|
|
Open issues (change requests and bug reports) are now tracked by the
GNATS database
Candidate Manager issues reported with version 1.1
o Candidates with multiple failures are not handled correctly.
=> added in version 1.2
o Have the user's time selection persist across candidate
selections, rather than defaulting to "current time" with every
candidate selection. Even with a whole new set of candidates this
is desirable.
=> added in version 1.2
o Convey information about why the search terminated, to
provide the user with an understanding for why these are the only
candidates in the list (as per Shirley's e-mail).
=> for CBFS search, added in version 1.2
=> for COVER search, planned for version 1.3
Stanley issues reported with version 1.1
o Step 3 of CB Diagnostic Test 1 takes a long time without truncation,
raising issues of what to do when the engine has to think a while.
The user is told to select a candidate from the manager before there
are any available. Think about what the user needs to be told while
waiting for the engine to provide the candidate manager with a set
of candidates.
=> added user prompts for version 1.2
o In the above situation, the schematic could be updated with the
assignments and commands prior to the issuance of fc so that what
the user is looking at while waiting reflects what the engine is
thinking about. Consider removing the labels on predicted values
from the previous step while waiting for the user to select a
candidate.
=> added in version 1.2
o The user might want to interrupt the scenario if it is taking longer
than expected, say to modify the truncation parameter. The DISMISS
button doesn't appear to stop the scenario.
=> The DISMISS has been removed. The larger issue is under study.
o I don't know if it's the candidate manager, stanley, or the engine,
but it seems to really hog my machine. I had to shut it down to
use other applications.
=> unresolved
o (Feature): might have Stanley come up in the upper right corner
if the candidate manager comes up in the upper left corner.
=> in version 1.2, Stanley's default preferences have been changed.
o (Feature): the Step and Run buttons seem to more appropriately
reside in the scenario window. For scenarios being generated
on the fly, the scenario window would help show what's there as
the user is generating it.
=> added for version 1.2.1
o (Feature): probably do not want to kill the candidate manager
when the user quits Stanley, even if stanley started it up.
Or perhaps that can be selected in the ensuing Quit dialogue
when appropriate:
{Stanley} {Skunkworks} {Cancel}
=> added for version 1.2
o (Feature): the single preference for the number of candidates to
search is only for cbfs. There should be an separate preference
for cover search to specify the maximum-rank.
=> added for version 1.2
o (Feature): more generally, obtain a complete list of engine
parameters that the user/modeler may want to specify, and have
them represented. Some that are engine compile-time parameters
may need to be pulled out for specification by a modeler; I'm
thinking of things like ok-prior-probability and whatnot ...
Shirley can probably quickly provide such a list.
=> planned for a future release
Major Issues reported with version 1.0
o the java layer is designed to forward a random candidate to Stanley
for display; skunkworks needs some way of conveying the full set of
candidates to the modeler, ideally in a clear way.
=> added in version 1.1
o there is no good way to do distributed model development using
Stanley.
=> simple locks may provide basic protection when sharing the same
working file set -- put off until later
Minor Issues reported with version 1.0
o the first release was announced but failed with some environment
variable not being properly set. The upshot was we decided that
the release czar and someone else will check that the integrated
system works before announcing releases (ah, memories of remote
agent integration ...).
=> release procedures are now in place.
o there are several steps required to start the system. A one-line
shell command is forthcoming.
=> added in version 1.0.2
o the models that are distributed with Stanley should be akin to a
shared library, rather than requiring that source files be copyied
into the user area. Projects may want access to sample models but
users shouldn't have to each house copies of them in with their
project files.
=> put off until later
o doing Load&Go on a sample model (cbAndLeds) results in a dialogue
to set the initial modes of each component. I was unable to make
changes after scrolling down in the dialogue, but the flakiness was
not reliably reproduceable.
=> not reproducable
o while Load&Go was processing, I switched rooms in my window manager,
and Stanley crashed because tcl needs to have the window be the top
window to draw correctly. Will says it's an inherent tcl limitation.
=> problem with CDE, ok with fvwm
o the only way to reset the test module to the initial state is to do
another Load&Go, respecify all of the initial conditions, and wait.
A Reset button would be quicker and more convenient.
=> added in v 1.0.1
o when a scenario is selected for stepping, the canvas should probably
be inactive to prevent the user from setting additional values that
conflict with the current scenario.
=> added in v 1.0.1
o after stepping through a 3-step scenario, the step button's bubble
help stated that I was in step 1 of a 3-step scenario.
=> added in v 1.0.1
Back to Main Page
|