they need to show, IE. they are just different ways of showing the same information. I guess a better phrase to describe what I mean, BrailleBlaster's GUI shows a print representation of BrailleBlaster's document, liblouisutdml will produce a Braille representation of that document. Ideally the Braille representation needs to conform to official Braille rules (I mean more than just the translation, I mean follow the standard for formatting), so may be the BrailleBlaster document model will be swayed towards Braille rules. Notes on what I have used, to show this was thought out: * BrailleBlaster semantic action files are tied to the underlying document format: Appendix B says "the key part of a line in a semantic-action file is a reference to markup in the document. This may be literal markup or an XPath expression". * Section 4 of the liblouisutdml documentation has the title "Connecting with the XML document - Semantic action files", this sounds very much like it goes right back to the underlying XML as well, also supported by mentions of using xpath expressions to identify nodes, etc. * My statement of documents should go to a intermediate document model, so that the various views need not be concerned with the underlying document format, this is a key part of object orientation. If the document model were represented by a interface, then the view only needs to know that what it has been handed implements that interface, it need not know how it works or even what class type it is. This is why frameworks which use dependency injection, such as the spring framework, can be extended so easily, although it must be noted that I am not suggesting we go as far as using dependency injection or spring framework. * My thought on BrailleBlaster's GUI and the Braille being just different representations of the same information, a similar example might be having data in a table and the same data plotted on a graph, it may look very different but it contains exactly the same information. It must be noted that even for some data there may be a number of different choices of what sort of graph could be used (eg. peoples' responses to a multiple choice question, it could be shown on a bar chart, it possibly could be shown in a PI chart). * Further to that last point, the data may be held in many ways, it might be in an excel spreadsheet, it might be in a database, it might be in a CSV file, may be in XML, etc. It just seems crazy duplication to map straight from the input to the output rather than going to an intermediate model (eg. for responses to a multiple choice question I have identified three outputs and I have named 4 input formats, direct mapping requires 12 mappings, mapping to an intermediate model would only require 7 in total, 4 mappings for input and 3 mappings for outputs. It gets worse as you add more inputs/outputs, add another output you will need 16 direct mappings, but only 8 if going through a common model, increase outputs to 5 and you need 20 direct mappings but only 9 if going through the model, and so on). * How many inputs and outputs do we wish to support in BrailleBlaster? There are certainly a large number of inputs, there are many file formats out there and as I mentioned it is possible to think that other outputs could be conceived (eg. conversion to speech document). I certainly do not wish to support such a level of duplication as my calculations seem to suggest may need supporting in BrailleBlaster. Michael Whapples On 13/10/2012 02:36, John J. Boyer wrote:The configuration files are analogous to MSWord templates. The semantic-action files enable BrailleBlaster to work with any flavor of xml. The different flavors all use different markup. BrailleBlaster uses a different set of configuration and semantic-action files than liblouisutdml because print is not Braille and BrailleBlaster also does editing. The algorithms in liblouisutdml have been successful, so it is worthwhile to consider adapting them for wider application. The reply does not seem to be well-thought out. John On Fri, Oct 12, 2012 at 08:42:58PM +0100, Michael Whapples wrote:I guess the main thing which crosses my mind is that there seems to be quite a bit of duplication between BrailleBlaster and liblouisutdml, particularly configuration files and semantic action files. If focusing on configuration files and semantic action files, I just don't get the difference of those for BrailleBlaster and liblouisutdml. OK, its said that the BrailleBlaster ones are for how BrailleBlaster displays documents and the liblouisutdml ones are for how Braille is produced, but what is the difference, surely the idea of the GUI of BrailleBlaster is so that people can see the structure of the document and how that might be handled for being put into Braille? Why does processing for the specific output have to always go right back to the underlying document format? Let me give some examples: Documents normally contain certain common features, headings, paragraphs, lists, emphasised text, etc. All the GUI is interested in is whether it has a heading, a list, a paragraph, etc, it should not be interested in how XHTML, docbook, epub, etc store these elements. Equally, there are two outputs (at current) which the user may encounter, the BrailleBlaster GUI and Braille/tactile document. Requiring a mapping from file format to output system leads to a large number of files required. In this case where there are two views (BrailleBlaster's GUI and Braille from liblouisutdml) we need twice as many as supported document formats. If additional document views were added (eg. may be someone wanted to add a way to produce a spoken version using TTS, not suggesting it as an actual feature at this point) then another whole set of files for all supported formats would need producing. I guess I am thinking down the MVC design