Testing Guide for 4.6-RELEASE
Goals
As part of our on-going effort to improve the release engineering process, we have identified several areas that need significant quality assurance testing during the release candidate phase. Below, we've listed the changes in 4.6-PRERELEASE that we feel merit the most attention due to their involving substantial changes to the system, or having arrived late in the development cycle leading up to the release. In general, our goal in the QA process is to attempt to check a number of things:
- The system has not regressed with respects to stability, correctness,
interoperability, or performance of features present in prior
releases.
- New features result in the desired improvement in stability, correctness, interoperability, or performance.
To effectively determine this, it's desirable to test the system in a diverse set of environments, applying a wide set of workloads, forcing the system to operate both within and outside its normal specification. Particular focus should often be placed on the continuing (or new) capability of the system to perform correctly when used in concert with systems from other vendors.
Features to explore carefully:
- DHCP client support. This release will feature a major update of the ISC DHCP client (v3.0.1 RC8). The common cases have been well tested, but testing in more demanding or unusual environments would be beneficial. In particular, the new dynamic DNS update functionality has not been well tested.
- sppp(4) upgrade. The ISDN project (i4b) once maintained an off-spring version of the SyncPPP driver, with a number of enhancements and bug-fixes made to it by people contributing to the i4b project. On the other hand, other improvements and bug-fixes of the regular tree's sppp driver didn't make it back to that version, so eventually two different SyncPPP drivers developed. The i4b version has been merged back in FreeBSD-current a few months ago, and these changes have now been folded into FreeBSD-stable as well, thus eliminating i4b's separate version. sppp users (of both flavours) should carefully test the new version. Users who previously used the i4b version should find all those improvements still being present (most notably, VJ header compression). Users who use sppp on hardware (HDLC) devices (like ar(4) or sr(4)) should see no breakage, but might notice that the negotiation of VJ header compression is now enabled by default. Use spppcontrol(8) to modify the default behaviour. Please report any breakage or oddity you observe to <[email protected]>.
- XFree86 4.2.0. sysinstall now installs XFree86 4.2.0 instead of XFree86 3.3.6; the XFree86 installation now uses ``normal'' binary packages instead of the special tarballs in past releases. The ports infrastructure now uses the XFree86 version 4 as the default version for satisfying dependencies.
The release notes will always be a good place to look for things to test.
Known Issues
- Due to a buffer truncation in sysinstall, the GNOME meta-port (x11/gnome) in 4.6-RC1 does not install correctly. This has been fixed and is functional for 4.6-RC2 (and any later release candidate snapshots).
- nslookup(1) had some bugs in 4.6-RC1 and 4.6-RC2. These have been been corrected and should function correctly beginning with 4.6-RC3.
- Some users have reported stability issues with tagged queueing and the ata(4) driver. These problems have only been observed when using tags on motherboard-based ATA channels.
- Potential problems with the TCP default window size (see kern/34801). The default window size has been reduced from 65536 to 57344 for 4.6-RELEASE. (Errata item needed.)
- Some of the XFree86 configuration utilities write their generated configuration files in places where sysinstall isn't looking for them, thus complicating XFree86 installs that are initiated from within sysinstall. These have been changed to /etc/X11/XF86Config for 4.6-RELEASE. (Errata item needed.)
- Some filenames in the ports tree are too long, thus resulting in odd /usr/@LongLink files after a sysinstall. These filenames have been shortened for 4.6-RELEASE.
- Under some circumstances, dhclient can go into an infinite loop.