Software Engineering.page

TODO merge in notes from sweng class

game development

- asm is still freq used
- c++ is the de facto std bc of physical memory control (rather than raw perf)
- often have heavy duty scripting infrastructure
- equal numbers of engineers/artists, with slightly fewer designers (scripters)

concepts

- scrum
- dependency injection: a way of structuring your code for testability
  - banish global state/singletons
  - explicitly pass objects/object factories around to ctors
  - ctors are very simple
  - initialize object/object factory graph at startup
  - easy to substitute different implementations for testing

characterizing and predicting which bugs get fixed: emp study of ms win (philip guo)

- bug fixing is imp; triaging is hard
- ms emp data on vista/7: bug lifetime, employee info
- qualitative survey to win devs with 20% response rate
- main question: how do factors related to ppl and bug report edits affect whether bug is fixed?
  - reputation, interest, reassignments, distance
- prev paper on bug triaging: bug opener reputation = |opened cup fixed| / (|opened| + 1)
  - +1 encourages more bug reports
- built predictive model: logistic regression with 7 factors
  - source, rep of opener, rep of first assignee, severity, opened by temp?, opener/assignee in same team?, same bldg?
  - 68% prec 64% recall (.5 cutoff)

anecdotes

- oopsla09: nasa jpl uses model checking (heavy duty theory) and static analyses (less theoretical but still highly useful)

Exploding Software-Engineering Myths

- <http://research.microsoft.com/en-us/news/features/nagappan-100609.aspx>
- code coverage is not a good metric
- TDD team 60-90% better in terms of defect density than non-TDD
- assertion/verification count negatively correlated with bug count
- org structure matters a lot: org metrics can predict SW failure-proneness
  with precision & recall of 85%
- geo distance doesn't matter much

writing solid code

- use warnings, static checkers, unit tests
- use assertions and debug-only backup algos
- use subsystem integrity checks: dangling ptrs, lost mem blocks, use of uninitialized mem
- use debugger and step
- use good design to create simple single-purpose functions
- use safer alternatives to risky algos/lang idioms
- avoid bad practices
- attitude & habits

companies
---------

facebook software engineering

- minimze downtime, scale fast, rapid deployment
- limited QA but many code reviews
- A/B testing, gradual/staged (geo) deployment
- PHP, memcached, mysql, hadoop, scribe, hive, hipal, haystack, thrift
- user:dev ratio is ~1.1M
- <http://cacm.acm.org/blogs/blog-cacm/51564-extreme-agility-at-facebook/fulltext>

google chrome

- trybot farms; can submit your changelists
- use gtest, gmock, valgrind, thread sanitizer

linux
-----

- mostly-hierarchical graph of development trees
  - tree per subsystem
  - staging trees: -mm, linux-next
  - <http://ldn.linuxfoundation.org/book/24-staging-trees>
- maintainers
  - david miller, greg kroah-hartman, andrew morton, ingo molnar
  - stats: <http://lwn.net/Articles/385949/>