fsv/TODO

58 lines
2.8 KiB
Plaintext

1999 Jan 27
* Put content in README and INSTALL.
* Write NVStore/uXML backend (or find a suitable existing one) to store
configuration parameters in ~/.fsvrc.
(The rc file should use XML formatting, but I don't want fsv to link
against libxml/expat to do this-- those libs are overkill. uXML is a tiny
yacc/lex-based XML I/O library I started writing for just this purpose,
but that will take some time to mature. Anyone know of a good, clean, <50K
DOM-based XML parser/generator library for C?)
* Rewrite the MapV layout algorithm. Currently, it works, but doesn't
generate very optimal layouts-- some nodes are too wide, some too tall,
and the ones in the frontmost row/rows invariably turn out paper-thin.
I've discussed an alternate approach with a friend of mine (hi Louis!),
one heuristically based that would run slightly slower but give much
better (even if not provably optimal) solutions. I'm definitely open to
ideas here. It's a rather tricky CS problem.
* Implement a somewhat smarter node pointing/selection handler for faraway
nodes (e.g. when in bird's-eye view mode). Right now, it gives you exactly
the node that is under the cursor's hotspot-- even if it's so small you
can barely see it. fsv should instead return an ancestor node, that is
larger than a set pixel-based minimum.
* Replace the bulk of GTK+ cruft (see dialog.c/gui.c/window.c) with
Glade-generated code, to improve view abstraction. All of the main
window's callbacks have already been moved into callbacks.[hc]. Still need
to figure out 1) if/how Glade can handle compiled-in XPM's; 2) how to set
up the GtkGLArea widget, which isn't directly supported; 3) how to specify
the minimum window size as a proportion of screen width (as opposed to
giving absolute pixel dimensions-- ick!)
* Implement a better way of bringing up the online help. I'm almost
embarrassed by the cheap /bin/sh hack currently employed :] (Not to
mention it leaves zombie processes hanging around until the main program
terminates-- bad bad bad!)
* Perhaps add scan caching. This would store the complete state of a
directory tree (say, /usr/lib as an example) in ~/.fsvcache/@usr@lib, and
the next time fsv is invoked there, it loads the cache and skims through
all the subdirectories, only scanning those with an updated timestamp.
(Result: much faster loading of '/' and the like)
ColorByTimestamp mode could take the time of the previous scan as the
start point of the temporal spectrum, allowing something of a "graphical
diff" on the state of the filesystem.
* Add a way to have fsv not scan certain directories, such as top-level
AFS mount points (stat'ing those can take *centuries*)
* Oh, and did I mention I'm taking 6.170 this semester...? :-) Some of the
core modules (camera.c/geometry.c) could probably stand some improvement,
and by the time summer comes around, I should have a good idea how!