pattern, the document is the model and implements a common model regardless of where it came from, the views being BrailleBlaster's GUI and the Braille document (at the moment) and the controller being there simply to introduce the two together. If a new document format was supported then neither view needs any modification, if a new view were added then none of the support formats would be affected. Sounds much more maintainable. Additionally with what was originally suggested, I get the feeling that BrailleBlaster's GUI could show document elements differently depending on what file format is being viewed (IE. a heading may look different if viewing a epub rather than docbook). OK, one might work to try and not let this happen, but the potential is there from what I understand. Just my views. Michael Whapples On 12/10/2012 18:39, John J. Boyer wrote:Here is the block diagram with some clarifications. I would appreciate comments. It is important to be quite sure that this is the best approach for BrailleBlaster. -------------------- This is a sort of informal block diagram in narrative form. It is intended as a guide for the future development of BrailleBlaster. Input may be in the form of an xml file or it may be a utd working file which has been saved so that work can be resumed later. xml files may be original well-formed files of any flavor, for example, dtbook, docbook, etc. They may also be derived from other formats such as MSWord, rtf, and so on. Another source is file sets, such as NIMAS or epub. In this case the manifest is opened, and the file to be processed is chosen from it. A means is provided to concatenate several files into one. xml files can also be derived from plain-text by calling the translateTextFile method with formatFor utd or from brf files by calling backTrnslateFile, also with formatFor utd. Whatever its source, an xml file is rendered by calling translateFile with formatFor utd. BrailleBlaster then works on the utd file produced. In the case of formerly saved utd files or those produced by importing plain text or brf files, BrailleBlaster works with these directly. The file is first parsed to produce a parse tree. The configuration files indicated in the user's settings are then read and used to begin the construction of a semantic table. This table is used to specify how markup in the document is to be rendered on the screen and how styles and actions are to be associated with markup for editing. For more on configuration files see Appendix A. Semantic-action files are then read. A file is chosen by looking for a file with the name of the root element and the extension .sem or according to an indication in the configuration files. The information in the semantic-action files is used to complete the semantic table. For more on semantic-action files see Appendix B. If the semantic-action files contain XPath expressions as keys these are applied to the parse tree, and the selected nodes are modified by adding an attribute indicating the entry in the semantic table to be used. The value of each key will already have been entered into this table. The keys containing markup in the semantic files are then applied to the parse tree, and a similar attribute is added to the matching nodes, unnless it is already present because it has been added by an XPath expression. This forms the DOM of the document. This DOM is then used to display the document on the screen. Both the print and Braille windows are filled in this process. Editing of the print document can then take place. If the contents of a text node are altered the new contents replace the old. They are also dynamically translated and the translation is shown in the Braille window. If an element node is deleted its entire subtree is deleted. If a new block of characters is created the user is prompted to asign it a style and a node with the appropriate markup (derived from the semantic-action files) is added to the document at the place where the new block was created. if focus is shifted to the Braille window and the user has checked the Edit Braille box on the advanced menu the window can be edited. Any editing is highlighted in both the Braille and print windows. The print window also changes to show the part of the original text that corresponds to each line of Braille. Dince the user may wish to view the result of editing Braille in the context of the entire document, The translate and back-translate items on the menu are replaced with retranslate and reback-translate. The file can be saved as a utd file so work can be resumed later or it can be saved as the original xml file with enhancements. These consist of edited Braille which has been moved into the print document with proper markup (specified in the semantic-actionn files). The attributes used to create the DOM are removed during the save process. Editing in the print window is handled automatically as part of the conversion of the parse tree to a file. Besides saving the file as utd or as an envanced document, it can be saved as a brf file or embossed. The enhanced document can then later be rendered by liblouisutdml with any liblouisutdml configuration and semantic-action files that the user wishes. This can be done either by BrailleBlaster or by another application which uses the loblouis-liblouisutdml transcription engine. Appendix A: Configuration Files Here is a description of BrailleBlaster configuration files. They specify how BrailleBlaster shall operate. The one which the user has selected is referenced in the userSettings.properties file. Selection, creation and editing are done via a dialogue. Note that BrailleBlaster configuration files are completely separate from liblouisutdml configuration files. They are stored in the configurations directory within the programData directory. Below is a sample file. More lines will be added as development proceeds. Explanations follow the sample. -------------------- hyphenate=no pageNumberAt=top topMargin=0.5 leftMargin=1 rightMargin=0.5 bottomMargin=0.5 paperHeight=11 paperWidth=8.5 xmlheader="<?xml=version='1.0'=encoding='UTF-8'=standalone='yes'?