add 'rdf' (#804)

[kevinushey]

Adds rdf as a little function helper for creating tbl_dfs inline, e.g.

rdf(
    a | b | 1 | 2,
    c | d | 3 | 4
)

Comments appreciated (probably most pertinent one -- what about column names?)

Other things to think about:

• is rdf the best name?
• should we provide a column-wise analog (ie, like rdf but transposes the created frame)? Such a function could accept named arguments (for named columns)
• allow for leading or trailing commas?

[hadley]

Maybe the first row should always be column names?

I definitely don't like rdf() as a function name.

Another approach that would be a bit easier to parse would be:

rdf(
    hdr(a, b, c)
    row(a, b, 1, 2)
    row(c, d, 3, 4)
)

Another approach would require a sentinel value at the end of each row:

rdf(
  a, b, c, d, END
  a, b, 1, e, END,
  c, d, 3, 4
)

(But I don't think you could make that as elegant as the other two approaches)

[kevinushey]

I definitely prefer not having to add 'extra' syntax to make the table; we should keep it as 'markdown-like' as possible.

Perhaps a dummy 'delimiter', or 'delimiter-like' symbol, could be used, e.g. a series of dots:

rdf(
    these | are | names,
    ...................,
    these | are | cells,
)

)

I would much prefer making the 'parser' more complicated if it meant keeping the user experience good + language simple.

[hadley]

I like the series of dots idea.

[kevinushey]

Just added some code to support the dots delimiter, and also we drop missing arguments. So we can now write e.g.

df <- rdf(
   , these | are | names
   , ...................
   , these | are | cells
 )

(note that symbols are implicitly converted to character)

Still need a nice name -- any thoughts? Maybe something like make_tbl() verbatim_tbl()?

[coveralls]

Coverage decreased (-0.77%) to 66.26% when pulling cbe46cd on kevinushey:feature/rdf into 2e298fc on hadley:master.

[coveralls]

Coverage decreased (-0.57%) to 66.46% when pulling 0b1c9eb on kevinushey:feature/rdf into 2e298fc on hadley:master.

[coveralls]

Coverage decreased (-0.57%) to 66.46% when pulling 0b1c9eb on kevinushey:feature/rdf into 2e298fc on hadley:master.

[lionel-]

doodle() ?

[lionel-]

another idea for specifying column headers:

data <- doodle(
  ha : ho : hi,
   1 |  x | NA,
   3 | NA | NA,
)

[hadley]

@jennybc I know this is something that you're interested in. Any thoughts on name, syntax, etc?

[jennybc]

Here's a mash-up of the above ideas

tibble(
  these, are, names,
  ---,
  these, are, values,
  these, are, values
)

Who doesn't like tibble()?!?

I am much faster at typing commas than | (tho the nod to markdown is not lost on me).

You could use the markdown-inspired "3 dashes" to signal the presence of variable names. A series of dots might lead to confusion with ...?

Maybe instead of a sentinel value to signal the end of one observation, just require variable names and get length there? That might be less of a drag than repeatedly typing the sentinel.

[lionel-]

tibble() is an awesome name.

I like your --- idea. It could be 3 characters or more, so that a fully formatted table looks like:

tibble(
  these, are,  names,
  ------------------,
  these, are, values,
  these, are, values
)

[hadley]

I love the name tibble, but I don't think - will work as a separator because it's not a valid R expression. Unless some else has a really bright idea, I think we're stuck with .......

I like the idea of forcing a header row so that we can automatically determine how many columns there are, so can use the implicit newline as a row separator.

[lionel-]

ah yes... and with the dots there's another issue, the function won't work with only three dots.

the following is maybe a bit annoying to type

tibble(
  these, are,  names,
  `----------------`,
  these, are, values,
  these, are, values
)

This would work but looks like a hack:

tibble(
  these, are,  names,
  -----------------.,
  these, are, values,
  these, are, values
)

[jennybc]

I definitely don't understand exactly what will work as the thing that separates variable names from the data, but I get the general drift.

That said, can the twiddle be pressed into service? It's easy to make a valid expression with it.

[lionel-]

maybe the simplest thing:

tibble(
  'these', 'are', 'names',
   these,   are,   values,
   these,   are,   values
)

possibly with an optional ruler:

tibble(
  'these', 'are', 'names',
  '---------------------',
   these,   are,   values,
   these,   are,   values
)

[hadley]

The problem with using an infix function (like ~) is that you need to have something on the right hand side of it (i.e. ~~~~~~~~~~x would work but, ~~~~~~~ would not). I think we need to use something that's a valid R variable name (i.e. can start with letter or ., and contain letters, digits, . and _. That doesn't leave many options, but we could do something like:

tibble(
  "col1", "col2", 
  HEADER,
  1, 3
  2, 6
)

I also don't think we want to drop the quotes for strings altogether because that will confuse people who want to put spaces in their values. So maybe we could flip @llionel-'s idea around and do:

tibble(
  col1, col2,
  1, 2,
  3, 4
)

But then the header doesn't stand out much. Another idea would be to eliminate the NSE altogether and require formula quoting for column headers:

tibble(
  ~col1, ~col2,
  1, "a",
  3, "b"
)

I quite like that.

(I also just thought of a another name: frame_data() because it's a transposed data_frame().)

[jennybc]

@hadley's last proposal looks really good to me:

tibble(
  ~col1, ~col2,
  1, "a",
  3, "b"
)

[lionel-]

and once again ~ saves the day

[hadley]

@jennybc how do you feel about frame_data() as a name?

@kevinushey do you want to take a shot at implementing this? Or should I? I think it should be fairly simple now.

[jennybc]

I like frame_data() and think it's clever. I still like tibble() too: brings back fond memories of the epic tibble diff discussion. You have a great track record with naming things so I'm not bothered either way.

[kevinushey]

@hadley I'll give it a shot and update this PR.

[kevinushey]

New PR at #1358.

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/