<sb0>
rjo, shouldn't the "start release-1 branch" commit be on master too?
fengling has joined #m-labs
kuldeep has joined #m-labs
<GitHub124>
[artiq] sbourdeauducq merged release-1 into master: https://git.io/vaPWE
<sb0>
whitequark, can you clean up the clock error reporting?
<sb0>
(see commit comments)
<whitequark>
yes, I have seen the comments
<sb0>
re. this new analyzer bug, i guess the dds rate test makes the ringbuffer pointer rollover and there is some issue with that
<sb0>
having a single clock failed exception that is raised both at clock switch time (and the clock switching is ignored) and while a kernel is running (and the session is aborted) would be fine
<sb0>
rjo, why doesn't coredevice/__init__.py import all exceptions from coredevice.exceptions?
<GitHub159>
[artiq] sbourdeauducq pushed 2 new commits to master: https://git.io/vaPlu
<GitHub159>
artiq/master d0af58d Sebastien Bourdeauducq: coredevice: only import common RTIO exceptions
<rjo>
sb0: i need a commit to differentiate the two branches. the buildbot config was not written with multiple branches in mind and needs debugging.
<rjo>
sb0: that __init__ just mirrors what was imported before.
<sb0>
rjo, git supports empty commits
<rjo>
sure it does. i'd also like to see how the merging between master and release-1 works out. eventually the changes in release-1 should be in master as well.
<sb0>
merge/cherry-pick ...
<rjo>
yes.
<whitequark>
rjo: what was the problem with multiple branches?
<whitequark>
ahhhh
<whitequark>
no.
<whitequark>
it *was* written with multiple branches in mind, that's the problem
<whitequark>
it explicitly ignores everything except master
<rjo>
conda broke when i built for another branch. i fixed that. then there is the doc upload. ignoring is fine.
<whitequark>
right.
sandeepkr has joined #m-labs
<rjo>
ah. a thing to note: once we tag something in release-X, that tag will need to be merged immediately into master.
<sb0>
rjo, what?!
<sb0>
isn't a tag supposed to be attached to a given commit on a given branch?
<whitequark>
it is
bb-m-labs has quit [Quit: buildmaster reconfigured: bot disconnecting]
<sb0>
whitequark, you removed the behavior to close the connection when len < 0. not only the analyzer uses that, but the very code you modified (session.c) does, too.
<sb0>
if(!process_kmsg(umsg)) {
<sb0>
*length = -1;
<sb0>
there's even that comment above
<sb0>
* (the session must be dropped and session_end called)
<sb0>
*/
<sb0>
/* *length is set to -1 in case of irrecoverable error
<sb0>
* (the session must be dropped and session_end called)
<sb0>
*/
<sb0>
this not what it does anymore.
<sb0>
whitequark, and doesn't tcp_output() have slightly nasty effects, such as locking the comms CPU if the PC doesn't ack the TCP data?
<GitHub48>
[artiq] sbourdeauducq pushed 1 new commit to master: https://git.io/vaPAv
<GitHub48>
artiq/master ffe4d81 Sebastien Bourdeauducq: runtime/net_server: restore connection close behavior when len<0
<sb0>
whitequark, look at how the analyzer closes the connection ...
<GitHub123>
[artiq] sbourdeauducq merged master into release-1: https://git.io/vaPAT
<sb0>
okay, after RTFSing it seems tcp_output doesn't wait for acks. so good.
<rjo>
sb0: if somebody tags 1.0 in the branch but master only has 0.1 in its ancestry, that 1.0 conda package will be the highest one in the dev label even though master is what we are interested in.