Add weekly sync meeting notes 2020-01-20

meeting-notes/2020/2020-01-20--ipld-sync.md

@@ -0,0 +1,66 @@
+# 🖧 IPLD Weekly Sync 🙌🏽 2020-01-20
+
+- **Lead:** @vmx
+- **Notetaker:**
+- **Attendees:**
+  - @vmx
+  - @warfork
+  - @rvagg
+  - @mikeal
+  - @stebalien
+  - @riba
+- **Recording:** https://www.youtube.com/watch?v=ZtPGrwamYxg
+
+
+## Agenda
+
+- General
+  - Start recording
+  - Start live stream
+  - Find a notetaker
+  - Ask everyone to put their name into the list of attendees
+  - Ask everyone to put their items that they've been working on the past two weeks (should be done prior to the meeting)
+  - Ask for last minute agenda items
+- This meeting
+  - _add your agenda item_
+
+
+## Weekly Update
+
+@vmx
+ - Catching up after vacations
+
+@warpfork
+
+- slogging away on implementation details of codegen'd maps (and corresponding improved `ipldfree.Node` maps).
+	- not much that's worth high-level report report -- didn't get as far as the benchmarks i wanted that would've made a nice summary.
+	- found a need for key&value child assemblers to need a pointer back to the map assembler -- irritated.  increases the complexity of recursion somewhat and adds another wrapper that adds another vtable.  it's all viable, but i looked for an alternate way around... and didn't find one.  biting the bullet and continuing; will hope for a clever idea to strike later.
+- one technical detail worth reporting up to the group: we've debated about whether maps maintain order?  can update the considerations to note: it's not high cost to do so in go-ipld-prime (which we might've previously suspected).
+	- turned out "free"... or more precisely, swallowed up by another cost unambiguously worth paying: the slices used for amortizing a bunch of allocations and preventing 'convT' costs can also be used for ordering.
+
+@rvagg
+ - Mostly Filecoin focused, CommP - understanding, digging in Go & Rust and extracting utilities to perform CommP calculation outside of Filecoin, including a Lambda service to do it for S3 resources. Leaning heavily on https://github.com/filecoin-project/go-fil-markets/ for the algorithm so we don't have to own much of it.
+
+@mikeal
+ - reviews, okrs, and other managey stuff
+ - resolved some threads on car files and data processing, unblocked now.
+
+
+## Notes
+
+<!-- After each call, the notetaker submits a PR to https://github.com/ipld/team-mgmt to store the notes on the meeting-notes folder -->
+
+ - Ordering of map keys
+     - Discussion regarding map key ordering, coming out of  https://github.com/ipld/specs/pull/230 discussion re deterministic selectors - how to be deterministic over maps? We need consistent key ordering rules.
+     - @mikeal: JavaScript perserves the order the map was written in
+     - @warpfork: (sorry for dropping out, ~~technical~~ consent-based-software difficulties)
+     - @warpfork: this is interesting in go-ipld-prime because most of the Node stuff is written *codec-agnostic*.  So my *nodes* will maintain order... because they might be used with two different codecs, and maybe those two codecs have different ideas about sorting requirements.  /2c.
+     - Conclusion:
+         - Dag* formats (like DagCBOR): sort in a certain way in write time, on write time they just read in how it was written => ordered
+         - Validation should be applied on our strict formats (DAG*) on read - it's an error if the keys appear out of order
+         - ( riba ) When writing spec make abs. sure **unicode normal form** is specified
+            - ( warpfork) dissent
+              - ( vmx ) solved by a general comment on the Data Model that the normal form should be NFC: https://github.com/ipld/specs/pull/350/
+ - Discussion about rejecting things on the network level
+     - @riba: Do it there
+     - @stebalien: Always do it at the same place, i.e. not one time at parsing time, and one point at networking time