Release plan/Lessons learned
Jump to navigation
Jump to search
This is a list of lessons learned from the previous releases, things that turned out well and should be kept for the next release as well as thing that didn't turn out so well and should be changed for future releases. Ideally, the release plan should be updated and augmented so that the lessons learned are incorporated accordingly.
3.40
Scenery not loading
|
it just says "Loading Scenery" and doesnt load.
— JackT44 (Mon Mar 09). Why is my Flightgear still saying Loading Scenery?.
(powered by Instant-Cquotes) |
|
Selecting Parking Positions
|
|
not able to choose under menue Select an Airport /Aircraft Position /Parking an exist Parkposition. |
|
Qt5/Mac launcher debate
|
|
Related to Qt, I can say that up to now - as far as I can tell - only a simple replacement for the Mac Launcher has been introduced. This is not related to the internal user interface of FlightGear, at least for now. If, how and when the internal gui can leverage Qt is under investigation and in such an early state that those responsible preferred to not announce anything yet. If this ever happens, this will be discussed on the mailing list
— Torsten (Sun Jan 11). Re: FlightGear GUI hell: PUI, Canvas GUI, Mongoose, .
(powered by Instant-Cquotes) |
|
Mid-term Project Goals ?
|
|
Has FlightGear got a list of goals for 5-10 years? How about making or adding to it?
|
|
|
|
One of my observations as an observer with just enough knowledge to understand the situation, from a distance, is that it would be extremely helpful for this project to come to some kind of a consensuses. This free-for-all is confusing and counter productive. Even if a few, or many, core developers know what is going on and are in lockstep, it is very disheartening to contemplate helping or contributing to anything "major" as you have the appearance of no consensuses.
I see advertisements for needed help everywhere I look, but many are not willing to spend significant time on anything that may be in vain. |
|
FlightGear on 32 bit Windows
|
|
The point I am getting at is we have a ton of 32 bit windows users trying to download and run FlightGear. Out of the box with the startup defaults that pretty much doesn't work right now. This is leading to a ton of bug reports, a ton of frustrated new users, and a general perception that FlightGear is a POS that isn't worth messing with.
— Curtis Olson (2015-02-05). Re: [Flightgear-devel] FlightGear on low (<=4GB) memory systems.
(powered by Instant-Cquotes) |
|
|
|
In FG 3.4 (full screen mode) I wondered what happens when I try to set e.g. the Screenshot Directory ?
the new and nice designed Fileselection box is just not displayed, because it is hidden below the full screened FG main window ... |
|
No LOD for aircraft previews in fgrun causing lag
|
|
Models (and liveries) take some time to load, if the model (livery,...) is very detailed, that time may be substantial. That's all there is to it. Just wait it out. (When I ask the launcher to get me a preview of the Vostok for instance, I get a 5 second freeze. Given the vertex count of the beast, I'm not surprised.)
(In principle, we could load a simplified 3d model in the launcher, but I doubt it would be worth the trouble asking people to provide that). |
|
Broken FGCom cleanup triggers crashrpt during process termination
|
|
Shut off FGcom, no more need to report constantly.
— clrCoda (Mon Feb 16). SOLVED: Re: FG3.2 wants to always send error report.
(powered by Instant-Cquotes) |
|
repare_ground_cache(): scenery_available returns false
|
|
A fortunate coincidence, I have just experienced this, too and was able to gather some logging. — Torsten Dreyer (2015-03-06). Re: [Flightgear-devel] prepare_ground_cache(): scenery_available
returns false.
(powered by Instant-Cquotes) |
|
|
|
at least on my 10.7 build, I’m seeing an issue with the nav-cache not initialising. This is actually good, because it gives me a way, finally, to trace down the ‘nav-cache never initialises’ bug that some folks have reported. But it will take a little time. Since the same binary works perfectly on my 10.10 box, I assume it’s something /very/ subtle, probably uninitialised memory or some tiny change in zlib’s gzread functions (since we use the system Zlib library which hence could be different between 10.7 and 10.10).
— James Turner (2015-02-18). Re: [Flightgear-devel] Release 3.4.0 is coming.
