Bugs

• New Refresh Mode that updates variables individually by sending javascript blocks is a potential problem for forms that stay up for long (hours, days) time periods. Need a way to replace entire form at some interval, or after a large number of hformSet's have been transmitting (since Netscape otherwise saves them all up for a potential "view source".
  
• Titler program has bug that prevents it from working when there is a #...# emacs file in a directory (not really related to rpm at all, but just here to remind me to fix it.
  
• Use of the pubname attribute in the {Var} tag is discouraged for anything other than read-only displays. When putting back a variable whose name was different from pubname, RPM will incorrectly try to write it back as "name" instead of "pubname".
  
• Option lists where the options are padded with 0's may not work correctly with the new rpmrefresh mode because the JavaScript code does not yet check for these 0-padded entries the same way that RPM does. This may affect {Icon} and {Menu} tags. The leading zeros are used to control the order that menu options appear on the widget, but a cleaner way of doing this is really what is needed. If this bug causes a problem, forms will still function, but may display numeric values or "?"'s for icons instead of the approriate descriptions.
  
• Need a way to detect internet explorer and disable server-push type documents, since it doesn't seem to handle them properly. Unfortunately it identifies itself as "Mozilla/X.Y (compatible)", so we'll have to look for the "compatible" in the string. Users of internet explorer may not be able to view all demo copies properly.
  
• In order to provide "instant" feedback for status/control forms (refresh mode=3), check boxes and radio buttons must react at the time the mouse is clicked. To avoid incorrect status information being displayed (say, if for some reason whatever the checkbox was controlling failed, or if some else flipped it back right away from another location) a timeout is set after the action is completed. From that time, the status information has a couple of seconds to trickle back the browser. If there is some kind of delay, the interface may change back and forth before finally settling. This could be a little disturbing to users, but should only happen if the network is really slow. The timeout is implemented using javascript code which is loaded into netscape navigator when it first receives the form from RPM.
  
• If a datafile is replaced while RPM is watching it then it may get the content-length wrong (i.e., it stats the file for the length, then opens a different one without knowing.) This should only affect watching text or html files in "dynamic" mode and the damage should be transient. (For example, in the first example in the documentation, if you change stuff over and over enough, there can be some glitches in what's displayed in the par file view window.
  
• Proxy Server Cache may get corrupted by requests for cgi scripts with virtual paths, if the cgi script doesn't generate a redirect when the trailing slash is omitted. Fix: Proxy server needs to rename a file and put it in a subdirectory if it later finds out that the path continues.

Limitations

• Variable Name conflicts:
  1. Case insensitivity: A variable's content is always case sensitive. The variable names always retain whatever case their were originally entered in, but look-ups are case insensitive. This is only a problem when a variable "filter" and "Filter" existed together and were supposed to be distinct variables.
  2. Same-named variable from two sources: Cached up copies of variables are always used if the name matches, regardless of the location specified in a LOC. The LOC only tells it where to go if a variable by that name hasn't been found yet. Once a variable is fetched, rpm does not keep track of its original LOC. This is only an issue when the same named variable appears in two different data files, and you need to distinguish.
     Possible fix: store locations with names of cached up variables (for example, as location:varname)
• rpmdynamic mode on non-rpm files will resend the whole file every time the timestamp changes, regardless of whether the content was really modified or not. For *.rpm files, however, the server compares the output with what it sent last time, and doesn't resend if there were no changes.
• rpmtail mode is broken because sometimes the browser hangs with a blank screen, waiting for something to trigger the buffer to flush or the screen to refresh. Highlighting with the mouse will often allow you to see the "hidden" text in Netscape. Instead of rpmtail, logs are displayed with rpmdynamic for now, but when a new log entry appears at the bottom of the screen, the whole document must be re-sent to the browser, so it is recommended to view the logs with a browser running on the local machine, to avoid slowing things down. (I.e., start netscape displaying from saturn or neptune if you want to look at saturn or neptune's log files.)
• Client disconnecting from a cgi script that doesn't detect broken connections properly may cause hung processes. I don't know that it really does, but I haven't tested it to verify that it doesn't yet....
• ArgMap streams must contain only 7-bit data. (It uses the 8th bit internally to keep track of which characters are escaped and which are not.)
• Some clients, such as Lynx-2.3 and older, don't look at the VALUE attribute of and OPTION element. Option menus will not work correctly with these browsers.

To Be Implemented

• Need to set the status icon on refresh forms to a busy-icon while the server is being contacted..
• Need to add a small built-in document explaining what the status icon on the new refresh forms is form. The icon itself should always be a link to this document.
• ArgMap_stream doesn't handle "prefix" yet.
• Fits file headers are not yet supported.
• Only variables that already existed in a par file are updated in that par file. (Fixed?)
• Make sure rpm tags can't be sent in the _basefile internal variable
• Mechanism for polling/finding out changes for non-file methods
• I don't understand what the issue is with synchronizing libg++ streams with stdout, since they are supposedly the same thing under Gnu C++. Despite this, the code to the nsgmls parser has this in it:

  #ifdef __GNUG__
    // cout is a performance disaster in libg++ unless we do this.
    ios::sync_with_stdio(0);
  #endif
  

Fixed Bugs

• Quoting tests (rerun if quoting code is changed!)
  • Make sure it adds quotes when value goes empty
  • Test nested quotes
  • Check Copen characters for all stream types
  • Test URL escaped characters in html (> , ", etc)