forked from rust-lang/rust
-
Notifications
You must be signed in to change notification settings - Fork 0
Meeting weekly 2012 01 03
Graydon Hoare edited this page Jan 6, 2012
·
1 revision
- Shall I push the new block syntax?
- marijn: Yes, seems harmless enough, we'll "see how it goes"
- Replace for with block
- break, cont, etc
- graydon: this was all discussed
- nmatsakis: we could add another kind of proto (loopblk) that permits break, cont, etc wraps result in a special tag
- nmatsakis: Been working on cleaning up the task library
- interfaces:
- marijn: making good progress on interfaces, starting to Just Work
- method resolution (compile-time) quite a bit slower this way
- added a lot more vectors, causing more malloc/free
- long term perf graph might help for monitoring perf regression
- maybe just exposing all the data we have
- rustbot:
- graydon: we're not getting "multiple builds" anymore
- msg broker was sending multiple build requests, causing multiple builds
- reseting the poll server seemed to solve this
- maybe we can use an EC2 cluster to achieve better latency, more builds
- currently very limited testing of ia32
- we could perhaps use multiple bots per machine as build is not particularly parallel
- perhaps we could just ignore stage3, or not build it
- change stage3 to once per day if at all?
- EC2 is expensive ($$), Cloudkick or Rackspace is better
- constant on (like rustbot) is not good on EC2
- graydon: one other problem with virtualized machines:
- linux machines super easy, windows hard, mac impossible
- maybe testing linux is enough
- marijn: get better control over work queue of bots, allow to cancel jobs etc
- graydon: all bots are listening to one another, in an async. comm. pool
- easy for bots to listen to each other and if one fails they could all fail
- maybe not a good idea, interesting to know if one platform fails
- add a command for rustbot to edit work queue
- graydon: working on documentation drift
- wants docs to autogenerate whenever we commit
- have a tool, rustdoc, that is evolving and would allow
- command line query support for documentation
- would allow emacs modes to better integrate for documentation
- Release management
- shall we hold release until interfaces are in place?
- new code would be shaky though
- should discuss with bigger group
- usable in another week, maybe can replace some objects
- graydon: do we need objects + interfaces?
- lots of overlap
- marijn: pcwalton still plans to go fwd with classes
- would integrate with interfaces and allow for classical patterns
- current obj system doesn't add anything to new work
- nmatsakis: I'm inclined to agree, obj system doesn't add a lot
- ad-hoc per object extension largely unused, so not an imp't feature?
- maybe used in the I/O system?
- marijn: nominal records alongside structural ones?
- better for recursive types
- better for specializing interfaces
- graydon: personally he likes keeping nominalness orthogonal
- nmatsakis: I don't like having nominal and stuctural recs,
- but calling nominal records classes might be ok
- graydon, marijn: absolutely do not want to give up structural records
- timeline for release: this month
- focus on the release
- anything missing from issue milestones?
- shall we hold release until interfaces are in place?
- nmatsakis: no copying proposal?
- marijn approves
- unique vectors are a pain
- nmatsakis: I started work on converting compiler to use vec:map() etc
- so that we know how well fixed-size arrays will work
- but more to be done