(powered by Instant-Cquotes) |
|
|
|
Is this a fresh "clean" install or are you installing it after 3.2? If after 3.2 try renaming |
|
|
|
there's currently a bug in FG 3.2 (under investigation) that makes the simulator crash when you upgrade from FG 3.0 to 3.2, and it seems FG 3.2 crashes even before writing to the log.
|
|
Instant Crashes
|
|
FlightGear 3.4.0 Fails after initializing subsystems and then I received a crash report with other data. — matt1717don (Thu Feb 19). FlightGear 3.4.0 Fails after initializing subsystems.
(powered by Instant-Cquotes) |
|
|
|
It gets to "Loading Navigation (something)" then crashes. As opposed to, essentially, flashing up for a split second like it does with Terrasync enabled.
|
|
Route Manager Crash
|
|
The only problem is that the route manager crash every time I want to open it. Thats the same problem I had with FG 3.2
|
|
Multithreading Issue
|
|
Looks like a multi threading issue to me. When I run FlightGear with multiple screens and multiple cameras, the highes threading mode for OpenSceneGraph that works reliable is CullDrawThreadPerContext while the CullThreadPerCameraDrawThreadPerContext quickly leads to a crash.
— Torsten (Sun Dec 28). Re: Error when trying to run FlightGear with three monitors.
(powered by Instant-Cquotes) |
|
Intel GPU Support
|
|
Has any dev tried to disable their nvidia card and use their built-in intel GPU for once?
|
|
More widespread testing
|
|
We could be using our most talented, experience sim pilots to produce more of the necessary tests during release candidates rather than after release.
|
|
|
|
ideally, people doing such flights regularly would provide a corresponding "fgtape" file using a standard aircraft and get this committed to $FG_ROOT into some kind of "Tests" directory containing pre-recorded flights for doing regression testing in an automated fashion. The thing to keep in mind here is that those among us able (and willing) to troubleshoot such issues are most unlikely to ever actually use FlightGear like this, which certainly includes Zakalawe. Thorsten recently also mentioned that he'll typically just start up FlightGear a few dozen of times using a simple aircraft to test things, without really doing any "real" flying at all. So there are certainly a number of use-cases that aren't getting much attention unfortunately. Which is why it would be a good idea for people to provide flight recorder tapes covering use-cases they care about, and get those committed to fgroot (fgdata) - so that we can grow a library of useful flights to do regression testing - as long as those flights are using standard aircraft and features, they should also be useful for people entirely new to FlightGear. |
|
legacy GUI Z-ordering issue (PUI)
|
|
if i have the radio dialogue (F12) and the map dialogue (CTRL-M) open with one overlapping the other, when i release my left mouse button on the top dialogue in a position over a portion of the lower dialogue, the lower one will pop to the top
|
|
|
|
Dialog z-ordering (fighting?) is my biggest annoyance with the current GUI
|
|
ALS impact on PUI styles for ATI/AMD cards
|
|
There are a lot topics concerning AMD graphic cards with flickering problems, but that seems not to match with the issue i have. Maybe there is someone who has or had the same problem and could solve it.
|
|
|
|
When I use either of Rembrandt or Atmospheric Light Scattering, some of the text in the menus disappear, with various bits of text disappearing then reappearing every second
|
|
|
|
n the HUD (windows, menus), labels tend to be clipped. You can see this, on the previous image, in the rendering options window. And that's more visible when I enable the atmospheric light scattering — Saga (Tue Mar 04). Flightgear doesn't use 100% of the workload and Fonts bug.
(powered by Instant-Cquotes) |
|
|
|
I've succeed to partially circumvent the problem (not solved it). Menu, windows are displayed well unlike labels directly displayed on the screen like "Airspeed exceed vne". You must change the style of the gui in data/preferences.xml
— Saga (Fri May 02). Re: Flightgear doesn't use 100% of the workload and GUI clip.
(powered by Instant-Cquotes) |
|
|
|
JUST CHANGE THE 1 to a 2 in
|
|
|
|
the fact to just change the font seems to solve the bug (just for some labels). I've spotted a font which don't provoke it (lucida.txf)
— Saga (Sun Jun 15). Re: Flightgear doesn't use 100% of the workload and GUI clip.
(powered by Instant-Cquotes) |
|
Release Schedule
|
|
I cannot guarantee for this to be the case in the future, nor is it even likely to happen again. Thus, if the release cycles of Ubuntu and FlightGear stays the same, I fear the next newest FlightGear won't make it into the newest Ubuntu release.
