Stream: t-compiler/wg-prioritization

Topic: t-libs issues in the meeting agenda?


Santiago Pastorino (Apr 30 2020 at 13:30, on Zulip):

@pnkfelix one important question, as part of this discussion compiler-team#267, I wonder if we should add some T-libs stuff into the agenda

Santiago Pastorino (Apr 30 2020 at 13:30, on Zulip):

I wonder if we want to include something as part of this move

pnkfelix (Apr 30 2020 at 13:30, on Zulip):

oh right

Santiago Pastorino (Apr 30 2020 at 13:30, on Zulip):

at least to mention them maybe for now?

pnkfelix (Apr 30 2020 at 13:30, on Zulip):

given that the meeting is starting in 30min

pnkfelix (Apr 30 2020 at 13:31, on Zulip):

lets not add T-libs stuff today

Santiago Pastorino (Apr 30 2020 at 13:31, on Zulip):

I have already added stuff

pnkfelix (Apr 30 2020 at 13:31, on Zulip):

but lets add an announcement about it?

pnkfelix (Apr 30 2020 at 13:31, on Zulip):

oh okay

Santiago Pastorino (Apr 30 2020 at 13:31, on Zulip):

check the agenda and let me know if looks ok

pnkfelix (Apr 30 2020 at 13:31, on Zulip):

well then lets definitely announce it

pnkfelix (Apr 30 2020 at 13:31, on Zulip):

yeah okay clearly I need to look at the agenda again; I overlooked this detail yesterday

Santiago Pastorino (Apr 30 2020 at 13:31, on Zulip):

I can't promise right now that every single corner of t-libs is covered because I'm not sure what do we want to do

Santiago Pastorino (Apr 30 2020 at 13:32, on Zulip):

in particular I didn't include nominations, because if I got that wrong, it's like the design and things like that of t-libs are not what t-compiler team should do

Santiago Pastorino (Apr 30 2020 at 13:32, on Zulip):

and I guess in that sense nominations are for the t-libs team

Santiago Pastorino (Apr 30 2020 at 13:32, on Zulip):

but issues and regressions are for us?

Santiago Pastorino (Apr 30 2020 at 13:33, on Zulip):

this was my guess and more or less what we have done during pre-triage

pnkfelix (Apr 30 2020 at 13:33, on Zulip):

so from my understanding of agenda as currently written

pnkfelix (Apr 30 2020 at 13:33, on Zulip):

is that you have posted some beta-backport and stable-backport nominations

pnkfelix (Apr 30 2020 at 13:33, on Zulip):

that are tagged as T-libs

pnkfelix (Apr 30 2020 at 13:33, on Zulip):

and other than that, there is nothing explicitly T-libs on the agenda. Is that correct?

pnkfelix (Apr 30 2020 at 13:34, on Zulip):

I have to admit, I'm starting to worry that we might not be well equipped to evaluate backport nominations

simulacrum (Apr 30 2020 at 13:35, on Zulip):

you mean t-libs backport noms?

pnkfelix (Apr 30 2020 at 13:36, on Zulip):

yes

pnkfelix (Apr 30 2020 at 13:36, on Zulip):

but I guess it shouldn't be too bad as long as we are conservative

simulacrum (Apr 30 2020 at 13:37, on Zulip):

they're usually infrequent (I think we've had maybe 10 over the last 20 releases or so)

pnkfelix (Apr 30 2020 at 13:37, on Zulip):

i.e. the truly worst-case scenario is that, due to a combination of pressure and inexperience, we end up approving backport nominations and injecting bugs

simulacrum (Apr 30 2020 at 13:38, on Zulip):

I am personally not too worried, but then again I personally am not sure how much the backport process is buying us today -- not that I have a better one in mind :)

pnkfelix (Apr 30 2020 at 13:38, on Zulip):

@simulacrum you mean in the sense of having the firewall approval process in place, rather than just trusting our test suite?

simulacrum (Apr 30 2020 at 13:38, on Zulip):

yeah, and the 'normal' of just one reviewer being enough

simulacrum (Apr 30 2020 at 13:39, on Zulip):

e.g. I do think beta backport should be explicit approval, just not sure it needs.. a team

simulacrum (Apr 30 2020 at 13:39, on Zulip):

and certainly it seems like it takes up a lot of time to approve things that could maybe be better spent

pnkfelix (Apr 30 2020 at 13:40, on Zulip):

I prefer to give everyone a chance to object, and I do not trust the async FCP process to get all feedback in a timely fashion

pnkfelix (Apr 30 2020 at 13:40, on Zulip):

there have definitely been backport proposals where there have been only one or two people who voiced concerns that eventually swayed the group

simulacrum (Apr 30 2020 at 13:41, on Zulip):

I'm not saying we don't reject things

pnkfelix (Apr 30 2020 at 13:41, on Zulip):

right right, I know

simulacrum (Apr 30 2020 at 13:41, on Zulip):

