Fix crash in `(scheme bytevector)`, and adjust the procedures to the spec.

jpellegrini commented 1 year ago

I've found two bugs in the (scheme bytevector) library.

How the bug was found -- `test/error` was wrong!

I have noticed that the test/error macro in tests/test.stk doesn't actually do anything: it ignores errors, but doesn't complain when no error is produced! So I fixed it (will include in a later PR), and... Found several problems.

This is one of them -- now the test suite complains that we didn't error when the wrong index (non-aligned) was used.

Later I'll send a PR with a new test macro, and several fixes in SRFIs, in the core, and in tests.

1. Check arguments in `bytevector-ieee-*-set!`

Otherwise STklos crashes!

Also, convert all numbers to imprecise (except complexes) before storing them in the bytevector (only complexes can't be turned into reals -- for bignums, if they don't fit a double, we store an infinity; for rationals, if they're too small, they become zero.)

2. Check indices in some of the IEEE bytevector procedures

The spec (the only part of $R^7RS$ Large that was taken from $R^6RS$) says we need to access with proper alignment (multiples of 4 for single precision, and multiples of 8 for double precision). Fixed.

egallesio commented 1 year ago

I have noticed that the test/error macro in tests/test.stk doesn't actually do anything

Funny. This is a way to not have problem :wink:

egallesio commented 1 year ago

Merged. Thanks.

egallesio / STklos