Haiku activity report - February and March 2020

Blog post by pulkomandy on Sun, 2020-03-22 10:14

Hello!

The previous report involved a lot of travelling around and attending various events and conferences. This month is quite different as some of us are locked home due to the ongoing pandemic. We already know some of the next planned events such as the JDLL and FLISOL are cancelled (for good reasons).

Anyway, the activity on Haiku has not slowed down, so let's see what's happening there. This report covers hrev53875-hrev53995.

Unit tests

Kyle Ambroff-Kao continues the work on fixing and improving our unit tests. The focus is currently on tests for the network kit, with a rework of the HTTP tests to not rely on external servers, and instead run a local one using Python.

Media and Audio

Korli fixed some issues in the media kit related to buffers ownership, making things more reliable and fixing the crashes on shutdown for good.

He also updated the HDA driver to handle some “quirks” (that is, workarounds for hardware that does not fully follow the HDA specification), extracting the needed information from the ALSA Linux driver.

Bootloader

Kallisti5 worked on various bootloader things, fixing the generation of the PowerPC bootable CD (the PowerPC still is not in an usable state, but at least we can test it, now), added a volume icon to the EFI system partition so it looks nice in Apple machines boot menu, and reworked the EFI code to not be specific to Intel CPUs. The goal here is to reuse the EFI code also for ARM64 and possibly ARM, since U-Boot and other firmwares are standardizing on providing the EFI API there.

We had problems getting the ARM port to run because U-Boot is a very minimal bootloader, it only runs our executable and lets us figure out the hardware. This means the bootloader would need to handle many platform-specific things, such as UART drivers, framebuffer, etc. EFI on the other hand provides us with a way to call back into the firmware and get it to do all that hard work for us. It makes it significantly easier to get the bootloader up and running, and pushes back the hardware-specific problems to the kernel (which can't call the firmware after it starts handling the hardware by itself).

Kallisti5 also worked a bit on Risc-V support, but that will probably be seriously worked on only after we get ARM64 running.

PulkoMandy received new SPARC hardware donations which led to various fixes to the SPARC and OpenFirmware boot code. The openfirmware bootloader now manages to find the network disk server, mounts the BFS and PackageFS inside it, and starts to load the kernel. It should even be possible to attempt booting from disk, as the openfirmware allows our bootloader to access disks without needing custom drivers. This work also helps reviewing code common with other architectures and led to at least one fix for the EFI bootloader as well.

Kernel and drivers

Korli implemented getting the current CPU number from the rdtscp instruction on x86_64 machines.

waddlesplash improved the handling of the KDL shortcut to make it work on PS/2 keyboards that don't properly implement the SysRq key (which has very special handling in PS/2) and instead report a normal Print Screen press.

waddlesplash continues his work on XHCI and USB stack to improve compatibility and fix bugs. He also improved the NVMe driver to use interrupts instead of polling.

PulkoMandy continues working on the Intel grpahics driver, with some fixes for SandyBridge cards, which should eventually lead to dual screen support (but we are not quite there yet).

Korli improved the ACPI battery driver to handle “_BIX”, which is a different way to get the battery information. He also is working on loading microcode updates at boot for Intel CPUs, but not all of the changes have been merged yt to make this available.

Korli also implemented getting the MTU from virtio network interfaces.

User interface

X512 reworked the window focus handling in Tracker, fixed Haiku3D to render its textures again; and fixed movement of the teapot in GLTeapot, as well as fixed drawing glitches in replicants and flickering issues in various other places, making some controls behave more closely to the BeOS version in small (but nonetheless important) details.

X512 also reworked BMenu code to simplify it, and fixed some longstanding bugs, for example with the “navigation area” handling which allows one to reach a submenu without accidentally selecting other menu items. Eventually this allowed to remove a now useless delay in opening menus, making them feel faster.

John Scipione reworked the Deskbar to fix various glitches and issues, reworked the auto-raise and auto-lower code to be less annoying, and finally adding a new mode for DeskBar where it shows as much information as possible while using only a single “line”, that is, fitting in the unused space at the right of window tabs.

X512 also worked on Deskbar, adding new shortcuts: middle-click on an application in the task list will open a new instance of it, and shift + middle click will terminate the application.

There is ongoing work on menus from both X512 and John to allow better handling of dynamically generated menus, improve scrolling when menus are too large to fit on screen.

Preetpal Kaur fixed the layout of the default and revert buttons in Input preferences to be consistent between all devices, and made the whole window better handle different font sizes.

runtime_loader and application roster

waddlesplash reworked the termination functions handling. This is code in the runtime_loader that automatically calls destructors for statically allocated objects, and similar functions. There is a very specific order in which these functions must be called, and the way they are stored in the executables and libraries changes with different gcc versions, so we have to implement this carefully. He also fixed a memory leak.

PulkoMandy fixed a regression in the runtime_loader that prevented running single-instance apps from a symlink. A second instance would be started instead.

Applications

The Installer finally handles “optional packages” again. This functionality had been lost with the introduction of hpkg-based packages, and it was a problem because having lots of packages mounted prevents Installer to succesfully install the system on low-memory systems (less than 256MB of RAM). This makes it possible to move the source packages and some development packages out of the live DVD/USB, but still have them available for installation.

CodeForEvolution added new scripting commands to MediaPlayer to allow retrieving the current track number in the playlist.

PVS and Coverity scans

A new Coverity scan has been done. MT investigated some of the problems found by either Coverity or PVS and fixed issues like missing NULL pointer checks, memory leaks and allocator corruptions.

Documentation

Nielx updated our API documentation to the latest Doxygen version and fixed some markup issues that lead to broken rendering of affected pages.

He then worked on documentation for BNotification, BKeyStore, and the old layout builder classes.

app_server

Pascal Abresch fixed the font fallback system, which had been broken after switching to Noto as the default font, because Noto is actually a set of multiple fonts, making the code a bit more complex.

waddlesplash reviewed and simplified the BRegion code.

Filesystems

X512 added support in BFS for deleting corrupt files. It would otherwise not be possible to do anything with them, and they would stay there forever.

Google Summer of Code students are also already starting to contribute, and this month the first few lines of code for XFS filesystem implementation were merged.

Cleanup

Not every code cleanup can be mentionned, but let's note the removal of “atomizer” (an example kernel module that had been imported from a Be Newsletter article and never used for anything), removal of deprecated C++ features (auto_ptr) and multiple implementations of hash maps (standardizing on a single one instead), all done by waddlesplash.

Kyle Ambroff-Kao removed the “makeudfimage” tool which was not used in Haiku, and for which 3rd party replacements are already available.