eris/BUGS

Current
- Within the sandbox sudo is not working for tools.
- If vigil is started with eveything up-to-date the up-to-date message never
  appears.


Current (tool related)
- disassemble for python doesn't always work.
- gut had an error with utf-8. A traceback printed directly on the screen,
  garbling the interface.


Fixed
- The interface is hanging during large tool computations.
- Your cwd relative to the code base path should not be important.  Currently it
  causes the cached paths not to be found.
- Clicking the mouse button causes a traceback.
  (Started when the keyhandler was added)
- The cursor disappears when on the placeholders "."
- The filenames are only showing their basenames.
- After moving with the mouse, the new position is not remembered. So that
  moving with keys moves from the position you last had when moving with keys. 
- Not working correctly: Let rows in the table have different numbers of
  columns. If navigating between rows, from one row, to another with less
  columns, change the column of the cursor. 
- Sometimes when moving with the mouse the wrong result appears. The column
  doesn't change.
- on_file_saved is responding to everything including when the cache is saved
- It seems that on_file_saved won't work properly with subdirectories. The event
  path is always a basename.
- Tabs are appearing as question marks in the source code.
- When changing a file and the results are updated... The updates don't show on
  the screen, you need to move the cursor over them to force them to be shown.
  And in this time the CPU is 100%.
- An exception shows when the cache changes while it is saving. 
  No more autosave.
- Fix problems caused when displaying long path names.
  <- So far only fixed by truncating the file name.
- Bug in _closes_result when using min().
- urwid is not coping with large files well.
- The result does not change to "?", only the statuses. (urwid implementation)
- The screen does not change from the placeholders if you never move the
  cursor. (urwid implementation)
- Have seen a hang. It was running the 'file' program on 'table3.py' at the 
  time. Still responded to 'q'. On next run all the results were in the cache.
  Maybe the hang was only in the display. (urwid implementation)
- Quiting out sometimes hangs. (urwid implementation)
- Often the program crashes early on. (urwid implementation)
- One time the program was slow after running for a long time, and after
  restarting it was definately quicker. (urwid implementation)
- Views crash when scrolled too far.
- Ensure the terminal is in a good state if the process is interrupted.
- The listing's title is missing when the program starts.
- Its possible to be calculating the same result more than once at the same time.
- The scrollbar disappears on long summaries, when moving to the next page down.
  <- urwid related
- Sometimes the cache isn't being saved
  <- changed cache implementation
- signal_process_tree is signalling the root process when it shouldn't.
  <- not using signal_process_tree at the moment.
- "./table.py t *.py" shouldn't show the py files in the parent directory of t.
  <- Not doing globbing at the moment.
- Running out of space while saving the cache leaves a partially written .tmp file.
  <- File is deleted
- If a file becomes executable this is not detected and it will have the wrong
  color.
- When quitting out any currently running jobs are forgotten, and forever left in
  the running state after restart.
- Quiting out is hanging.
- Quiting out is not producing an error code of 0.   i.e.  $? != 0   If all the
  results are calculated it does give 0 when quiting out.
- If a tool is interrupted its result will be incorrect, and worse cached.
- The interface is very slow and almost unresponsive on larger projects.
  <- The whole summary was mistakenly being rendered.
  <- Changes to the cache weren't excluded from events caught by pyinotify.. 
- Sometimes when quiting out the screen is not restored
- Sometimes when starting up the whole screen is not drawn, only the first diff.
  <- The screen update thread was being started before the blessings fullscreen
     context manager.
- Currently any issue thats on the current row, to the left of the cursor, won't
  be found, even if its the last one.
- The metadata result isn't formatted correctly.
- The log is truncated, and is too short on large displays.
- The "All results are up to date" message doesn't show immediately.
  <- The most recent n messages weren't being shown.
  Also it ideally wouldn't be repeated by each worker thread.
  <- still being repeated
- Sometimes two results are yellow simultaneously, even though theres only
  one worker thread.
  - Caused when the pep8 column for some reason is the same as its neighbour
    pflask!
  <- The key in sync_with_filesystem was sometimes the same for different tools.
     Fixed by including the tool name in the key.
- All rows in the filename column need to change width together, or always stay
  the same size.
  <- For now its always a fixed arbitrary size.
- Shouldn't reuse "." to mean not applicable. Its hard to tell which jobs are
  really pending.
  <- Changed to a grey block
- The scrollbar is invisible on very large pages.
  <- Now its always at least one character in size.
- While the metadata is being calculated the width of the metadata column changes.
- If a file has many tools the statuses are truncated on the right. Probably
  should scroll over with the cursor. Or possibly wrap the statuses onto the
  next line?
  <- Fixed by scrolling over. Haven't tried wrapping yet.
- The cursor can dissappear behind the scrollbar when scrolling to the right in
  the summary view.
  <- Off by one... Forgot the extra space added at the end of the filename.
- lib/jquery.min.js (from luakit) takes a long time to display (by pygments) and
  the output is corrupted.
  <- Not a bug with vigil, and I can't see the corruption.
- Files which are broken symbolic links only show errors.
  <- Decided to not list broken symlinks in the summary view.
- The filenames are all white!
  <- ljust on a termstr returns a str. Fixed by avoiding ljust, but should
