Throw ValueError when 1 is provided to the second argument of log()

[alexandre-daubois]

This helps with early error detection and explicit intent, instead of having a potential NaN propagating.

[bukka]

Seems like a BC break to me so this might need at least some discussion. Not sure if NaN is worse than exception.

[bukka]

I would split it to two independent messages if we decide to do it.

[alexandre-daubois]

I understand, I would also be fine throwing a deprecation for now

[alexandre-daubois]

I removed this proposition from https://wiki.php.net/rfc/deprecations_php_8_6, rebased and added an UPGRADGING entry.

[Girgias]

Not sure if this falls into the same category. Returning NAN is technically standard behaviour in some other PLs. Might be a good idea to see what other PLs do with their log function, or just keep in on 8.6 deprecation RFC.

[alexandre-daubois]

Alright, let's wait for new inputs and I'll add it back to the RFC if there's no news in the coming weeks

[TimWolla]

instead of having a potential NaN propagating.

NaN explicitly exists so that one does not need to check each intermediate value. Specifically with floats experiencing implicit rounding, it's not always easy or cheap to avoid specific "malformed" inputs in a chain of operations (vs some class of inputs, such as "all negative numbers" that are reasonable to check for).

Trying to diverge from principle goes against the intentional design of IEEE 754 and I would suggest this not be done (and not even put to vote).