— Markus Wanner (2015-02-18). Re: [Flightgear-devel] Release 3.4.0 is coming.
(powered by Instant-Cquotes) |
|
Updating "Catalog Metadata" version, vars and data paths
Something that appears to have fallen through the cracks is the Experimental Aircraft Center Catalog creation and version update. Need to implement some automatic feature to handle these type of routine version updates.
Windows: Installer Issues
|
|
Today I installed FlightGear on a 32 bit windows laptop that had an existing v3.2 install already loaded. I ran into several issues that I — Curtis Olson (2015-02-04). [Flightgear-devel] FlightGear 3.4 RC1 (Windows) feedback.
(powered by Instant-Cquotes) |
|
|
|
The installer came up and wanted to install FlightGear-3.4.0 into FlightGear-3.2.0's folder. I had to manually switch this to — Curtis Olson (2015-02-04). [Flightgear-devel] FlightGear 3.4 RC1 (Windows) feedback.
(powered by Instant-Cquotes) |
|
|
|
If I don't choose any aircraft at all, the default comes up as the C172 with a 2D panel. Again is that really what we want? Is it time to — Curtis Olson (2015-02-04). [Flightgear-devel] FlightGear 3.4 RC1 (Windows) feedback.
(powered by Instant-Cquotes) |
|
|
|
Again, without selecting any airport, the launcher chose 17CL for me which is some random little bitty strip I don't even know where. Wouldn't — Curtis Olson (2015-02-04). [Flightgear-devel] FlightGear 3.4 RC1 (Windows) feedback.
(powered by Instant-Cquotes) |
|
|
|
FlightGear 3.4 is essentially unusable on a 32 bit windows system outside of very remote/simple areas with no scenery. Should we consider — Curtis Olson (2015-02-04). [Flightgear-devel] FlightGear 3.4 RC1 (Windows) feedback.
(powered by Instant-Cquotes) |
|
|
|
I seem to recall that there was some work on reducing the memory footprint of FlightGear. Did any of those changes make it into the 3.4 branch? — Curtis Olson (2015-02-04). [Flightgear-devel] FlightGear 3.4 RC1 (Windows) feedback.
(powered by Instant-Cquotes) |
|
Updating Dependencies
- Updating the SG/FG source trees for dependencies (JSBSim, OpenSceneGraph etc) should be done right at the beginning of the new release cycle, or at least be closely coordinated.
Bugfix releases
In case a serious bug is found, there should be a process in place to create and distribute a bugfix release as soon as possible.
--Elgaton (talk) 05:31, 16 March 2015 (EDT)
3.20
- The release should not be announced only on the main web page. It should also be announced on the following places:
- The official forum
- The developer list
- The newsletter
- As things turned out this time around, most of our users seem unaware that a new version have been released.
Advice users to delete residual settings from previous versions.
After upgrading to 3.2 on Windows 7, I struggled with strange rendering, misbehaving TerraSync and endless crash on load until I found out that the old 3.0 settings were still there in the Roaming folder. After simply deleting this folder ( C:\Users\USERNAME\AppData\Roaming\flightgear.org\ ), and reinstalling FlightGear, I think, there was never any more trouble and everything worked perfectly. --Jarl Arntzen (talk) 19:03, 22 November 2014 (UTC)
3.00
JSBSim Sync
|
|
One point that might be worth considering is whether we want to update JSBSim now or not. OTOH one consequence of updating is that there might be need to adjust some FDM configurations (I have no clear picture what has changed in JSBSim/CVS since the last update and therefore the size of that risk).
Otherwise we leave the update to post-3.0.0. [1] — Anders Gidenstam
|
|
|
|
I did not see this coming (feature freeze) and, yes, there are some important changes in JSBSim that should be updated. I need to merge some offline changes into JSBSim CVS first, though.[2] — Jon S. Berndt
|
|
|
|
I Guess it's too late to integrate the latest JSBSim code now?[3] — Erik Hofman
|
|
- ↑ Anders Gidenstam (2013-12-17 21:15:58). Release preparations - feature freeze starts today.
- ↑ Jon S. Berndt (2013-12-17 21:15:58). Release preparations - feature freeze starts today.
- ↑ Erik Hofman (2014-01-17 10:27:53). Release preparations: version number is now 3.0.0.
