Be explicit about the type and bound we're checking in BytecodeReader.nextInt32.

[allevato]

Thanks to a heterogeneous comparison operator on BinaryInteger, the expression someUInt64 < 1 &<< 32 was compiling incorrectly on 32-bit architectures, because the right-hand side was being inferred as Int instead of UInt64. And on a 32-bit system, 1 &<< 32 is zero.

Fixes #1858.

[thomasvl]

Aside - there's no public api addition here, but I do wonder if this should be semver/minor due to the bug in question.

[allevato]

Aside - there's no public api addition here, but I do wonder if this should be semver/minor due to the bug in question.

The bytecode was introduced in 1.31.0, so if we release this fix as the next patch release (1.31.2), then users who are using upToNextMinor in their SwiftPM dependencies would get the fix automatically. Likewise, if we released it as 1.32.0, then those users wouldn't get the fix without manually updating, which would be undesirable.

[thomasvl]

Aside - there's no public api addition here, but I do wonder if this should be semver/minor due to the bug in question.

The bytecode was introduced in 1.31.0, so if we release this fix as the next patch release (1.31.2), then users who are using upToNextMinor in their SwiftPM dependencies would get the fix automatically. Likewise, if we released it as 1.32.0, then those users wouldn't get the fix without manually updating, which would be undesirable.

And I guess the bug only happens to folks using TextFormat or JSON on watchOS. Binary on watchOS is fine, and TextFormat/JSON on everything else is 64bit and won't run into this problem.

[FranzBusch]

I think this is fine as a patch release since this looks like a relatively straightforward bug-fix.