ArcWeb Version 0.11

Version 0.11

• Embedded TAB characters are OK
• <pre> fixed again to honour empty lines
• There are now 4 different ordering system for nested ordered lists, and 4 different bullets for nested unordered lists.

Version 0.10

• Parser now handled <PRE> tags correctly, and does not make the (wrong) assumption that 2+ consecutive linefeeds imply a paragraph break.
• Delayed image loading partially implemented
• Changes to the frontend: URL is shown in title of document window; -iconbar option introduced to ArcwebImg & ArcwebLcl.

Version 0.09

• Hotlist window added, but it isn't very intelligent (it doesn't resize itself when it can't fit all the entries into it).
• Support for Replay movies added to ArcWebImg
• Latin1 character entity references can now be terminated by whitespace. This is not correct HTML, but is handled by Mosaic, and is produced by some broken providers.
• Fixed handling of % characters when they are not followed by two hex digits, also spaces are allowed in arguments to tags - eg. <a href = other.html > is allowed. In version 0.08 this caused a Fatal Error type=5.
• Special parameter settings to ChangeFSI for Risc PC when BASIC's MODE function returns a pointer to an information block, rather than a mode number. (Code by Julian Wright)

Version 0.08

• First FTP'able version. This version is funtionally the same as 0.07.

Version 0.07

SWI chunk, Wimp message chunk and error chunk are now officially allocated:
• SWI chunk: &4A240 (Arcweb)
• Message chunk: &4A240 (Message_Arcweb)
• Error chunk: &80D200 (ErrorBase_SBrodie)
• Fixed bug in requester.c which meant that link flag was always set in the cache when a document fetch is initiated, regardless of whether the target was already in the cache and not a link. This resulted in spurious files appearing in the ArcWeb directory when using the directory browser.
• Optimised parsing & rendering of non-HTML text format files. Files are no longer parsed into the usual intermediate form, but left intact.
• Resolver for non-riscos: URLs is now implemented. This should allow relative URLs to be used in other types (eg. http: URLs)
• A potential problem with SparkFS archives in specific formats has arisen which results in ArcWeb being unable to read files from within the archive. (ArcFS 2 may also be affected by this).

Version 0.06

• Render code changed to align objects properly. The images respond to align arguments in the <img> tag. As a result, an even line spacing is available between lines (sometimes when no there were no descenders, the text was misaligned).
• Cache stores expiry dates for fetched documents and images. Every 60 seconds the cache is checked and old documents expired. Currently, images are not expired at all (there is a very good reason for this, which will be fixed in the next version hopefully).
• Catches DataOpen messages for HTML files and accepts DataLoad's for a range of system filetypes which are known to be ASCII text files. (eg. )
• Activity indicator added to reoganised information window, although you will never see it if the fetchers and renderers do not multi-task while working. This does not affect the icon bar icon in this version.
• Active window list window added (with configurable fonts etc.)
• !ArcwebImg introduced. Calls ChangeFSI to convert images into sprites. The setup window allows the user can change some of the options. Unfortunately, the palettes are wrong for palette-less Wimp sprites. I can display JPEGs and GIFs, but have had major problems with RLE files (ChangeFSI gives me rubbish - but is this GIGO?). Risc PC mode numbers are causing me a problem, as I don't know what to give to ChangeFSI on the command line. Currently, I use the BASIC MODE function and if the result > 256, then I default to mode 28. There must be a better way ...
• Canonicaliser will now check that a given filing system is present before calling OS_FSControl 37. This should stop the backtraces being given when an unavailable RISC OS filename is specified. Instead, you get the root on the current drive (which hopefully will be there!)
• The extent of the window is not adjusted to take into account any extra width required by non-wrappable items - ie. wide images.
• Full interactive help supported by !Arcweb, !ArcwebLcl and !ArcwebImg (including on menus). Use !Help for precise help on menu items and icons.
• Setup window added to !ArcwebLcl to control the new feature of autobooting applications for which it does not have a sprite (when it is constructing directory listings); and also the auto-quit feature; and the expiry delay. Configuration stored in a Choices file.
• All fetches (I hope) are now terminated correctly. In previous versions, this resulted in ArcWeb sitting in a dead state not able to do anything because as far as it knew, a fetch was still in progress.
• Some of the framework for handling forms is now in place, but nothing more than that.

Version 0.05

• Paragraph handler put in to recognise <p> tags
• Now copes with empty arguments in tags (it was getting caught out by empty alt arguments in img tags
• Replaced web_calloc with one which calls web_malloc and then exits via memset. This was done due to a possible bug in the C library supplied with C5 compiler (alpha version) which meant that memory wasn't being zeroed in calloc.
• Inserted extra code to set the link list pointer to NULL on page creation. This was causing address exceptions when combined with the point about calloc above.
• Comment parser now accepts an end of line as a comment terminator if no proper --> symbol is found before the end of file. NOTE: If the author has mixed comment styles, then the terminator for a subsequent comment may be found in the case where there is no terminator for the first comment.
• Corrected bug in requester.c where the RISC OS handle to the diagram file wasn't being closed before an fopen call for write access was made on the file. This meant that the fopen failed and the diagram did not get saved to the cache.
• Directory browser now clears the symbolic link bit when it generates a directory listing
• !ArcwebLcl no longer tries to generate an HTML error file if the fetch was an error fetch. Instead, it sets the error bit.
• No error is generated if the Message_ArcwebFetchDone is received with the error bit set, the error handled bit clear and the image fetch bit set. Instead, the image fetch is terminated properly (with a failure) and the alternate text is inserted if there was any.