I'm saying that I don't know whether the time spent is worth the potential for regressions

simulacrum (Apr 30 2020 at 13:41, on Zulip):

Maybe it depends, and we could have a less rigorous approval process

pnkfelix (Apr 30 2020 at 13:41, on Zulip):

yes I believe I understand what you are saying

simulacrum (Apr 30 2020 at 13:41, on Zulip):

e.g. t-libs I personally don't really worry much about

pnkfelix (Apr 30 2020 at 13:41, on Zulip):

that's interesting

simulacrum (Apr 30 2020 at 13:41, on Zulip):

we usually have good test coverage and backports are well understood

simulacrum (Apr 30 2020 at 13:42, on Zulip):

as opposed to compiler backports, where at least I feel that the interactions are less well understood

pnkfelix (Apr 30 2020 at 13:42, on Zulip):

due to our stability promises, I myself am more concerned about T-libs backports

pnkfelix (Apr 30 2020 at 13:42, on Zulip):

i.e. accidentally introducing some change in behavior

simulacrum (Apr 30 2020 at 13:42, on Zulip):

I guess -- the reason I say this -- is that the compiler is huge and all of it interacts with each other quite a bit

pnkfelix (Apr 30 2020 at 13:42, on Zulip):

right, T-libs should be easier to grok, this is true

simulacrum (Apr 30 2020 at 13:42, on Zulip):

whereas the standard library is fairly well isolated

pnkfelix (Apr 30 2020 at 13:43, on Zulip):

There is at least one exception though

pnkfelix (Apr 30 2020 at 13:43, on Zulip):

the choice of what traits to implement

simulacrum (Apr 30 2020 at 13:43, on Zulip):

we've certainly had behavior changes, but people notice pretty quickly usually -- and most of the time, though not all, we just don't go ahead

pnkfelix (Apr 30 2020 at 13:43, on Zulip):

can have wide-reaching effects

pnkfelix (Apr 30 2020 at 13:43, on Zulip):

but that also is something that is not our choice

pnkfelix (Apr 30 2020 at 13:43, on Zulip):

since it should be part of the public API

simulacrum (Apr 30 2020 at 13:43, on Zulip):

yeah, though we'd pretty much never backport stabilizations (which trait impls essentially are)

pnkfelix (Apr 30 2020 at 13:43, on Zulip):

and therefore a true T-libs(-design ) question

pnkfelix (Apr 30 2020 at 13:43, on Zulip):

right

pnkfelix (Apr 30 2020 at 13:44, on Zulip):

so maybe I'm wrong to be so concerned about our ability to backport T-libs stuff

pnkfelix (Apr 30 2020 at 13:44, on Zulip):

and I should instead, as you note, be worried about the time suck it represents

pnkfelix (Apr 30 2020 at 13:44, on Zulip):

if we follow the same protocol that we do for T-compiler

simulacrum (Apr 30 2020 at 13:44, on Zulip):

e.g. the stdarch thing I'd personally feel comfortable self-approving

pnkfelix (Apr 30 2020 at 13:44, on Zulip):

which may lead to the question that we perhaps should follow a different protocol

pnkfelix (Apr 30 2020 at 13:45, on Zulip):

for T-libs backports

simulacrum (Apr 30 2020 at 13:45, on Zulip):

yes, and I think generally there may be a class of "not risky" backports

simulacrum (Apr 30 2020 at 13:45, on Zulip):

e.g. making something not ICE but a hard error

pnkfelix (Apr 30 2020 at 13:46, on Zulip):

yeah maybe

pnkfelix (Apr 30 2020 at 13:47, on Zulip):

well for this week, I'll follow the existing protocol

simulacrum (Apr 30 2020 at 13:47, on Zulip):

sounds good -- there's only one t-libs backport anyway that isn't approved

pnkfelix (Apr 30 2020 at 13:48, on Zulip):

but I would like to think more about ways to be more aggressive about backporting so-called low-risk backports

pnkfelix (Apr 30 2020 at 13:48, on Zulip):

without requirng full synchronous discussion

simulacrum (Apr 30 2020 at 13:48, on Zulip):

(one interesting note is that t-libs backports are essentially always individually approved currently by a t-libs member)

pnkfelix (Apr 30 2020 at 13:48, on Zulip):

we already basically have precedent for T-compiler leads to approve backports in "emergency" situatinos

pnkfelix (Apr 30 2020 at 13:49, on Zulip):

so maybe one way to go would be to have T-compiler lead(s) pre-approve backports (and just listing them in a separate section at the meeting, as a general notice)

simulacrum (Apr 30 2020 at 13:50, on Zulip):

yeah, I'd generally be in favor, though not sure on leads given that Niko and you are both very time-constrained with "more important" things.

simulacrum (Apr 30 2020 at 13:51, on Zulip):

(but I also am unsure who else, in some sense)

Last update: Jun 05 2020 at 22:45UTC