Merge Request Handling
|
|
Fact is, merge requests do get ignored most of the time, unless you know one of the active developers. I had a merge request pending for three months to update my own code in the tree, not touching anything else. It wasn't downright rejected, but was ignored for enough time for me to start working on
something else. Hence I'm not posting any more merge requests and instead I'm keeping my own separate tree (let's not call it fork, since I'm the only
user). Though I admit, sometimes it's a major pain to keep up with changes in the main tree.[1] — Adrian Musceac
|
|
|
|
I can sympathize with the developers. I do know they have their personal agendas, limited time, and a lot to do without having to bother
with external users's whining. On the other hand, I can't be expected to babysit my merge requests too much either since I'm in the same position (I do
contribute to other projects as well, and have my own to contend with). I think the best way would be to asign a developer by drawing straws every
month, and whoever loses gets to look at merge requests on Gitorious and make at least one comment (looks good/ugly code/don't like your nickname etc.).
Having one year old merge requests without even one comment is unacceptable in my opinion, no matter which excuses can be provided.[2] — Adrian Musceac
|
|
|
|
People get busy and miss things. If someone sits quietly for a year on a
commit request, they've got nobody to blame other than themselves.
I agree that some kind of scheduled review process should be put in place
- things shouldn't fall through the cracks if it can be avoided. [3] — GeneB
|
|
|
|
I agree pretty much entirely - not commenting one merge requests for
such a long time
is a Bad Thing, and indeed discourages contribution. I like the idea
of having a rota
of people responsible to at least reviewing merge requests. At
present I'm discouraged
from reviewing merge requests because I don't feel qualified to do so
in many areas,
but I could easily provide at least some useful feedback.[4] — Stuart Buchanan
|
|
- ↑ Adrian Musceac (2013-12-16 06:12:12). osgEarth integration into FG 3.0.
- ↑ Adrian Musceac (2013-12-16 06:12:12). osgEarth integration into FG 3.0.
- ↑ GeneB (2013-12-16 17:25:37). osgEarth integration into FG 3.0.
- ↑ Stuart Buchanan (2013-12-17 21:53:24). osgEarth integration into FG 3.0.
Distribution
- Seed torrents once the release is out
2.12
Release postponed:
Due to real life constraints and the low number of active core developers, we've hit a fluctuation where pretty much everyone is occupied with something else at the moment. As everybody seems to be caught in some real life trouble, we can't see a better way to get the release out than delaying it for a while. Also, it seems that it will take longer than usual to address any issues found in the RCs. Thus, we have agreed to postpone the upcoming FlightGear 2.12 release by a month, to provide sufficient time to handle release candidates and process end-user feedback. [1]
Distribution:
RCs:
- we should try to get out release candidates earlier to give testers a chance to actually run the RCs [4]
|
|
could we generate a full installation RC package for testing? It would make it easier for testers not familiar with Git to use it, and would be quite
handy for people like myself who do their development on Linux, but have Windows systems available for testing but without the git infrastructure or the time to download the entire git fgdata repository.[1] — Stuart Buchanan
|
|
|
|
Jenkins only does what it's told by the scripts (mostly in fgmeta besides the CMake files) - so we're still at the mercy of missing files in the
installer description and so on - I didn't yet automate a 'smoke test'[1] on Jenkins, since that would mean keeping a clean environment to run test
installs, and involve several expensive operations since we'd be launching the sim. That's all doable but requires VMs and more energy than I have. In general
I've been hoping to get enough people using the nightly builds that an automated smoke-test would be unnecessary but that's probably optimistic[2] — James Turner
|
|
|
|
MSVC has a very powerful source view level debugging, but at present this fails in some auto-generated ctor/dtor
code before it reaches 'main()' so can not be used ;=((.
