Labs/Jetpack/Weekly Meeting/2011-07-19
From MozillaWiki
< Labs | Jetpack | Weekly Meeting
Contents
Agenda
- FlightDeck 0.9.7 status
- SDK 1.1 status
- Development Process
- Roundtable
Minutes
FlightDeck 0.9.7 status
- bug fixes, polish
- should be done by friday (12-14 bugs outstanding)
- otherwise, we'll need another week
- afterward, we'll start on roadmap
- 0.9.7 is scheduled to ship next Wednesday, freeze is on Friday
- 0.9.7 will be shipped whether or not the remaining bugs are fixed
- next release is 0.9.8
- two week cycle, 0.9.8 ships two weeks after 0.9.7
- we know what makes addon testing and downloading slow
- https://builder-addons.allizom.org/services/graphite/prod
- bright purple spikes on graph are queue time
- something is going wrong occasionally with process of putting packaging job into "rabbit" queue (service for queuing requests)
- when it doesn't go wrong, 6-8 seconds with good connection
- when it goes wrong, a minute and a half
- working on this to try to resolve the problem by this friday, ship it in 0.9.7
- two separate perf problems: the occasional glitch, and the "normal" time of 6-8 seconds (want it to be under 4 seconds)
- cfx xpi doesn't seem to be the issue (takes less than a second)
SDK 1.1 status
- Irakli
- submitted bug regarding module exports that are supposed to be frozen but aren't, bug 672199, creating potential for malicious code to overwrite APIs; created bug, attached patch, brian to review
- another bug is adding strict mode to our modules, bug 553434
- brian: turn on strict automatically rather than "use strict"?
- irakli: we decided to turn it on for just our modules
- test failures; i could reproduce two errors, myk was supposed to push to mozilla-central to see if third is still reproducible
- myk: i haven't done so but will do so today
- bug 672087: module search has complex algorithm for finding a module
- talked about explicit require statements so no search for SDK's own modules
- irakli created patch; brian to review; patch might need updating to trunk
- bug : last week we talked about developer ergonomics...
- console.log is almost useless
- pull request that changes console.log behavior to dump object in way that gives access to properties and values
- brian: like Python's pdump?
- irakli: like nodejs
- brian: depth restriction?
- irakli: not sure, but it's configurable
- Matteo
- working on selection bug
- finally found a workaround that makes it work for multiple selection
- not sure if i have to create a bug for the firefox platform or for the SDK
- (some discussion)
- multiple selection doesn't work well in the platform, not in HTML, not in text fields
- there might be an API design issue here; selection module returns first range by default
- Alex
- content proxy: submitted multiple review requests to irakli
- multple issues with customized toolbars: tried to reach folks on #developers, but without much success
- will try to reach another guy
- uninstall event: submitted some requests for comments on google groups, needs feedback from brian, may require a change in linker, related to discussion about e10s
- bug about uninstall is bug 571049. content script/proxy ones are 666547,612726,669884,667949. customize toolbar issues are 660857,660860
- Will
- looking into removing the cfx docs server
- current idea is to generate static docs locally and open them in web browser
- the problem is that it'll be slow the first time you run cfx docs
- brian: i like it, seems cleaner
- will: code is simpler, will make it easier to get away from python
- brian: time to build docs?
- will: several seconds to build them from scratch
- brian: i'm happy with that, as long as there is progress marker
- will: and that's one time only
- brian: and we should be able to regenerate only changed docs
- the other thing i've been thinking about is migration guide for XUL developers
- trying to figure out what that should have in it and how to explain limitations
- that you can't manipulate browser chrome to the same extent
- brian: we need to figure out which of these restrictions are policy vs. priority vs. technical
- will: yes, and it'd be nice to have a UI guideline for addon developers, not necessarily related to the SDK
- (discussion about extent to which you can do what the SDK doesn't provide APIs for)
- Gábor
- built an XPCOM component, packaged it into XPI, installed it, made it work
- bit tricky because of lack of documentation
- built Mozilla from source, set up Windows development environment
- Windows development painful, mostly because of VM, issues with shared folder
- but in the end it worked, mostly just took a lot of time; tuned it to use six cores to reduce build times
- tried to reproduce sandbox leak, so far have not been able to reproduce
- digging into it, it's complex, feel a bit lost, but i will get there
- Eddy
- did a few measurements today about how long it takes to create a tab within content process
- jetpack process takes 70-90ms to create
- content process 120-150ms, if you also create tab is 210-240ms
- content process memory footprint is 22mb without tab, 27mb with tab
- talked to bsmedberg, he thinks we can do better than 27mb
- talked about content processes, he thinks it isn't necessarily a bad idea, but it might be expensive
- so there are possibilities, but for him it was more a big deal that we'd have to share addons between processes instead of one addon per process
- whereas with jetpack processes, we could have a one-to-one mapping
- and that's a harsh price to pay
- so he would like to see if we can do both, i.e. use content processes but have one process per addon
- his suggestion is next step is to post to dev.platform
- so i will make summary of tests and circulate it in email
- tomorrow is my day off; plan to make post by tomorrow or thursday morning at latest
- next week i want to get as much feedback as possible
- so we can move forward and make a decision
- one more thing: friday i'll be visiting an office location in utrecht area
- Myk
- working on test automation issues
- not much progress
- Brian
- not much to report, just got back from trip
- various cleanup tasks, want to hear back from dcm about priorities before jumping in on the larger items
Development Process Proposal
- get your feedback in to Myk
Roundtable
- Alex
- would like to have feedback from Brian and Irakli on how to handle uninstall events
- Alex to cc: Brian and Irakli on the bug
- Irakli
- simpler module loader: last week we talked about simplifying things
- i got so excited i worked on simplified module loader over weekend
- it uses platform's "JavaScript module" loader instead of our own loader
- it can load all modules in SDK
- but it needs some changes on the linker side, because our test infrastructure, doesn't rely on linker, it uses runtime linking
- started a thread on the mailing list, asked for feedback
- cc:ed Brian on pull request, interested in getting feedback
- this change has potential to fix issues like XPI unpacking and merging cuddlefish, securable-module, bootstrap, and harness scripts
- feedback welcome!
- Daniel
- alex, have you had time to look at pulling out the client-side linker for use by FlightDeck?
- alex: i started looking at how to build an addon for the SDK that would provide command line interface like the current SDK
- alex: but i haven't started looking at the client-side linker
- Daniel: when your internet connection is bad, Builder packaging is a problem
- Brian, Myk: it isn't clear that packaging is actually the issue, if packaging on the server is cheap, and if you have to push the files down anyway
- Daniel: ideally, the files will be pushed down already, when you start editing the addon
- Brian: can you cc: me on the bug so i can talk to Dietrich about the problem?
- Daniel to cc: Brian and Myk on the bug
- Alex: i looked at the bug, and there is one missing API that is easy to do that would let us start experimenting with alex's prototype
- Daniel to talk to Piotr about providing that API
