Nim: Obtaining string address results in IndexError

[endragor]

Commit: c12726d41e7df20bce452b6d7f8959d8b9e401a9

This code now produces IndexError, even though I think it shouldn't?

var s = ""
echo unsafeAddr(s[0])

[trustable-code]

From the changelog:

Accessing the binary zero terminator in Nim's native strings is now invalid. Internally a Nim string still has the trailing zero for zero-copy interoperability with cstring. Compile your code with the new switch --laxStrings:on if you need a transition period.

[Araq]

This always was an edge-case, not that this code never worked for seq. This code should be written as cstring(s).

[endragor]

@trustable-code there is no _accessing_ in the sample I provided, so the changelog doesn't explain the behaviour.

I'd expect taking address not to inject any runtime checks, especially when it's unsafeAddr.

[Araq]

I'd expect taking address not to inject any runtime checks, especially when it's unsafeAddr.

I agree and this will be done, but actually it never was possible. For strings it only worked because s[0] used to be no IndexError.

[zah]

Isn't the problem here that we don't offer a separate call returning the starting address of the string? I don't see why the combination unsafeAddr(s[X]) should be blessed when s[X] doesn't denote a legal location. Having a starting address allows you to do any kind of pointer arithmetic yourself.

[krux02]

As Araq mentioned, you should use cstring as the conversion function.