> semanticFiles=* internetAccessRequired=no newEntries=yes file:filename include debug=no styles=para,heading1,heading2,list -------------------- The first line specifies whether hyphenation should be used when a document is displayed onscreen. The second line specifies where the page number should be placed. Then come six lines specifying margins and paper size. These are used principally for making hard copies. the xmlheader line specifies the header that should be used at the beginning of new files. semanticFiles is a comma-separated list of the semantic files to be used. Here the value is an asterisk, which means to use the file which is named for the root element of the document. This is actually the default, so this line is not strictly necessary. InternetAccessRequired specifies whether the Internet will be needed for processing a document, e.g., for getting a DTD. newEntries specifies whether a record should be kept of the markup in a document which has not been assigned to a semantic action or style. This is useful when processing a new flavor of xml for which a semantic-action file does not exist. It applies to all documents. Semantic-action files can also have a newEntries line. It applies to that type of document only. The include line, with a preceding filename, specifies that another configuration file is to be read at the point where it is encountered. debug specifies whether extra testing should be done, experimental features used, and extra logging done. The styles line gives a comma-separated list of styles that will be used in displaying and editing a document. These styles are in the styles directory. The order of entries in a configuration file is generally not important. However, an include line does cause the referenced file to be read at the point where it is encountered. Appendix B: Semantic-Action Files Semantic-action files associate the markup in a particular type of xml document with BrailleBlaster styles, methods (actions) and macros. Usually they are named by concatenating the name of the root element of the document flavor with the extension .sem They are not to be confused with liblouisutdml semantic-action files. The latter are concerned with rendering an xml document into Braille and tactile graphics. BrailleBlaster semantic-action files are concerned with displaying the contents of a document on the screen and with editing them. They are stored in the semantics directory in the programData directory. There is a dialogue for creating and editing them. the key part of a line in a semantic-action file is a reference to markup in the document. This may be literal markup or an XPath expression. There are a few exceptions, which will be discussed later. The value part contains the name of a style or action, or of a macro, which can combine several styles and actions. it may also contain parameters. Literal keys may have one of the following forms: an element name; an element name, a comma, and an attribute name; an element name, a comma, an attribute name, an attribute value. XPath keys begin with the characters &xpath with the XPath expression imediately following and enclosed in parentheses. The key may also be the word newEntries. If the value is yes markup which has not yet been associated with anything is recorded and placed in a prototype semantic-action file. The key may also be the word file, followed by a colon followed by a filename. In this case the value is the word include, and the line specifies that another semantic file should be read at this point. Values start with one of the words action macro style. This is followed by a space. If action is specified, the action is one of those below. If style is specified a style name follows. The extension .properties is added to it and it is looked up in the styles directory. Likewise, macros are looked up in the macro directory. All three may have parameters preceded by a space and separated by comas. The following actions may be specified. no, Do nothing except insert a space. skip, Skip the subtree of which this markup is the root. generic, Apply the parameters. cdata, Special processing for CData sections. htmllink, Insert a link htmltarget, to this target. configfile, Specify a configuration file. configstring, Specify a configuration string. configtweak, Use new configurations while rendering. # reserved styles. These styles are predefined, but may be altered document, Assign to the node that actually contains the content, such as the book node in dtbook. para, A paragraph. heading1,various levels of headings heading2, heading3, heading4, heading5, heading6, heading7, heading8, heading9, heading10, contentsheader, The heading of the table of contents. contents1, Various levels within the contents. contents2, contents3, contents4, contents5, contents6, contents7, contents8, contents9, contents10, # General text attrtotext, Transform an attribute value to text. runninghead, Specify a running header. footer, Specify a page footer. boxline, Specify a line of identical characters. italicx, Italicise the text within the markup. boldx, Bold it. underlinex, Underline it. linespacing, Number of blank lines between lines of characters. blankline, Leave a blank line. softreturn, Start a new line, but not a new style. newpage, Start a new page. brl, Process the <brl> subtree rooted at this node. music, Display/edit the music notation in this subtree. math, Display/edit the MathML notation in this subtree. chemistry, Display/edit the chemistry notation in this subtree. graphic, Display/edit the graphic pointed to by the markup.