This PR doesn't have any linked issues. Please open an issue that references this PR. From there we can discuss and prioritise.

See nodejs/node#35644 for why we have to do this to avoid potential regressions when users upgrade to Node 16.

I read the associated Node PR for fs.statSync but unless I missed something, it looks like it's an explicit flag that has to be passed in to get the new behavior? Otherwise you get the original (throwing) behavior. So I don't understand why a backport is necessary.

@fatcerberus I don't think that's the case. My understanding is that if you don't pass in the flag, exceptions will be thrown with an arbitrarily large call stack capture. @amcasey understands it way better than I do though.

@fatcerberus, yes, the new flag is the mitigation. Prior to adding the flag, we were using a different mitigation (suggested by node) - we set the call stack depth to zero for the duration of the stat call. Node PR 35644 circumvents this mitigation by resetting the call stack depth to infinity while construction uvExceptions (like the one for non-existent files). More details: nodejs/node#35644 (comment)

The change looks good, but strictly speaking, we need to review all the statSync calls in case 3.9 had some that were dropped by the time I made the change in 4.2 (unlikely).

Edit: confirmed for 3.9, 4.0, and 4.1.

@amcasey Thanks, I was assuming this was a mitigation for a breaking change in Node 16, then I got confused because it didn't seem like throwIfNoEntry was a breaking change (in the Node API). I didn't notice there was a different change causing a perf regression at the time.