Jankey/TODO at master · gmatht/Jankey · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
Short Term

Make WANT_CRASH_ID optional
Force all Titles to start with lyx::
Start moving out lyx specific stuff
Make variables customisable
Support LessFS
do not mark as "Reproducable" unless KEYCODEpure is short enough to report to user
Debug problem with "./cache-bisect.sh examples 1276005507.KEYCODEpure" when ./.lyx directory exists

in cache_bisect.py,
  - do make install, make clean rather than leave in src dir
  - Tar up old source dirs
  - compile in a tmpfs
  - deduplicate installed files

Longer Term

Support mouse clicks.
Support other failure modes (e.g. freezes and Python tracebacks)
Tally of failures vs successes
If come across different bug when trying to reproduce another, reproduce *both* bugs
Add report as to how easy bug is to reproduce
Try e.g. electric fence
Automatically update from svn every 24 hours, recheck if bugs have been fixed.
Automatically add bisect to bug report.
Add a GUI
Try eliminating each keycode in turn.
Do performance regression testing
fuzz test inputs and outputs as well.
use ulimit -u to protect against fork bombs.

Pause when memory pressure high.

Integrate SIKULI: http://groups.csail.mit.edu/uid/sikuli/

DONE:

After eliminating keycodes, remove restriction that backtrace has to match original and see if we can eliminate more keycodes.
Try removing alt, shift (but not ctrl) modifiers from key presses.
semi protect home dir for repeatibility.
HTML out with links.
Move all temporary files to a single dir which can be mounted as a tmpfs to reduce HDD writes
Fix gdb PID bug, use readlink -f lyx/src/lyx
Try to get KEYCODES to be repeatible and useable (DONE)
Do not spend hours reproducing a bug that has already been reproduced. (DONE)
Make it easy to use Xvfb instead of VM. (Done)

Store state so can continue replay after restart

Adaptive keypress removal

We could speed up the rate at which keycodes are discarded by using an adaptive algorithm. E.g.
if the "splice" techinque was last succesful three attempts ago and the "independant" tecnique
was successful 9 attempts ago, we could have a 3:1 chance of picking the splice technique.

A similar method could be used to choose between discarding a small number of keycodes vs a
large number of keycodes.