spectrum4 / spectrum4 Goto Github PK

See warnings in CI:

   552) [0.003s] src/spectrum128k/tests: z80-unknown-elf-as -o test_po_fetch.o test_po_fetch.runner
test_po_fetch.s: Assembler messages:
test_po_fetch.s:44: Warning: 8-bit overflow (-231)
test_po_fetch.s:60: Warning: 8-bit overflow (-163)
test_po_fetch.s:76: Warning: 8-bit overflow (-231)

and later on:

  546) [0.006s] src/spectrum128k/tests: z80-unknown-elf-as -o test_po_store.o test_po_store.runner
test_po_store.s: Assembler messages:
test_po_store.s:26: Warning: 8-bit overflow (-231)
test_po_store.s:48: Warning: 8-bit overflow (-163)
test_po_store.s:71: Warning: 8-bit overflow (-231)

Sometimes fuse tests don't run, and the process hangs. This is typically resolved by rerunning tup. However,

1. The tests shouldn't hang, so we should find out what is going on. I think the hang occurs on startup before the individual tests are run, rather than e.g. midway through a test. Need to validate this hypothesis.
2. There is a rerun mechanism built into fuse-tests.sh that is intended to workaround the issue, yet it doesn't work, for some unknown reason. It implements retries for the test suite, yet if one fails, they all seem to fail. This is curious, since rerunning tup normally fixes it. Perhaps a sleep is needed? In any case, would be better to fix the problem at source.

One reason I've not looked too deeply is that the problem may lie in fuse, tup, bash, or something outside of this code base, and that may be tricky to diagnose.

This is the Spectrum 128k runtests.s equivalent change to 149d2ae (where the same feature was implemented for Spectrum +4 in aarch64).

For example, instead of reporting:

FAIL: po_msg_paper_ink_9_1: RAM page 5 offset 0x1c8f changed from 0x38 to 0x13, but should not have changed.

report something like this:

FAIL: po_msg_paper_ink_9_1: System variable ATTR_T changed from 0x38 to 0x13, but should not have changed.

Note: not all system variables are one byte, so some care is needed to generate appropriate output depending on the storage size of the system variable. Also note, space is tight, so routine and data should be as compact as possible.