In the Debug build 'new' is replaced with a 'new_dbg' which deliberately fills the allocation with 0xcc...
so if a person does NOT initialize ALL variables simple dtor code like 'if (buf) delete buf;' crashes.[3] — Geoff McLane
|
|
FGData:
- Stuart's PDF generator in bin/makereadmepdf.pl should be run as part of the installer creation process
- Once again, some users were asking for a more lightweight/essential FGData distribution [5]
- Some users reported broken sound configurations for a number of aircraft due to stereo files [6]
- We still keep seeing issues due to aircraft developers contributing resources with files, file names, paths that would break support on OS with case-sensitive OS or a different locale, it would make sense to add some form of automated/scripted validation during startup to detect such issues, without requiring a manual review, possibly through Catalog metadata [7] [8] [9] [10]
Usability/GUI:
- the menubar is increasingly getting a little cluttered, especially the "debug" entry - given the inflexibility of PUI, this also means that there are usability issues, because certain menu items are only accessible with a certain minimal screen resolution, other items cannot be accessed easily. A short term suggestion made, was moving all "Reload X" items into a dedicated "reload" menu, and moving tools to a separate "tools" menu, so help reduce the size of the debug menu [11] [12]
Core:
- A few users reported issues that seemed possibly related to aggressive compiler optimizations, exposing the cmake/build flags via the property tree would seem like a good idea [13]
- Some windows users reported that TerraSync would only work for them in 32-bit compatibility mode [14]
- Providing support to new users would be greatly simplified if we could expose debug related info via the property tree, i.e. to ensure that OSG/SG/FG were not built in DEBUG mode [15]
- We should look into exposing OSG plugin information via the help/about dialog, i.e. writing the info the property tree [16]
- There we still some reports about "SQLite API abuse", presumably due to non-English characters in installation paths,maybe the installer could check paths prior to installation? [17]
- We kept seeing SQLite/NavDB Cache related bug reports [18] ticket #1227
- Like it used to be the case with detailed log files, most end users are usually very willing to help, but unable to provide backtraces - reconsider adding google breakpad support [19]
- under some circumsances, initial navcache/POI processing took up to 10 minutes, even on powerful Linux boxes with plenty of horsepower and RAM using just the default settings [20]
- the way we are incrementing version numbers in SG_SRC/FG_SRC and FGDATA seems to be causing issues ("base package check failed"), which were reported already during the last release cycles (all the way back to 2.8), even one core developer reported the issue during the 2.10 cycle (many people affected by this seem to be using brisa's download & compile script, but it doesn't appear to be Linux-specific, it also seems to happen on Windows). It seems fgdata version is 2.12 while the SG/FG source trees in next were still looking for 2.11 - could be also because of cmake caching, and the cache not being updated properly once certain files are touched in SG/FG source, still investigating... [21] [22]. The problem seems to be this.
- some rendering issues related to OSG 3.1.9 were reported [23] [24] [25] [26]
- it would probably be a good idea to explicitly check for the latest supported OSG version in our SG/FG CMakeLists.txt, to ensure that people do not build FG against OSG versions that are not yet supported, with some option to override/disable the check (for developers) [27] [28]
- while the delayed tile loading is a big improvement in the responsiveness, delayed loading of models caused issues for helicopter pilots, because they could no longer start on buildings - it might be better to make the new behavior property-configurable through a prop switch, so that helicopters could override the setting during startup in their aircraft-set.xml file [29]
Version numbering:
|
|
There was already a couple of people on IRC confused that 2.12 is different to 2.1.2 (since minor version numbers > 9 are something of a rarity in many people's perception).[4] — James Turner
|
|
|
|
Give our release pattern is date scheduled, an Ubuntu style numbering scheme would actually make more sense, but a bit more effort to move too.[5] — James Turner
|
|
|
|
Externally, 3.0 is going to be considered a bigger deal than 2.12.0.[6] — Stuart Buchanan
|
|
|
|
I suggest that we zero-pad the minor release digit after 3.0.0 so we have 3.02, 3.04 etc. to reduce confusion if we reach double digits.[7] — Stuart Buchanan
|
|
|
|
many computer systems sort file names in ascii order, many people don't seem to pay careful attention to where the decimal points are placed, etc.
