Flip expiration times #13912
Flip expiration times #13912
Conversation
|
These all correspond to millisecond timestamps... and now they’re reversed? If our model weren’t so tied to a timeline and these were merely relative priorities, I would like this. But since they represent deadlines I find it confusing. |
|
ReactDOM: size: -0.2%, gzip: -0.2% Details of bundled changes.Comparing: bf9fadf...29e1b1f react-dom
react-art
react-test-renderer
react-reconciler
react-native-renderer
scheduler
Generated by 🚫 dangerJS |
|
You win some, you lose some. Just pretend you're looking in a mirror. |
|
Want me to make them all negative? 🙄 min 31-bit int - Sync Then 0 is in the right place and it goes in the right direction… |
|
I think we can find a mental model where this makes sense. E.g. a count down until the app stops working (NoWork). |
|
I’m cool with this if we rename the field, too. Like |
| (childChildExpirationTime !== NoWork && | ||
| childChildExpirationTime < newChildExpirationTime) | ||
| ) { | ||
| if (childChildExpirationTime > newChildExpirationTime) { | ||
| newChildExpirationTime = childChildExpirationTime; | ||
| } |
There was a problem hiding this comment.
At this point we could even go so far as to do
newChildExpirationTime = Math.max(
newChildExpirationTime,
child.expirationTime,
child.childExpirationTime,
);
so simple… wow
|
I kinda like the negative clock (that becomes less negative over time) actually… what about you two? |
|
Nah. Negative is weird. That's not Rust/WASM/et al friendly since we'd want the unsigned form there. |
|
If we didn’t already have something called “priority” I would say let’s rename it to that and merge your PR as is. I don’t like the “expirationTime” naming regardless because it only acts like an expiration time when we schedule the callback with Scheduler. Everywhere else it acts like a relative priority. Precedence? Urgency? Rank? (I’m just reading the thesaurus) |
|
"Threshold" maybe. |
|
Importance? |
|
Note that we also plan to add another public priority API which is not related to time (the priority of fallbacks returning to their unsuspended state). |
|
Urgency? |
|
We're hopeless. |
|
Oh I missed @acdlite already suggested urgency. |
|
@sebastian yeah and I also don’t think we understand the relationship between the two concepts yet, either. |
|
restlessness |
Agree, we use the priority name before but then use expirationTime to abstract too many things for processing starvation and competition better. But just from a developer perspective, the expirationTime is confused and hard to understand(lower means higher, higher means lower), and when you compare it to some names like |
``` codemod --accept-all -d packages -i '((?:expiration|suspended|pending|current|remaining|absolute|renderer|pinged|completed)time(?:(?:before|after)commit)? )<' '\1dnkvjchbdddtngeurvlbvbktrilldeeu' codemod --accept-all -d packages -i '((?:expiration|suspended|pending|current|remaining|absolute|renderer|pinged|completed)time(?:(?:before|after)commit)? )>' '\1<' codemod --accept-all -d packages -i '((?:expiration|suspended|pending|current|remaining|absolute|renderer|pinged|completed)time(?:(?:before|after)commit)? )dnkvjchbdddtngeurvlbvbktrilldeeu' '\1>' git checkout packages/scheduler/src/Scheduler.js ```
Ran these, then prettier, then these again, then prettier again. ``` codemod --accept-all -d packages -m '\(\s*([a-zA-Z.]+) === NoWork\s*\|\|\s*\1 (<=?) ([a-zA-Z.]+)\s*\)' '(\1 \2 \3)' codemod --accept-all -d packages -m '\(\s*([a-zA-Z.]+) === NoWork\s*\|\|\s*([a-zA-Z.]+) (>=?) \1\s*\)' '(\2 \3 \1)' codemod --accept-all -d packages -m '(?<![.\w])([a-zA-Z.]+) !== NoWork\s*&&\s*\1 (>) ([a-zA-Z.]+)' '\1 \2 \3' codemod --accept-all -d packages -m '(?<![.\w])([a-zA-Z.]+) !== NoWork\s*&&\s*([a-zA-Z.]+) (<) \1' '\2 \3 \1' codemod --accept-all -d packages -m '(?<![.\w])([a-zA-Z.]+) === NoWork\s*\|\|\s*\1 (<=?) ([a-zA-Z.]+)\s*\?\s' '\1 \2 \3 ? ' codemod --accept-all -d packages -m '(?<![.\w])([a-zA-Z.]+) === NoWork\s*\|\|\s*([a-zA-Z.]+) (>=?) \1\s*\?\s' '\2 \3 \1 ? ' ```
See facebook#13912 commit messages for how this was done.
Now:
0 - NoWork
1 - Never
...
big - low pri things
even bigger - hi pri things
max 31-bit int - Sync
Per @sebmarkbage in #13904.