Issue

Currently nodes returning NULL will be considered "invalid":

"null": The errored target continues and returns NULL. The data hash is deliberately wrong so the target is not up to date for the next run of the pipeline.

As a result, even if the underlying result should be NULL, the pipeline will still consider the node to be invalid. One of the consequences I often find is when making targets with shortcut=TRUE, the pipeline raises errors if NULL values are encountered.

Is there any way to avoid raising errors in such cases? Maybe users can set options to ignore NULL... but then I found out if a node is truly invalid, then we still need to raise errors.... Duh

Propose

Maybe we can consider a less frequently used symbol (you can create it via alist(a=)[[1]]) to store "invalid results"? This symbol is very special, normally you are unlikely to create it elsewhere in base R (for example, str2lang("") and quote() will raise errors) unless using the low-level C interface via R_MissingArg symbol. Also trying to evaluate this symbol will trigger "missing" error (instead of "object not found"):

missing_arg <- alist(a=)[[1]]
eval(missing_arg)
#> Error in eval(missing_arg) : 
#>   argument "missing_arg" is missing, with no default

This symbol perfectly fits in the situation where a target is errored (the corresponding value is missing!).

One propose I want to make is to

1. consider both NULL and R_MissingArg symbols as invalid values by default, so current users won't be affected
2. store R_MissingArg as "invalid" node values instead of NULL. The hash values of errored nodes will change, but the value is out-dated anyway, so...
3. provide options to treat NULL as valid. This can be enabled by default, or just in shortcut mode.

If you don't like alist(a=)[[1]], I think rlang::sym("") also generates the symbol.