Once we clear past the 2.10, 2.12, etc. series, I'd like to go back to keeping things single digits in the major and minor version numbers and when we run out of a single digits bump up the major number (so 3.8.x -> 4.0.x). Number are numbers, but this one confused a lot more people than I expected it would or should so maybe it's good to be sensitive to that after we clear the 2.x series of versions.[8] — Curtis Olson
|
|
|
|
That sounds OK to me, as it would imply a full release every 2.5 years, give a clear flag ahead of time when we're nearing a major release, and save having these discussions in the future. For reference, with the current plan there will be 4 years between 2.0.0 (Feb 2010) was and 3.0.0 (Feb 2014). [9] — Stuart Buchanan
|
|
|
|
I've set the version files on 'next' to be 2.99, on the assumption the next release will be 3.0 as discussed.[10] — James Turner
|
|
Backwards Compatibility:
|
|
I think I sensible step in that case is to keep 3.0 as backward compatible (in terms of Aircraft APIs) with 2.12 as possible, which mostly means my resisting the urge to clean up legacy stuff :)
Obviously it's tricky to offer a 100% guarantee, but I don't have anything planned for 3.0 that will require aircraft changes - I'm sure new technologies such as state machines, knob/slider animations and tooltips will mature and gain some new features but hopefully aircraft developers will be able to work against 2.12 with confidence that things will work the same in 3.0[11]— James Turner
|
|
|
|
It's a bit tricky because I haven't had much feedback from aircraft developers about the new APIs (since they aren't in 2.10), but once 2.12 ships we would want to keep them compatible, so fingers crossed the current design is sensible [12] — James Turner
|
|
2.10
!!! NOTE: None of these issues have been incorporated into the release plan yet !!!
- FlightGear Core related :
- a number of users reported segfaults related to the new flight recorder system [30] [31]
- the download & compile script in fgmeta should be updated for each release [32]
It is great news if you are able to crank out full installers right from Jenkins. That will save me a bunch of downloading and hours of uploading for every new release candidate [33].- But it might be a good idea to create a script that will distribute the new builds to the various mirrors. That way I'm less likely to throttle the build server to 10k/sec [34]
- we could also automatically seed them in BitTorrent, on a Linux box and use "btmakemetafile" which I use here to generate those update packages on the tracker [35]
- IMPROVE RELEASE PLAN: perform a sync with JSBSim sources before the feature freeze
[36]. - IMPROVE RELEASE PLAN: decide early on if/when navdata can be updated [37]
- IMPROVE RELEASE PLAN: merge requests that didn't make it into the previous release should probably be handled early during the upcoming release cycle
- IMPROVE RELEASE PLAN: distro-specific repositories should probably be updated, too [38]
- IMPROVE RELEASE PLAN: the "FlightGear & friends" SuseStudio images should probably be also updated for each release cycle [39]
- every now and then, people raise the issue of the major/minor version numbering scheme being a little confusing to people not familiar with software development, thinking that 2.8 must be newer/better/more recent than 2.11 - using code names or release dates instead was suggested [40]
- there are usually reviews posted on blogs, forums etc after each release - we should specifically collect links to those and evaluate all opinions [41] [42]
- IMPROVE RELEASE PLAN: the release plan should be augmented for the sub-release procedures [43]
- there were a number of navcache/SQLite related issues reported via the issue tracker and the forum/devel list [44] [45] [46]
- a little irritation/frustration was caused due to the conflicting review statements concerning the new radio propagation code [47] [48] [49] - some of this boiled down to coding style related issues, highlighting the fact that different core developers have different "coding styles" and requirements when reviewing merge requests, because we still lack an official "FlightGear coding style guide" [50]
- according to Windows users, the installer created by jenkins could use some optimizations [51]
- a number of Windows7/Windows8 users reported issues that needed a "force 32/64 bit" workaround during startup [52][53][54]
- IMPROVE RELEASE PLAN: new GUI widgets, new fgcommands and new Nasal APIs should ideally be documented prior the release, at least through updated README files, preferably also through the wiki
- Better bug reports and troubleshooting:
- show HLA/OpenRTI availability
Fixed since 2.11+ [55] - add a string property with list of startup arguments used by the user, for use in the about dialog
- add a property specifying if the binary is 32/64 bit for use in the about dialog (to check if segfaults are related to 32bit RAM limits)
- add a property specifying the CMAKE_BUILD_TYPE variable used during compilation, for use in the about dialog (debug, release, RelWithDbgInfo) [56]
- show the threading mode in use in the about dialog
- show average frame rate and frame spacing in about dialog
- add a property specifying how much physical RAM is detected (to see if people are running out of RAM) [57]
- is there a portable ARB/WGL extension to determine the amount of dedicated VRAM available ?
