Keyboard: don't auto-toggle CODE/KANA lock while that key is held down - #7

…n thickness setting and manual controls.

…er support

With a <level_2_via> tag. According to the discussion in:
  openMSX#2138

I didn't do _any_ testing yet. I ran out of time and did already want to
push this experiment. I'll continue later, and then ask for help with
adjusting all the machine/extension configs.

Summary: added a new tag <level_2_via> with possible values:
* read   (=default)
* write
* interlocked_write_read
* only_level_1
"read" implies "shared_address=false", "shared_counter=false", the other
modes have "true". Open question, maybe this "read" mode can be dropped?

I only adjusted the machine configs
* Sanyo_PHC-70FD2.xml
* Sony_HB-F1XDJ.xml
* Yamaha_YIS-805-256.xml

Thanks to melovictor:
  https://www.msx.org/forum/msx-talk/openmsx/missing-topic-listing-for-machine-info-in-tcl-console

See again: openMSX#2138

Apparently also 'level_2_by=read' must have a shared address/counter
(because of measurements on 'Sony HB-F1XDJ').

I realized that we can greatly simplify our implementation by always
assuming a shared address/counter. As there are no known non-emulator
implementations anymore without sharing. And if we ever do need an
unshared version again, then we can instantiate the MSXKanji device
twice: once on ports 0xd8/0xd9 and once on ports 0xda/0xdb.

(After IRC discussion)

Reading from the level2 ports when only level1 ROM is present (128kB
ROM, not 256kB) should return 0xff. Before we mistakenly thought it
would mirror level1 content.

Includes the used test program and explanation.

Commit for bsittler. Thank you so much!

From 2026/03/08 to 2026/06/27.

Again a commit for bsittler. Thanks again!

…t it

Background info (at least how I understand it currently):
* Historically (openGL 1 and 2) textures object contain both the image
  data and sampling properties like filtering (nearest neighbour,
  bilinear interpolation, ...) or wrapping mode (tile, clamp to edge,
  ...), ...
* From openGL 3 onwards it became possible to separate the texture data
  (the image itself) from the sampling-parameters. The latter is a then
  a "sampler object".

OpenMSX is mostly still only using openGL2 features, so we still set the
sampling parameters on the texture objects.

Recently Dear ImGui started using openGL sampler objects (we got this
change since 4644985 when we update our Dear ImGui version).
Unfortunately this broke a few things:

For example the "Bitmap Viewer" debug window explicitly uses "nearest
neighbour" filtering to show the zoomed-in image. Before that worked
because we set the filtering mode on our texture (which we then draw
with ImGui::Image()). But the new ImGui version has a sampler object
active which overwrites the filtering mode to 'bilinear interpolation',
resuling in a blurry zoomed image.

Another example is drawing the grid on top of the zoomed image. We did
that via a texture containing a single grid-cell, and then using the
tile-wrapping mode on that texture. Though again because there's now a
sampler active it overrules that mode and uses clamp-to-edge.

This patch (hopefully) temporarily disables the new Dear ImGui behavior
(it forces the openGL2 fallback path where openGL sampler objects didn't
exist yet). Though there's probably a good reason why Dear ImGui want to
use sampler objects (and we should try to avoid patching the Dear ImGui
code .. makes upgrading to newer versions harder). So we should find
better solutions. ATM I have some ideas, but I need more time to work
them out.

)

A Hit Bit U owner has run the test and provided me with screen photos that show this behavior.

Co-authored-by: Benjamin C. Wiley Sittler <bsittler@gmail.com>

Due to incomplete constexpr support for std::string in the current C++ code I need to through some hoops here (thanks to Gemini) to get this constexpr. But I think I made it. If someone has a better idea on how to solve this, please change the code :)

Fixes openMSX#2149.

Tiny improvement on previous commit. More lines, but overall cleaner...

… older versions to reduce confusion.

This version newly incorporates the `BACONLDR.BIN` loader+runtime library for the (MIT-licensed) [MSX-BACON](https://github.com/hra1129/msx_basic_compiler) BASIC compiler, which is used to build a precompiled version of the test suite. This precompiled version loads and runs quite a bit faster on machines with disks and sufficient RAM (64K or more). This precompiled version of the test suite is assembled using [zma](https://github.com/hra1129/zma). To rebuild the precompiled test suite requires an MSX-BACON compiler bugfix from hra1129/msx_basic_compiler#32 to make `INP(`...`)` usable. Prebuilt binaries are included to ease testing on real hardware.

A small, standalone public domain Python CAS-to-WAV converter generated by Google's Gemini search assistant is also included, and was used to generate the WAV version of the cassette test suite from the CAS image.

The cassette version of the test suite excludes the precompiled MSX-BACON version as it would complicate cassette loading.

The X-BASIC / BASIC'n / MSXべーしっ君 `_TURBO` compiler is still recommended when loading from cassette, and for loading from disk in machines with 32KB of Z80 RAM. Machines with only 16KB of Z80 RAM are stuck with slow interpreted MSX-BASIC. 8KB of Z80 RAM isn't enough to run these tests.

This is related to openMSX#2138

Fix KEYPAD_COMMA input bug introduced in d24d37d and add POSITIONAL mode support

Use a dummy value for SCANCODE as it is not used in unit tests.

Changed the mouse speed correction from a fixed 1/2 multiplier to a ratio based on the display size and the MSX screen internal resolution (240).

Thanks uniskie for reporting!
    openMSX#2154

We added backwards compatibility serialization code in MSXKanji, but
then forgot to bump the version numberr, oops.

…penMSX into uniskie-Mouse_speed_adjustment_feature

National FS-4000 I/O port kanji behavior is confirmed by Jun Nishikawa using kanji test v3.1

see openMSX#2138 (comment)

this already happened to be how openMSX emulated this machine, but this makes it explicit and documents the behavior as matching test results

For discussion and background info see:
    openMSX#2152

This commit only changes the implementation of the `getUserHomeDir()`
function (which is used by tilde expansion). It does not yet do a full
audit of all openMSX commands that use filenames. That's for follow-up
patch(es).

See:
    openMSX#2157

In 'fast blink mode' (undocumented bit 2 in R#2) the blink status
changes per line instead of per frame. This was already implemented for
the bitmap modes. This commit also adds support for the TEXT2 mode
(screen 0, width 80).

Thanks to:
  openMSX#2057   (measurement tools by bengalack)
  #5          (analysis done by Claude (Opus), an AI assistant)
See comments in code for details. Read the discussion in the above
links for even more details.