johnmyleswhite

If you make a change to Julia Base by adding a new file, you'll probably want to do the following:

• Add the file as an include in base/sysimg.jl
• Add the few exported names to base/exports.jl
• Add a test file in test and then activate the test by modifying test/runtests.jl

If you only modify an existing file, you should just update base/exports.jl.

johnmyleswhite

> type.unstable <- function(x) {
+     if (x > 0) {
+         return("foo")
+     } else {
+         return(NA)
+     }
+ }
> 
> class(type.unstable(0))
[1] "logical"

johnmyleswhite

Values vs. Bindings: The Map is Not the Territory

Many newcomers to Julia are confused by the seemingly dissimilar behaviors of the following two functions:

julia> a = [1, 2, 3]
3-element Array{Int64,1}:
 1
 2

johnmyleswhite

• Replace NAtype with Nullable{T}
  • This will substantially change the semantics of missing data
    • You will have to fully specify how missing values are handled
    • You cannot leave off thinking about missing values to the first time you hit one
    • Your code will become type-stable and will run much faster (~100x)
    • The name NA will be replaced by NULL
      • This implies isna will become isnull
    • There will be no more attempts to define functions over NA
      • This includes Booleans: there will be no more three-valued logic
• Replace DataArray with NullableArray

johnmyleswhite

Missingness as boxes with unknown content. Cf. the Smarties task: http://www.jstor.org/discover/10.2307/1130386?uid=3739920&uid=2&uid=4&uid=3739256&sid=21104622658277

Understanding pointers and mutable containers. Confer the development of object permanence and object identity: http://en.wikipedia.org/wiki/Object_permanence

johnmyleswhite

> v <- c(1, 2, 3, 4)
> v[1]
[1] 1
> v[4]
[1] 4
> v[0]
numeric(0)
> v[5]
[1] NA
> v[-1]

johnmyleswhite

> v <- c(1, 2, 3, 4)
> ind <- NA
> ind
[1] NA
> v[ind]
[1] NA NA NA NA
> ind <- -ind
> ind
[1] NA
> v[ind]

johnmyleswhite

Papers

Benchmarking seems not to be a main focus of any specific academic field, although the problem has been addressed by many different groups in CS.

Some papers I found interesting:

• http://research.microsoft.com/en-us/um/people/nswamy/papers/bottlenecks-ecoop04.pdf
• http://www.sigmod.org/publications/sigmod-record/0806/p45.dewitt.pdf
• http://daniel-wilkerson.appspot.com/trend-prof.pdf
• http://sape.inf.usi.ch/sites/default/files/publication/EvaluateCollaboratoryTR1.pdf

johnmyleswhite

# TODO: Need different behavior for the REPL
function set_stdin_mode!(raw::Bool)
    res = ccall(
        :jl_tty_set_mode,
        Int32,
        (Ptr{Void}, Int32),
        STDIN.handle,
        int32(raw)
    )
    return res != -1

johnmyleswhite

> x <- c(4, 1, 3, 5, 2)

> # Sorts levels automatically
> f1 <- factor(x)
> f1
[1] 4 1 3 5 2
Levels: 1 2 3 4 5

> # Does not sort levels automatically
> f2 <- factor(x, levels = c(3, 4, 5, 2, 1))