Version 0.04

• Parser augmented to parse start tags so that it can determine whether to go into comment mode, or verbatim. The code to do this has moved from render.c into parse.c and is thus available to the standalone parser.
• Strikeout capability added
• When document is rendered, the window width is precalculated so that the document fills the whole visible window width (not the extent). You can now resize the window, and click Reload and the image will be adjusted.
• Destination anchors are supported within the same document. For example, you could view the notes for version 0.01 (check the destination with ADJUST).
• Need to fiddle alignment so that text bases all line up correctly. At the moment, sometimes there is a misalignment with bold and emphasized text.
• Images are now fetched using the Fetcher protocol
• Cache Manager symbolic links re-enabled (they were disabled in v0.03 due to me not having put in an API to reverse the setting of the symbolic link bit.
• !ArcwebLcl changed to use the new sprite naming (see for more details)
• The _kernel_register_slotextend function is now called upon initialisation so that the flex blocks can be moved to make way for the heap and stack.

Version 0.03

Parser now recognises (but does not necessarily act upon) all tags given as mandatory in the HTML 2.0 specification. Most of the optional tags are also recognised and passed to the renderer. Corrected sprite colours (adds a 16-entry palette to each sprite which doesn't have one already. Uses values from Wimp_ReadPalette to create those entries. Is this right? pre and /pre added for preformatted text. Any file which does not begin with an HTML tag (whatever that tag is, except para) will be treated as though it began with a pre tag and ended with a /pre tag Handles relative riscos: URLs (by calling OS_FSControl 37 to canonicalise the filename) Although it is a little suspect with ^ characters, as it gets confused between files and directories Window management handled properly. New pages are rendered to the size of the current window; info/draw window pairs are not joined and can be moved and closed separately (closing BOTH will free the memory associated with the page). Foreign characters allowed - like á and ÿ (&aacute; &yuml;) Forms are not handled The # field is interpreted as an anchor marker - special handling is required for RISC OS URLs which are on filing systems which use special fields. Symbolic link handler added to cache manager so that large local files need not be copied from one place to another. This feature has been disabled temporarily since I changed the protocol, and haven't got around to putting in the new API in the cache manager. Bug in View Source corrected (didn't put the extra zero byte in memory to mark the end of the text, so you got a load of rubbish after the end of the document (if you were unlucky - depending on what was in the memory you got allocated).

Version 0.02

Colours in sprites without palettes is wrong Memory management greatly improved (does not release pages yet though!) However, by cutting out the draw_createObject function and just using flex to extend my internal buffer and put the stuff directly there has helped. Also, memory is only claimed for the page as it is under construction (ie. not 128K claimed and hope it is enough). View Source works (provided it was an HTML or plaintext file!) Again, MemMgmt: far less calls to [cm]alloc, using stack/static heap for as much stuff as possible. All calls to [cm]alloc, free, flex_alloc, flex_extend and flex_free all indirect via the webmem.c module. If an error occurs while looking for a sprite and the sprite name is not null, then it will be looked for in the Wimp sprite pool. If the src field for an image does not contain a '.' or a '#', then the name is looked up in the Wimp sprite pool. Unfortunately, I have had to use Wimp_BaseOfSprites (marked as 'do not use this SWI for future compatiability' in the PRMs). Cache Manager does not go into a infinite loop if it fails to remove a file from the cache directories during the sanity checks. !HTML introduced to act as a fetcher for Risc OS URLs and it doubles as an example program which uses the Fetcher Protocol. This program handles directories as well as files, and now pulls in sprites from the Wimp pool of application sprites as required.

Version 0.01

Cache naming scheme was inconsistent (used hex numbers in some places, and decimal elsewhere). Fixed. Now uses decimals throughout. Cache routine goes into infinite loop if it fails to remove a file during the cache sanity checks. HTML file type not official (awaiting allocation from Acorn) Wimp message ids not official ( " " " " ) Memory allocation problem remains. Arcweb eventually malloc's all memory available. More intelligent use of memory will be implemented. From 13/8/94, remote fetching is available (see !HTML.!help) May specify a local directory URL: eg. riscos:SCSI::SCSIDisc4.$ The RISC OS fetcher generates a pretty display for a directory. Inlined images limited support (src field of img tag must be a RISC OS filename. Only supports filetype FF9 (Sprite) at the moment. (filename with optional #spritename at the end to select which sprite from the file. Omitting # field means use first sprite in file). Can NOT use a URL for the src field yet. Alignment ignored. May hyperise images (a blue box is put around them). URL resolver not implemented, except to check for a : after the first colon in the URL and prepending riscos: if it finds that that is the case.

Future Enhancements