There are still a few regressions to address, but keyCode-to-simCode-to-scanCode conversion feels more robust/cleaner, and new features like programmable repeat rates will actually be doable now
HTML controls with bindings are now identified by the single class attribute of "{APPCLASS}-binding" instead of the somewhat arbitrary "{APPCLASS}-input" or "{APPCLASS}-output" classes
The old VisiCalc demo state file (state.json) was triggering a spurious (keyboard?) event that would generate a beep and wipe the copyright message; I'm kind of curious what was causing that, but the simplest solution for now is to generate a new state.json.
If a Markdown document contains an embedded machine that specifies a version number (eg, "1.15.6"), then the web server will switch to serving a compiled version, since that's the only way it can guarantee the requested version
I spent some time fixing the latter because browsers have started warning that synchronous XMLHttpRequest calls on the "main thread" (as Firefox calls it) are deprecated. Since the days of such calls seems to be numbered, I decided to act sooner rather than later. All PCjs XMLHttpRequests go through a single library function (weblib.loadResource), which accepts an fAsync parameter, and the only code that is still allowed to pass false for that parameter are "emergency" shutdown requests (which I believe browsers must still allow).
What this means is that XDF disk images formatted as 80-track 23-sector-per-track images will work with XDF code that honors that format; however, I've not yet been able to test with "real" XDF disk images (ie, disk images created with DiskDump using the experimental --xdf flag). "Real" XDF disk images almost certainly need more work (eg, setting all the sector IDs properly), but the groundwork has been laid.
However, performance will suffer until I can implement RTC interrupts;
also, there may be a problem exposed by the BIOS memory test following a soft reset
