Mark Klik

A _fst\_table_ object is expected to mimic the behavior of a _data.frame_. Therefore, `df[0, ]` syntax should return a zero-row table without an error or warning: ``` r tmp_file 1...

Required API for slice()

5

Method `fst::read_fst` needs an API change to accommodate `slice()`. In addition to parameters `from` and `to` a (possibly randomly ordered) `integer` vector should be accepted. The [current implementation](https://github.com/krlmlr/fstplyr/blob/master/R/tbl.R#L50-L53) for `read_from_meta`...

Use fst to write a subset of rows and columns

2

Following the discussion [here](https://github.com/fstpackage/fst/issues/124#issuecomment-357175069), `fst` could provide the API to write a subset of the supplied dataset. That would have the advantage that no copying would be required in column-...

Just-in-time computation

1

Hi @krlmlr , In `fst`, calculations on the main thread can be done while loading data on background threads. For many column types this means that user operations can be...

Method lazyvec_skeleton() generates a full sample implementation

1

To create a custom ALTREP vector, method `lazyvec()` needs overrides for methods from the ALTREP API. For the average user, it will take some time to study this API to...

At default, ALTREP's are expanded when attributes change. By a smart override of the duplicate method, a full expansion can be avoided (with a mechanism identical to wrapper objects).

Currently, setting an attribute to an ALTREP vector, triggers a copy and after that the vector is no longer an ALTREP: ``` r x [1] TRUE class(x) [1] FALSE ```...

When an ALTREP vector is de-serialized, custom code is executed in the `UnserializeEX` client method. That code can contain anything so can be be potentially harmful. Normally, `R`'s de-serialization mechanism...

Apart from the method above, many other `ALTREP` methods are not exported yet on `R-devel` on linux. This seems to be caused by a _attribute\_hidden_ definition in front of the...

Phase 3 of [milestone 1](../milestone/1): method `lazyvec()` has the option to enable/disable checkS on the result of user-defined `R` methods. Typically, the users will enable these checks when first developing...