New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Lines take too long to print #625
Comments
Isn't this the more generic "need a high-level memoization cache" problem? I remember people saying a few other things are hit by not honoring siglus on the top road, and was one of the things that Curtis worried about for proper implementation of an It might be worth mentioning that large stacktraces take an extremely long time to print for me, and only appear to start buffering after everything is finished. That is, while :talk fetches the chat backlog it doesn't print while it's getting the rest, and only starts the long printout after fully synced. I'm not sure if that's because :talk returns the lines all at once, though? |
I'm uncertain if this is a 'bug' or a feature request. @ohAitch can you clarify? |
It is a bug, introduced by a4abd16. If you revert it the problem goes away, but there's a permanent store leak that it did in fact fix. |
Closing as I'm not sure this is still relevant, and I'm biasing towards a scorched-earth approach on older issues. |
Does the linked commit still cleanly revert? If it does, is there a noticeable difference in the responsiveness of running (I vaguely remember something about memoization being restructured, but "not sure this is still relevant" seems like the opposite of having a reason to close a ticket. Bug bankrupcies are not beneficial, though I suppose if this is effectively a MAYFIX classification I shouldn't take offense.) |
Definitely don't take offence -- as you may have noticed, I've just been trawling through the issues list trying to clean up things that I don't expect can be worked on as-is. I expect collateral damage, but I appraise a particularly-actionable issues list as being worth the cost. A longer justification for closing this one is that "Lines take too long to print" is probably no longer an accurate characterisation of a current issue; the specific problems you describe may certainly still be relevant, but should probably be repackaged/re-presented, if so. |
The purpose of the issue open/closed flag is not to track actionability,
capacity for action varies over time - "we expect to work on this in the
next 6 months" is better served by a "high priority" tag, which people can
filter on should they be searching for high priority things to work on in
particular.
"Prints slowly" is of course subjective, but I believe I did see the issue
on the app/djay livestream, which despite having the benefit of the new
bytecode engine had noticeable lag compared to an old development branch I
had with the relevant commit reverted. At its core, this is an issue with a
specific patch breaking tank-renderer big-O by disabling a specific
performance annotation via idiosyncratic outer-road behavior; which to my
knowledge remains accurately characterized by "all the printing is slow".
I welcome any and all re-presentation via renaming the issue or such.
…On Wed, Jun 19, 2019 at 00:19, Jared Tobin ***@***.***> wrote:
"not sure this is still relevant" seems like the opposite of having a
reason to close a ticket. Bug bankrupcies are not beneficial, though I
suppose if this is effectively a MAYFIX classification I shouldn't take
offense.
Definitely don't take offence -- as you may have noticed, I've just been
trawling through the issues list trying to clean up things that I don't
expect can be worked on as-is. I expect collateral damage, but I appraise a
particularly-actionable issues list as being worth the cost.
A longer justification for closing this one is that "Lines take too long
to print" is probably no longer an accurate characterisation of a current
issue; the specific problems you describe may certainly still be relevant,
but should probably be repackaged/re-presented, if so.
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#625?email_source=notifications&email_token=AAOFPBVJTDOZUFDRO6TQNE3P3HMYXA5CNFSM4BVWU7GKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODYA6ADI#issuecomment-503439373>,
or mute the thread
<https://github.com/notifications/unsubscribe-auth/AAOFPBWNH3H2V2H5OLT2S7LP3HMYXANCNFSM4BVWU7GA>
.
|
Slow printing is definitely among our lowest-hanging performance fruit, just waiting to be plucked. It's probably gotten worse since this was opened -- there are multiple unicode conversion passes over I haven't looked into As far as this issue is concerned ... On the one hand, the entire printing pathway should be rethought (or perhaps burnt in a cleansing fire), on the other, this is a specific characterization of a specific problem that is likely still active (poor performance caused by disabling a useful but broken feature). Per the test of "actionable as is", I favor keeping this one open. |
say no more |
Ping. Performance 👀 |
Well, @philipcmonk do you know if "the outer road doesn't have access to memoization" is still an accurate statement in the current system? I do vaguely remember seeing a PR about this. |
Memoization is still disabled on the home road, and the memoization cache is still not promoted across roads. Enabling either is simple, having confidence in the correctness and/or righteousness of either is somewhat less so. Note that the I tried virtualizing Finally, I enabled |
Can you confirm whether or not doing so causes a subjective change in the experience of typing |
There isn't anything actionable on this issue anymore. If we're concerned with printing performance we need to come up with some way to measure it and detect regressions. As is, there's not really anything that we can do to complete this issue. I'm going to close it. If there's some specific performance issue or action here then we can handle whatever that is separately. |
@ohAitch's last question is perfectly actionable, and even includes acceptance criteria. Experimenting with home-road memoization would be good. There are lots of things that could be done to improve print performance, and it would be great to have measurements. But this issue is more specific than that, includes important historical context, and clear next steps. |
Would the task be to:
Also sounds like there may be another spinoff issue wrt rendering The original context was from a change that can no longer be cleanly reverted which makes comparisons hard, but the issue to come out of this could be something specific about "Enable home road memoization to improve print performance" linking to this for context with specifics about correctness risks, etc. Maybe pulling the general value of "fast print performance" into the wiki and linking it back to this issue too. I could condense that down into one that links here. One of the goals of this is to pull out the important stuff so people don't need to read through old comments where only parts are still relevant, basically making it easier to read and understand quickly for people looking at an issue. I'm not interested in just closing old issues (which is why I'm going through all of them), but a lot of time these old ones can be restructured in a way that's clearer. For example like this: #6033 |
This bug was introduced by a4abd16, is the root cause of #507, and of me running into #592.
Somewhere in the printing system, as
~+
hint is being thrown, and the hint is not being honored. Doing so causes a memory leak; not doing so causes unreasonably slow performance forut
(sample contains the type of ++twig several times over).The text was updated successfully, but these errors were encountered: