Subject: Admin
Keywords: ::NUPRL6
::TODO
Title: TODO
--------------------------------------------------
(c-q) and simultaneous refresh get editor stuck with x-in thread
and terminal io fubarred. prob should do some history for x-in, x-out??
rebuild queue (maybe other bots) does not catch send_notify errors
and remove bad viewers (as it should)
- recode with new version of notifiable_bot_basic and test
also resending info when nothing changes (ie get idle notify and nothing to dispatch)
web-projection fails to find URL's for other theories
web projection should pre-process and
show objects with bad refs so they can be fixed (rebuilt or fixed by hand)
editor starts transaction
editor calls lib
lib calls editor
editor fooa due to memory-exceeded.
things get stuck.
activate_prf on command line does not update middle T of STM!
but prf_save instead of activate does work.
make sure ref and refdby graph stuff works.
rebuild all prfs modified since 1/1/2018 to get extract_by_obids refs back
xlib
- attempting draw of window after window killed puts editor into debugger
in such a way that terminal-io is fubarred.
- xlib:drawable-error
x-flush calls display-finish-output which enters debugger
maybe some handle error around x-flush can catch, print, and ignore.
- also stop calls to x if we know window is gone or going
SaveComplete in replay failures should
propose list of items to save,
remove saving from replay failures and pop up dialog but remember failures.
then in separate transaction save, any that fail to save should
have dialog offer to restore to replay failures.
Still using MAIN for batch but maybe only if all batch busy?
tree refine not propagating node status up the tree on fail/incomplete.
- would be good to not attempt refine of placeholder
add command in proof tree show path-to-root tactics.
change io-history to record destination and time sent, internal-elapsed good enuf.
fix tx clash bug
- fixed missing write clash case
- still pondering read clash
- clash errors aren't handled well?
- sequentialize prf updates somehow to prevent clashes when updating proofs
not actually sure how they aren't sequential since refine bot should be sequential.
review end-hooks see if can do more inline.
change find connections
- instead have a queue of idle batch refiners.
Refiner places self on queue when active-threads-info shows no threads.
plausable that idle refiners may somehow not be on queue
- have lib ping refiners on some schedule
- check if refiner on queue when adding to queue.
- refiner could specify if appended or prepended to queue
ie have some that work a lot and more that cycle through when the eager are busy.
maybe a second queue for interactive refiners.
view_showc -- now evals in BATCH but not interruptible yet.
refiner crashing while lib is calling db can cause cleanup iob
from refiner link loss to be queued to call waiting for db rsp.
Doc replay/rebuld/refine buttons -- faq to describe bots and introduce terminology.
Get Faq Search to paste automatically from show menus into search parms.
why do replays in green vs black have diff num of thms?
Users should see only their refinements in refine_queue viewer.
Trivial edits shouldn't use global trans -- nor should replay/rebuild
cons_left - there may be a general soln to malformed term in editor
- check for malformedness (like null subterm)
at creation and fail or abort
doesn't seem like a trans-thread should run if only updating tables.
ie if no eval requests between begin and end then no need for trans-thread.
could delay trans-thread until req encountered then resolve.
sleeper-thread or delay-thread maybe.
bots ought to be able to test if pending request viable on demand,
and then remove after dialog confirmation.
if push fails (rightclick) then start dialog to remove.
- verify stuck tries to do this automatically but isn't working quite right yet.
do-tents-collects ????
backthru lemmasx
ParallelFirst -- remove.
gc push doesn't check for interrupt?
-only if it can't find trans-thread
add a limit_exceeded throw catch for refiners to replace abort
- gc limit could be used outside of refiner, so not generally refiner specific.
- set flag if memory-limit is exceeded and then do real full-gc after refinement completed.
batch builds from create data type are lost.
-done but need to restart rebuild queue to test.
need to add edit for refine_queue yet.
transaction-pending-requests not being discharged, there is a discharge
in eval-toploop but must be other paths that leave it undone.
fttb, added some prints.
- think fixed, used eval-thread-lock to sync :
- dispatcher receiving requst with tstate pending
- eval-thread-toploop finishing transaction begin.
changed pending dialog to ping bot if push exited at start.
- changed pending_dialog code for rebuild but not for refine yet.
need to update refine_pending dialog to take view arg and update macros to call with view.
replay takes diff path, but probably should be changed to use common path
- difficult to test since need unusual failure to get a pend stuck.
activate old batch and kill during rebuild.
need RefineInterrupt to not be catchable by tactics.
- changed to use guid_interrupt instead of throw_abort
newAutoStep output with_tty_file
interrupts from bot get fast-tracked now.
Show* button on refine_queue noops.
Add edit capability to rebuild_queue properties,
bots : change appendp to property from arg.
Check that right-click refine and replay work as expected.
clicking on filtertree* when filter tree open causes window lock problems.
interrupt-step on rebuild/replay fails,
Show* button on replay viewer fails.
Add dforms for property edit terms.
Fixed in latest disksave?
(cm-a)psp -- made new edit disksave and psp fails there!
can't reproduce : filter/filter-tree is not on pop-up menu
Fixed: probably fixed by ref view_show fix -- RmGrp on a .recall fails
Done: but need to oed_rhash -- right-arrow in faq pointer doesn-t pop up faq
⋅
--------------------------------------------------
Authors: NUPRL:t
Contributors: RICH:t
:t
⋅
Home