Haiku monthly activity report - 12/2017

Blog post by PulkoMandy on Sun, 2017-12-31 17:26

Welcome to the last report for the year 2017!

Stats

Who doesn't like them? I updated the Haiku stats to keep track of the activity in our git repository. The overall number of commits is very similar to 2016 (which was our quietest year so far) with more than 1300 commits (far from the 5555 commits in 2009). Our author of the year is waddlesplash with 213 commitsi, followed by PulkoMandy, Korli, Humdinger, Kallisti5, and Skipp_OSX. 65 different commiters made changes to Haiku this year, a net increase from 48 in 2016, but not reaching as high as 2012 (83 different committers). In the week-by-week graph you can also clearly see the effect of the coding sprint, which is of course the week with most commits.

Anyway, let's dig into this month changes to see what's cooking for 2018. This report covers hrev51623-hrev51722 (exactly 100 hrevs!)

User interface

Lots of activity in this area, as polishing the UI is what makes the system look smooth and finished. A subtle but important change was implemented by Skipp_OSX, so that maximized windows will now avoid the deskbar. While maximizing windows is not recommended (they should pick a size appropriate for their contents instead), sometimes it does make sense to use as much screen space as possible.

Jua continued his work on improving HaikuDepot, with a better look for the screenshot window (work he had started during the coding sprint).

Andrew Lindesay also worked on the network side of HaikuDepot, improving the loading of data from the haiku depot server.

The work on HaikuDepot spawned debate on the "barber pole" control, to use as a progress bar when the total progress is unknown. Jua started designing a new control for that, and Janus reworked the look and integrated it in places where a barber pole was already used, such as Zip-O-Matic.

Janus also reworked the "key store" window (which you can see when saving a wifi password to the keystore, for example). It now looks nicer with an icon and other minor details.

Janus fixed a crash of StyledEdit when the search/replace window was open and the main window was closed.

Owen (one of our GCI students) made MediaPlayer accept URLs from the command line (not just from the GUI). He also worked on improving scripting support in MediaPlayer and Magnify, making it possible to control them using the "hey" command line tool, for example.

Hrishi Hiraskar (another new contributor) made it possible to move the notification window to different places. It does not need to be near the deskbar anymore.

Legal

Kallisti5 started a survey of licences used in Haiku. While most of our code is MIT, some parts have a different history and for various reasons ended up in our sources with different licences.

In the case of BFS recovery tools, they were integrated without a clear licence mention. Axel confirmed that using MIT was fine, so this is now clearly indicated in the files.

In the case of MediaPlayer, there was actually a different licence used. However Kallisti5 contacted all authors who had edited the files since integration in our source tree and they all agreed to relicence it as MIT.

Package repositories and build infrastructure

As mentioned in a separate news item, the package repositories are now generated automatically. This means as soon as a recipe is added to the haikuports repository, buildbots will make a package from it and upload it there, making it available for everyone. This considerably reduces the workload in maintaining the repositories, so Haiku developers can focus on more important tasks.

This led to a cleanup of the repositories as well as 3rd-party sources included in the Haiku tree, making our codebase easier to manage as well.

Waddlesplash also cleaned up support for building Haiku with clang. While the resulting images are currently unbootable, this allowed him to streamline support for scanning our code at Coverity, and a new scan was finally made available, in time for Google Code-In students to have a look at it.

mmlr did wome work on restoring support for "bootstrap" builds. These make it possible to build Haiku witohut any binary downloads (normal builds rely on pre-built packages for all 3rd party sources).

Waddlesplash cleaned up our configure script, made it autodetect some settings (like host support for extended attributes), and made building on Haiku more similar to cross-compiling, allowing simplification on the most ugly parts of our build system.

Servers

Axel added a test_launch_daemon, to help debugging launch_daemon problems. Like the test_app_server, it allows to run the launch_daemon code in a somewhat sandboxed environment, without interferring too much with the running system.

DeadYak worked on Debugger, with the ability to debug the runtime_loader, and review of the code to track various memory leaks.

Standard library

As GCI is running and students are attempting to port all kind of software to Haiku, some problems in our API were discovered.

Leorize added support for some secure BSD functions (explicit_bzero, readpassphrase, etc) as a first step to porting LibreSSL. This is now pending a proper implementation of arc4random (which needs a syscall and kernel-side entropy sources).

ohnx fixed a crash in calloc when called with strange parameters (such as negative sizes and item counts)

Drivers and kernel

Greg Cain continued his work on the USB3 stack, fixing some more problems with it. While still not perfect, USB3 support is good enough for me to use my mouse on an USB3 port (I ran into problems with usb to serial adapters last time I tried, however).

Ayush contributed some more fixes to the TCP stack after his work done during GSoC.

PulkoMandy fixed a corner case problem with stack alignments. The stack was aligned only when entering the main() function, so use of sse2 functions in early init code (static initializers) or late deinit code (static destructors) could fail. This had already been fixed for 64bit architectures, but x86 32bit had never been updated.

Hy Che's patches to btrfs (done this summer during GSoC) were merged. The btrfs write support is still incomplete and disabled, but someone else can pick up the work from there, with a lot of the basic support already in place.

waddlesplash fixed some locking problems in the management of teams and their attached resources.

tqh fixed the UEFI build (still not included in anyboot images), fixed some configuration issues in our ACPI port (some changes were overwritten last time we updated from upstream) and he is now investigating some lockup on shutdown problems.