Grail To Do List

(See also the Grail 0.2 bugs list, from this was derived by editing.)

Here is a list of things that ought to be fixed in Grail and aren't being taken care of yet. Obviously, these are not in any particular order.

General Project Items

The font and color assignments are hardcoded (and incomplete). This should be done via a preference panel; only a partial solution has been implemented so far. (Somewhat improved in 0.6.)
The indication of a partially loaded or broken image could be better. When an image load fails, the "ALT=..." text should be displayed, if present; not a blank space. The stop sign is confusing.
Pages containing an excessive number of in-line images run the risk of running out of file descriptors, which isn't handled gracefully.
Grail is unresponsive during the initial connect() to a server. This is especially noticeable for dead servers, where you have to wait the full time-out period. (Work-around: type a Control-C in the shell window where you started Grail.) (Should use connect_ex() instead for real sockets; may require changes in *API handling.)

Need to speed up the Tk rendering code.
The <TABLE> support isn't good enough for many of the commercial Web sites out there today; this should be fixed.
There's a problem where a <DL> inside a <DT> doesn't nest unless it's inside a <DD>. See the lay-out test page.)

Running two Grail processes sharing the same cache directory is not a good idea. The last one to quit wins.
When a username/password combo is known for a particular page, it should be submitted by default for all other pages in the same directory (maybe even subtree?).
The $no_cache environment variable should support domains as well as hosts. (Probably a preferences panel issue.)

Applets can't use modules that are imported as shared libraries. (I tried to fix this but the package support using ni seems to have broken it again.)

Applets can read arbitrary files (they can only write files in $TMPDIR or /usr/tmp though).
Applets can open arbitrary sockets.
Applets have access to most browser internals and can modify them.
Applets can quit the browser (in a number of ways!).
There is no limit to the resources used by an Applet.

Restricted execution mode doesn't allow access to self.__dict__, making implementation of __getattr__ and (especially ) __setattr__ hard.
Fred Drake's <LINK> support. This will probably be an HTML plug-in.
Memory leaks. (Yeah, right.)
Frozen binaries and packages won't work together. (Is this still a problem with Python 1.5?)
No built-in support for these protocols: gopher, wais, telnet. (Later.)
No content-encoding support (well, it recognizes it and offers to save; this should be enough). (Support for gzip in release 0.5.)
Applet code is executed synchronosly, so any blocking I/O they might do (such as loading URLs) blocks the entire browser.
There could be an applet debugger.
The save dialog for files of unknown type should display the URL and file type, and suggest a filename. (Partly done; a filename is suggested, and the MIME type is shown.)
<FIG> support.
Support for the Secure HTTP protocol (https: URLs)
Support for <MATH>.
"Cookies". (Originally expected in release 0.5, still not in 0.6.)
There should be an automated process for downloading and installing patches and plugins.
No in-line scaled images. (Works if PIL is installed.)
Print the actual contents of form fields.

Occasional "grab failed" messages resulting in failures (Tk bug?).
A socket error in the http protocol handler doesn't seem to be handled properly.

Directory listings of local files don't work (it invokes "ls -l").
Creating frames makes it go beserk.
Stability. (Wait for Tk.)
Performance. (Wait for Tk.)
SIOUX steals events from Tk, especially activate events.
Some pathname operations are still using Unix style paths. (Which ones?)
No way to generate button-2 or button-3. (Wait for Tk; need work-around.)
No way to generate Alt-key events. (Wait Tk; need work-around.)
Can't import standard modules from applets. (I know what it is -- Mac pathnames look like URLs so are removed from sys.path.)

These items no longer require our attention.

In the announcement, add a pointer to netpbm (perhaps even mirror on python.org). (Done.)

Most protocol handlers read their metadata synchronously, thereby blocking Grail while waiting for the metadata to be transferred completely. (Fixed for HTTP.)
Issueing a Reload command in the window created by the View Source command will redisplay the file formatted as HTML. (Done.)
No scroll bars in the load/save file dialogs. (Done.)
The load/save file dialogs should remember the last directory used. (I have this working for the save dialog, but in a too ad-hoc fashion to be useful. It should be a feature of the load/save dialog itself.) (Done.)
When dragging (as opposed to clicking) in a link, the link should not be followed. (Done.)
Printing of non-HTML documents doesn't work. (Done.)
The image objects for the animated logo are not shared between browser windows. (Done.)

A <P> tag following a <BR> tag does not insert a blank line. See my lay-out test page. (Done.)
A <LI> tag outside a <UL> or <OL> list doesn't get rendered properly. See also the lay-out test page. (Done.)
Check Grail against some torture test pages, e.g. Digital's Forms test suite and BrowserCaps. Anthony Baxter once reported: "Grail passes some of these tests, but not all. So far, the major failures of things that BrowserCaps claims is required are: ALT text in <img> H4-H6 all use the same font. No support for   (non-breakable space)." (Partially done.)
``Multiple BR's create more space (they are not meant to).'' (Done.)

There is no persistent disk cache. (Done.)
The memory cache is never flushed. (Done. There is no longer a memory cache, but the disk cache is properly flushed.)
Cache entries are never tested for freshness. (Done.)
Form input and queries shouldn't be cached. (Done.)
Reload doesn't re-read images. (Done.)

(Surprisingly enough, some of these have been done. :-)

Crash on JPEG images: os.popen not defined. (Done -- no crash, still no JPEG suport though.)
Can't write user preferences file? (Done -- was a GUSI chdir() bug, fixed by Jack.)