[DNSGeek]

Summary

Seven confirmed bugs from the recent code review, one commit each. The C64 build is behaviour-identical apart from the fixes; the C128 editor image still links below $4000.

Editor (data loss)

• do_new_file / quit_dirty_prompt: a cancelled or failed save no longer clears the buffer or quits. do_save_file clears IS_DIRTY only after a successful write, so a still-dirty buffer is treated as cancel.
• F3/F6 load now prompts "SAVE BEFORE LOAD (Y/N)?" when there are unsaved changes, as README and the editor manual already promised. The Y/N/STOP wait shared by new, load and quit is factored into one dirty_prompt routine.
• run_module_by_index sets IS_DIRTY on status $02 (buffer replaced, e.g. Renumber) and $04 (gap repositioned, e.g. Search/Replace), so module edits trigger the save prompts.

MODREN (silent truncation)

• Pass 2 walks the linearised staging text, but advance_src still warped SRC_PTR across the editor's old gap. With pre-gap text larger than the gap, gap_size bytes of program were dropped mid-line. After setup_staging, REN_GAP is pointed at REN_GAP_END so the warp is a no-op.

I/O (session wedge)

• loadsave.asm load and save, and modasm.asm open_output: a failed OPEN now goes through CLRCHN + CLOSE. The Kernal registers the LFN before the serial open can fail, so one DEVICE NOT PRESENT used to leave the file open and every later load/save/assemble failed with FILE OPEN. open_output also records I/O ERROR, which its caller already assumed.

MODASM (miscompile)

• Relative-branch range check now also tests the lo byte's sign, so +128..+255 and -129..-256 report BRANCH RANGE instead of emitting a wrapped offset.

C128

• mpop_show_error / asm_show_err read MODASM's error block via MOD_LO_BASE + $07..$0A instead of literal $C007..$C00A. On the C128 that state is at $9000 and $C000 is ROM, so assembler errors degraded to the generic message. Unchanged on the C64.

Verification

• bash make_petproject.sh: both targets build and link.
• python3 tests/test_include.py: all 15 checks pass, including the new branch-range regression test (BEQ/BNE over 130 NOPs must fail, +120 / -128 must assemble). The unfixed assembler accepted the +130 case and emitted offset $82.
• MODREN fix reproduced on py65 with a nearly full buffer whose old gap start falls inside a line number: unfixed output 10 A / 20 B / 340 D ..., fixed output intact.
• C64 .prg size deltas: editor +35, modasm +11, modren +12 bytes; all other modules byte-identical.

Noticed, not changed

• MODSFR returns $04 for a plain find as well as a replace, so a find-only now also marks the buffer dirty (a spurious save prompt, no data loss). A distinct "edited" status from MODSFR would fix it.
• open_output's no-filename exit still sets no error message.
• A cancelled or failed save from any prompt leaves the gap compacted at the end of the text (cursor moves to end), as before.

🤖 Generated with Claude Code