When adding Nullable::operatorT() as a more idiomatic version of the existing as() getter, we encountered six packages among the 3000+ tested which would not compile. With this issue, we will look more closely but it seems that for all these six cases it is due to Nullable() being used in context of a ternary C operator (conf ? yes : no) and the compiler complaining that the 'yes' and 'no' cases had different types.
Presumably this goes back to how the compiler (implictly) casts and having operatorT() opens the door for a compiler error of the "type of Nulllable<someObject> different from someObject" which a simple cast will avoid. (And if one does the more commonly used explicit form of if (obj.IsNotNull()) yes else no it does not even arise.)
So I plan to send (simple) PRs to the six packages, monitor adoption and we should able to remove the currently-used #if defined(...) guard about Nullable::operatorT() that makes it a limited opt-in, and possibly switch to an opt-out reversing the boolean logic. To be seen.
Per the reverse dependency check, the packages are
- gdalraster (and @ctoney is already on top of things)
- ieTest
- localScore
- scellpam
- sorcering
- statgenIBD
When adding
Nullable::operatorT()as a more idiomatic version of the existingas()getter, we encountered six packages among the 3000+ tested which would not compile. With this issue, we will look more closely but it seems that for all these six cases it is due toNullable()being used in context of a ternary C operator(conf ? yes : no)and the compiler complaining that the 'yes' and 'no' cases had different types.Presumably this goes back to how the compiler (implictly) casts and having
operatorT()opens the door for a compiler error of the "type ofNulllable<someObject>different fromsomeObject" which a simple cast will avoid. (And if one does the more commonly used explicit form ofif (obj.IsNotNull()) yes else noit does not even arise.)So I plan to send (simple) PRs to the six packages, monitor adoption and we should able to remove the currently-used
#if defined(...)guard aboutNullable::operatorT()that makes it a limited opt-in, and possibly switch to an opt-out reversing the boolean logic. To be seen.Per the reverse dependency check, the packages are