4.2KB

Raw Blame 履歴

story: E02-S01 owner: e02_source_reader status: done started: 2026-11-07 depends-on: [E01-S03, E01-S05]

E02-S01 Claim

Intended change

Add the compiler's source reader. src_load reads a PETSCII SEQ file from device 8 once into a bank-1 cache, enforcing the line, line-count, and file size limits with stable error codes and exact positions. src_rewind and src_next_line then replay the cached source identically for every pass, one logical line at a time. Disk access goes through new diskio.s KERNAL wrappers that always close their channels.

File area

src/compiler/source.s, src/compiler/diskio.s
src/common/source.inc
src/compiler/mmu.s (bank-1 block copy only)
cfg/compiler.cfg (bank-1 source cache area)
tests/platform/map-skeleton.s (table-size constant only)
tests/source/
scripts/build-source-test.ps1, scripts/verify-source-test.ps1, scripts/lib/
docs/memory-map.md, docs/architecture.md, docs/decisions.md (D118)
planning/claims/E02-S01.md
generated files under build/

Acceptance evidence

Reads Vim-created SEQ source without host conversion: fixtures use Vim's exact save format (PETSCII, $0D lines, both letter ranges in decls.n8); the reader delivers the bytes unchanged, verified by a host Fletcher-16 match. A live Vim edit is E07-S03.
Accepts $0D lines and a missing final $0D: nofinal.n8 yields the same 50 lines and checksum as decls.n8 with one byte fewer.
Enforces line/file limits with stable errors: edge.n8 (80-byte line, 599 lines, 18,999 bytes) loads; longln.n8 -> code 3 at 3:81, manyln.n8 -> code 4 at 600:1, big.n8 -> code 2 at 500:38.
Rewinds identically for all compiler passes: two replays per load give identical line counts and checksums, and src_line_no ends at the line count.
Disk failure closes channels and leaves no partial output: LDTND is 0 after every load; after each failure src_len is 0 and src_next_line returns end of source at once; a later decls.n8 load succeeds unchanged.

Coordination notes

Source error codes are reader return codes. E02-S04 maps them to stable Exxx diagnostics. A true Vim round trip (edit in x64sc, compile in x128) is E07-S03; this story proves the reader consumes Vim's exact byte format (../vim/NOTES.md, vedit.c save path) with no conversion.

Work log

Confirmed Vim's save format in ../vim/NOTES.md and vedit.c (raw buffer, $0D lines, final $0D appended, 80-column wrap on load).
Recorded D118: one disk read into a bank-1 cache ($2000-$6AFF), tables keep $6B00-$BFFF (21,760 bytes). Map skeleton re-verified with it.
Added mmu_copy_to1 and mmu_src to mmu.s, diskio.s with a COMMON chunk reader run through kcall, source.s, and diskio.inc.
Added scripts/lib/Nova128Test.psm1 (link, PETSCII conversion, disk images, remote monitor, mailbox polling). The monitor helper waits for the (C:$xxxx) prompt; a fixed delay returned empty replies with a 1581.
Negative control: allowing 81-byte lines makes the C128 report FAIL and the host flag longln.n8 (code 0, position 0:0, 100 bytes kept). Restored.

Completion

Tests run: scripts/verify-source-test.ps1 (pass plus mutation run), scripts/verify-map-skeleton.ps1, scripts/verify-kernal-probe.ps1, tests/grammar/validate.ps1, tests/disk/prepare-disk.tests.ps1.
Results: all 8 fixture records match host expectations; screenshot shows each result code and NOVA128 SOURCE TEST OK; regressions pass.
Files changed: src/compiler/source.s, src/compiler/diskio.s, src/compiler/mmu.s, src/common/source.inc, src/common/diskio.inc, cfg/compiler.cfg, tests/source/source-test.s, tests/platform/map-skeleton.s, scripts/build-source-test.ps1, scripts/verify-source-test.ps1, scripts/build-map-skeleton.ps1, scripts/lib/Nova128Test.psm1, docs/memory-map.md, docs/architecture.md, docs/decisions.md (D118), this claim.
Remaining risks: device 8 is fixed; a read error part-way through a file (ST error other than EOI) is handled in code but not exercised, since VICE offers no simple way to inject one; only the 1581 (D81) was tested; error codes are reader codes until E02-S04 assigns user-facing Exxx IDs.

4.2KB Raw Blame 履歴