[affan-arch]

with autoconvert on, __config_setting_get_int(), __config_setting_get_int64() and config_setting_set_float() cast a double straight to int or long long, so a config value like x = 1e300; read through config_lookup_int() is undefined behavior (ubsan: 1e+300 is outside the range of representable values of type 'int') and the lookup reports success with a saturated or indefinite value; they now check the double fits the target type first and return CONFIG_FALSE otherwise, the same way the int64-to-int path already does, and the new AutoConvertFloatToInteger test fails without the fix.

[hyperrealm]

Instead of hardcoded constant 2147483648.0, why not (double)INT_MAX + 1.0

Similarly for the other constants.

[affan-arch]

done, the int bounds are (double)INT_MIN and (double)INT_MAX + 1.0 now. for the int64 upper bound i used -(double)LLONG_MIN instead of (double)LLONG_MAX + 1.0: LLONG_MAX already rounds up to 2^63 as a double, so the + 1.0 is a no-op there, and on x87 with excess precision the sum can come out as 2^63 + 1, which would let 2^63 itself through the check. can switch it to the symmetric form if you prefer.

[hyperrealm]

looks good, thanks!

[hyperrealm]

please rebase and then I'll merge.

[affan-arch]

rebased on master. the only conflict was tests/tests.c, where WriteLargeFloat from #307 landed in the same spot, so both tests are kept now and the suite passes.