Comments on lazy

[maelle]

@krlmlr

• it is confusing that read_parquet_duckdb() has an argument called lazy whereas duckdb_tibble()'s similar argument is called .lazy.
• The README state that you can only have automatic fallbacks for an eager data.frame but isn't this a design choice? Couldn't duckplyr use collect() before the fallback? I think this could be clearer in the docs.

[maelle]

Is what's in https://duckdb.org/2024/04/02/duckplyr.html#eager-vs-lazy-materialization still valid regarding laziness? Hasn't the default of lazy changed over time? (or are there two levels of laziness 🤪 )

Also note that https://duckplyr.tidyverse.org/dev/articles/developers.html#eager-and-lazy-modes has good content

[maelle]

A table of lazy vs eager might make sense.

Mode	Eager	Lazy
Materialization/computation triggered	at creation	when requested by collect()
Direct access with $	Works	Fails (?)
Automatic fallbacks to dplyr	Work	Fail
Default for	In-memory data (duckdb_tibble())	Remote data (read_parquet_duckdb())
Recommended for	small to medium-sized data	large data sets

[maelle]

If a vignette is created it should include a full example of compute(lazy = FALSE) usage.

[maelle]

In the reference there's a single section https://duckplyr.tidyverse.org/dev/reference/index.html#connecting-copying-and-retrieving-data

Why not split it into a section with functions for lazy frames, whose title would contain "lazy" and "materialization" (collect, compute*) and other functions (pull and explain).

[krlmlr]

Thanks. I now think this should be a vignette. The creation of the vignette is tracked in #207, below are my clarifications.

• lazy vs. .lazy : The leading dot seems mandatory for functions where the args in ... are user-supplied, can't find a reference now. On the other hand, it feels superfluous for the other functions.
• Automatic collect() : we protect the user against a huge result set. Implementation-wise, the only difference between "eager" and "lazy" is a flag that prohibits automatic materialization.
• Perhaps we shouldn't use "eager" and "lazy" after all: Rethink "eager" vs. "lazy" #473 . The table is incorrect too.
• https://duckplyr.tidyverse.org/dev/articles/developers.html#eager-and-lazy-modes can be picked up when rearranging vignettes
• Agree that a vignette about automatic materialization should contain code that demonstrates the behavior in detail.

[maelle]

The table is incorrect too.

Could you help me fix it? I think a table would help.