- From: Anselm Baird_Smith <abaird@www43.inria.fr>
- Date: Wed, 27 Nov 1996 15:57:51 +0100 (MET)
- To: Joel.Crisp@bristol.ac.uk
- Cc: www-jigsaw@w3.org, b.kelly@bath.ac.uk
Joel Crisp writes: > > Hi > > Thanks to Anselm for his very fast fix to my FormResource.lookup problem > - it all works ! > > Some points about Jigsaw and the SGI port of 1.02 to IRIX 5.3 > > 1) You need to use systune(1M) to set the kernel parameter svr3pipe to > 0 to turn on bi-directional pipes - this is not > sufficiently documented. Symptoms are InputStream.available() throws > IOException when used on file handle returned > from Process.getInputStream(). Prevents non-blocking io on process > streams ! > > 2) There are intermittent crashes with core dump signals - as of yet I > have been unable to trace this, but it was happening *before* I > loaded any of my own libraries ... ;-) I think this is in the SGI > port, possibly in their thread code ( IRIX threads are *weird*.. ) Thanks, I'll try to build an IRIX page out of this in my spare time :-) > Some other points > > 1) I have a crocked version of FileResource which runs SGMLS on the > submitted PUT data to validate it against the HTML 3.2DTD. It is a > crock because it is not clear how filters apply to PUT methods. I've > overloaded PUT - but a filter seems a neater way of doing this. > One nice point is that Netscape displays the failure message so you > can feed back the first couple of K of errors ( Question : Is there > a length limit on server failure replies ? Is it actual or > recommended ? ) No length limit that I am aware of. As you say this should be done through a filter, but you have to "tee" the request input stream (the put data) in order to have one pipe going to your checker and the other one saved (and feeded back again, so that the FileResource gets it). > 4) Is there a possibility of getting the StoreManager to run the > finalizers when it does it's GC ? Well, now it's time to tell the truth: although I have tested the ResourceStoreManager GC several times, it is unlikely to happen at this point in time, since it is trigerred only when 512 stores are loaded (eg you have to serve more then 512 directories to enable it). I will propabaly lower this limit in next release. > 5) How hunger is the attribute manager ? How does it do it's GC ? > Can you intercept resource dump/restore to throw away resources and > re-create them if they have not been modified from the defaults ? > Why ? I have a database of potentially a few hundred thousand > entries. When I use my database interface to get the search results, > I'm using form cards to display groups of 40 records per card, with > deferred creation of the cards. However, if I do page to a card I > suddenly require 40 new resource. Over time, this will accumulate > considerably. Now, each time I get a record, the meta data in the > resource is generated from the record itself - seems stupid to have > to store this as well, and knowing JAVA I just have to override the > right methods... ;-) You can intercept resource dump/restore (check the w3c.jigsaw.resources.AttributeHolder methods, in particular pickle and unpickle methods, which you can override in your resources). Check also w3c.jigsaw.resources.Resource:notifyUnload. There is no mean to explicitly unload a resource right now (is that what you're asking for, would this help ?) > 6) Is there a 'text/html only' field for form cards for, e.g. section > headings or comments ? With left/center/right justification ? > ( Aside - Style sheets ? ;-) ) Style sheets have been added by Yves Lafon (yes, Jigsaw team as doubled recently :-) recently to the HtmlGenerator. For the text/html field, I used somethink like that for links (there is a LinkField I think), it would be easy to add/define an HtmlField though. > 7) Form cards are nice, but I have a set of code for 'database views' > which implements a whole language for HTML page description for > records - in JAVA ;-) Any takers ? I'm working on getting it > interfaced to Jigsaw. Wow ! This would be nifty. You may also want to check server side includes (which can be extended in Java), but that's another view of what are I guess related - at least - problems. > 8) Correctness of HTML generated by Jigsaw - I've noticed that > sometimes Jigsaw generates non-HTML v3.2 html - any chance of a slow > move to merging this in the main stream ? Correct, I know I have to fix directory listing at least. > 9) CVS - got it all working. Re-reads the directory a lot tho - maybe a > bit more intelligence in caching the directory ? Or a 'Force Reload' > button ? ( Just stick a TTL in the HTML header and let the browser > handle it ? ) Maybe you should document the cvs.root variable a bit > better ( Sardonic ;-) ) Stay tuned on that one. Next release of Jigsaw has a much better support (thanks to Yves Lafon again), and it really enables collaborative editing of a web site, which I have been using for monthes to produce all the Jigsaw pages. This ought to be better documented, it's really nifty (beleive me) > Hope this is of interest/thought provoking Indeed, thanks for all that feedback, Anselm.
Received on Wednesday, 27 November 1996 10:02:43 UTC