Consider syntax for assigning intermediate value to symbol

[renkun-ken]

Consider syntax for assigning intermediate value to symbol

[renkun-ken]

A draft syntax is like

x %>>% (~ symbol)
x %>>% (~ f(.) ~ symbol)
x %>>% (~ x ~ f(x) ~ symbol)

It can be best described by Start with ~ for side effect and end with a symbol for assignment.

An example is

mtcars %>>%
  subset(mpg <= mean(mpg)) %>>%
  (~ smtcars) %>>%
  (~ dim(.) ~ dim_mtcars) %>>%
  subset(select = c(mpg, wt, qsec)) %>>%
  lm(formula = mpg ~ .) %>>%
  summary %>>%
  (~ summ) %>>%
  (coefficients)

              Estimate Std. Error   t value     Pr(>|t|)
(Intercept) 17.0183914  5.1411954  3.310201 0.0047583237
wt          -2.9781345  0.6044032 -4.927397 0.0001823504
qsec         0.6033051  0.3053237  1.975953 0.0668509086

Inspect the environment after evaluating the code above.

> ls.str()
dim_mtcars :  int [1:2] 18 11
smtcars : 'data.frame': 18 obs. of  11 variables:
 $ mpg : num  18.7 18.1 14.3 19.2 17.8 16.4 17.3 15.2 10.4 10.4 ...
 $ cyl : num  8 6 8 6 6 8 8 8 8 8 ...
 $ disp: num  360 225 360 168 168 ...
 $ hp  : num  175 105 245 123 123 180 180 180 205 215 ...
 $ drat: num  3.15 2.76 3.21 3.92 3.92 3.07 3.07 3.07 2.93 3 ...
 $ wt  : num  3.44 3.46 3.57 3.44 3.44 ...
 $ qsec: num  17 20.2 15.8 18.3 18.9 ...
 $ vs  : num  0 1 0 1 1 0 0 0 0 0 ...
 $ am  : num  0 0 0 0 0 0 0 0 0 0 ...
 $ gear: num  3 3 3 4 4 3 3 3 3 3 ...
 $ carb: num  2 1 4 4 4 3 3 3 4 4 ...
summ : List of 11
 $ call         : language lm(formula = mpg ~ ., data = .)
 $ terms        :Classes 'terms', 'formula' length 3 mpg ~ wt + qsec
 $ residuals    : Named num [1:18] 1.658 -0.813 -1.643 1.386 -0.376 ...
 $ coefficients : num [1:3, 1:4] 17.018 -2.978 0.603 5.141 0.604 ...
 $ aliased      : Named logi [1:3] FALSE FALSE FALSE
 $ sigma        : num 1.79
 $ df           : int [1:3] 3 15 3
 $ r.squared    : num 0.623
 $ adj.r.squared: num 0.573
 $ fstatistic   : Named num [1:3] 12.4 2 15
 $ cov.unscaled : num [1:3, 1:3] 8.217 -0.178 -0.437 -0.178 0.114 ...

[renkun-ken]

Given all syntax with (~ ...), operator ~ can be viewed in this context to be branching operator, which indicates that the following expression will be a side effect. It can either branch the left-hand side value to an expression (side-effect evaluation), or branch it to a symbol (assignment). After all, there's no point to evaluate a symbol for side effect (no side effect at all). Therefore this syntax seems not to create additional confusion or work at the expense of possible actions allowed in cases without this feature.

[renkun-ken]

Consider the = syntax suggested by @yanlinlin82.
See #38.

[renkun-ken]

The following code adopts the = syntax.

mtcars %>>%
  subset(mpg <= mean(mpg)) %>>%
  (~ smtcars) %>>%   # side-effect assign
  (~ dim_mtcars = dim(.)) %>>%   # side-effect assign
  subset(select = c(mpg, wt, qsec)) %>>%
  lm(formula = mpg ~ .) %>>%
  (sum_lm = summary(.)) %>>%   # eval and assign
  (coefficients)

[timelyportfolio]

Definitely prefer this. I think this is much clearer, intuitive, and more readable.

[renkun-ken]

Think so too. Thanks @yanlinlin82 for the great suggestion. I'll implement it at branch feature/assign soon and see how it works.

[renkun-ken]

The latest commit at feature/assign uses symbolic call to perform the assignment, which allows the following usage:

> z <- list()
> 1:10 %>>% (~ z$a = length(.)) %>>% mean
[1] 5.5
> z
$a
[1] 10

That is, the assignment no longer calls assign() but builds a symbolic call to perform the assignment, which does not require the expression on lhs of = be a symbol and allows the usage like names(a) = ....

[yanlinlin82]

That is more powerful!​

[renkun-ken]

In v0.5, <- and -> will no longer be interpreted as lambda expression and are allowed to perform assignment in a pipeline, which makes the code even more readable in some cases.