     add ljust to termstr.
- The cursor sometimes moves down one line while programs are being run.
  Probably a problem in sync_with_filesystem. Happened while leaving the
  cursor on pyc files.
- Its not working in the linux console. The terminal codes are producing
  garbled output there. Its probably the 24 bit color terminal codes.
  <- It wasn't about colour.
  <- The encoding was latin, it should have been utf-8.
- Jobs are still being started during shutdown
- Sometimes jobs hang. Started after the switch to multiprocessing with pausing.
  Accidently pausing?
  <- Fixed by removing the join which can deadlock and is unnecessary.
- Sometimes quiting hangs. At least with the q key. Test ctrl-c also. Started
  after the join was removed when running jobs.
  <- Fixed by making the multiprocess Process a deemon thread. But is the reset
     method really terminating the process?
- Switching status style doesn't work sometimes after restart.
- next_test.py passes in the console but fails when piped through 'less'.
  <- Fixed when changed to terminal_codes.
- Old results aren't being deleted from the cache.
- When the terminal window is resized the contents aren't immediately resized.
- The length of the grey highlight bar isn't always the full width of the
  view window.  Its only as long as the widest row currently showing.
- sync_with_filesystem is being run when changes occur in ".vigil" directories
  in sub-projects.
- Filenames are always colored white in gnome-terminal. This used to work, and
  was probablty broken when blessings was replaced.
  <- False alarm. Went away!
- When sorting by ext then path, the filesystem is always changing for some
  reason.
  <- The filesystem change was the log changing. Moved the log elsewhere.
- When a sync_from_filesystem occurs the files are always sorted one way.
  Which could be a change.
- While the watch manager is disconnected from the mainloop, events are being
  queued up, and are all played out when the watch manager is reconnected.
- First character in help is green
- Switching 'watching' on, is slow.  At least update the indicator text quickly.
- Something in ~/repos/pygame breaks convert_lscolor_code_to_charstyle
  <- Files with ".conf" extensions were breaking convert_lscolor_code_to_charstyle
- When syncing, why does the tool name in the title of the result pane briefly
  change colour?
  <- Can't see this any more because you can't manually sync. Also doesn't show
     when toggling the 'watching' switch.
- terminal.italic is None in the linux console. Then code_for_term fails.
- Vigil won't clear the cache if the previous vigil was still running and making
  changes after the new vigil was installed.
  <- Created a ".creation-time" timestamp for when the cache was created.
- Have all jobs calculated, and job runners waiting, then find new jobs with
  syncing-with-fs... The job runners aren't then being started to calculate these
  jobs.
- Sometimes the job runners stop and say all results are done, but I can see
  outstanding pending results. A restarted vigil processes them.
- Getting screen corruption after using curses for input. The corruption occurs
  when viewing particular reports. Must be in a different mode so characters
  are interpreted differently.
  <- This went away, for now...
- The mouse sometimes stops working. At this time you can also highlight parts
  of the screen. Maybe we've left raw mode?
  <- This also went away.
- Theres corruption the first time you type a key or press the mouse button.
  The screen gets cleared and only the diff lines are shown. Eventually with
  more diffs the whole screen is showing.
  - If can't fix properly... a hack could be to programatically type a key early
    on.
  <- This was some problem with my use of curses. Fixed by using urwid's raw_display
     instead.
- When the screen changes the cursor can be seen flying around quickly. It also
  ends up on the right hand side of the screen.
  <- I forgot I began relying on curses to hide the cursor. Put back the
     'hidden cursor' contextmanager.
- Sometimes when quitting: "close failed in object destructor",
  "sys.excepthook is missing"
  <- This has probably gone away now jobs are run differently
- disassemble.py is not always found
  <- The script isn't used anymore
- Some file's display is garbled. e.g. perldoc of FieldHash.pm
  <- pydoc was outputting ASCII term codes.
- Scrolling right on a result from disassemble_pyc (or pydoc run on termstr.py)
  causes screen corruption.
  <- pydoc was outputting ASCII term codes.
- If a pending file is deleted, while not watching the filesystem,
  then tools fail when they can't find the file. Also, even if watching, there
  would be a race. Do what?
  <- Decided to remove the 'watching' toggle. Now its always watching. Use
     pause to stop computation.
- Changing the status style with 'watching' off can result in recalculations.
  <- 'Watching' is never off now
- When the filesystem changes a lot vigil is syncing the summary repeatedly
  for each part in a queue of changes. Needs to empty the queue then sync.
- There is an exception if vigil's window is made too small.
  <- When the window is narrower than 10 or shorter than 20 the contents
     start to crop instead of shrink.
- The cache is not relocatable anymore. i.e. Restarting vigil after renaming
  the project directory is failing.
- Some jobs are always recalculated when restarting vigil.
  e.g. vigil.py pylint, BUGS metadata, BUGS _pygments
- Tracebacks occur if you pause during early startup.
- Scrolling in the help screen doesn't work with the arrow keys.


Won't fix
- If the summary window is narrower then max_path_length the paths are never
  visible.
  <- Only a problem for very narrow windows.
- There is no color in lxterminal, only shades of grey.
  <- Only testing gnome-terminal & stterm at the moment.
- Sometimes a lot (or all?) of the results are "?" even with correct statuses.
  <- Happens when the cache is deleted from underneath a running vigil.
- In stterm the previous console title isn't being restored.
  <- Seems to be a bug in stterm, because its okay in gnome-terminal.