- try to detect Intel GPUs and Mesa drivers ? (some of the more common issues were related to Intel/Mesa)
- show HLA/OpenRTI availability
- Changelog / Release Announcement:
- Walking through the list of "lessons learned" as part of the "Upcoming release" announcement was useful [58]
- IMPROVE RELEASE PLAN: Probably it would be even better to directly process all "lessons learned" items and improve the release plan after each release accordingly
- To get to the 3.0 goal sometime in the near future, it's probably a good idea to create a backlog of open items in the wiki and link the release plan document to that. As usual, we don't have to be perfect for a new major release number. But the new features being the reason for the new major number should work reasonably correct. [59] (also see Category:Developer Plans)
- Posting the link to the changelog for the upcoming release helped writing the changelog early, this should also be done for the Hardware Recommendations and Notebooks known to run FlightGear pages probably?
- The changelog can be easily written by using "git log", searching the issue tracker and by going through the last 6 newsletters published since the previous release. It might make sense to explicitly add a "ChangeLog" message to important commits, so that the Changelog can be compiled more easily ?
- Alternatively, request developers to add major changes also to $FG_SRC/ChangeLog again (last updated in 2001)?
- IMPROVE RELEASE PLAN: for the web-based release announcement, it would be helpful to have screen shots or even youtube videos to demonstrate new features - get the community involved EARLY
- IMPROVE RELEASE PLAN: it may make sense to also allow artwork contributors to contribute new splash screen images for use in the upcoming release. The screen shot contest should provide plenty of options [60].
- IMPROVE RELEASE PLAN: a screenshot/banner contest should be held early on, so that we can use the images for our promo work-NOT after the release [61]
- IMPROVE RELEASE PLAN: for the changelog we should early on invite volunteers to help translate it, useful for the release promotion
- IMPROVE RELEASE PLAN: having dedicated promo videos sounds like a good idea , see [62] Howto:Creating FlightGear Promo Videos
- The RC announcement should also contain links to 1) the issue tracker and 2) the RC subforum [63]
- Using wiki tagging, we could ensure that we can also tag our wiki documentation after each release, so that we can provide older versions of our docs for old FG versions [64]
- various files in $FG_ROOT haven't been updated in YEARS, either update them in the future, or just get rid of them: THANKS, NEWS, ChangeLog etc
- Shaders:
- the ATI viewport hack didn't seem enabled in the RCs, as reported by a number of users on the forum [65]
- the ATI viewport hack should only be enabled by default if ATI/AMD hardware is detected [66] [67][68]
- we should probably try to detect if software emulated OpenGL is in use (i.e. using Mesa) and show a corresponding warning [69]
- texture compression should be disabled by default [70] [71] [72] [73] [74]
- lowering the default shader level to 1 improved compatibility for older/underpowered systems [75]
- but there were still many users reporting issues like crashes/segfaults during startup, that seemed affected by changing graphics settings [76]
- we should make sure that the default shader level (and related shaders!) works for all common setups, including ATI/AMD cards (Mac!) and Intel GPUs
- IMPROVE RELEASE PLAN: GLSL shaders and effects should be treated like core code, and should be tested on different platforms before being enabled by default (i.e. signed-off by people using NVIDIA, ATI/AMD, Intel) [77]
- IMPROVE RELEASE PLAN: modified shaders should be tested with other shader-related features to prevent breakage [78] (there might be a way to automate this a litle by catching GLSL compiler errors?)
- to address all the intel GPU related issues, we may want to show an info dialog on computers where /sim/rendering/gl-vendor contains "intel" as a substring and provide an option to disable all shaders [79]
- we probably need a separate article detailing GLSL coding requirements to ensure that portable constructs are used [80] so that problematic shaders are not just identified at the end of the release cycle
- Installer:
- The installer should be updated to show a warning regarding TerraSync update time [81]
- When Flightgear releases a new version, can the staff create a way for the average computer users to install a new version without doing anything to the old version but still use the terrain files from the older version? [82]
- I believe Fred intentionally chose to use the same registry key from one version to the next. Thus if you install a new version over the top of an existing version it will end up in the same path under C:\<PF> [83]
- FGData related (Base Package):
- accessibility of README files in $FG_ROOT/Docs should be improved for our Win/Mac users [84] Fixed since 2.11+ (by Stuart)
- aircraft included in the base package should not require DDS support [85]
- a bunch of Intel GPU related issues were tracked down to be related to texture dimensions beyond 512x512 not being supported, suggested workarounds are mentioned at [86]
- "Funny how the 172P doesn't have it (crash detection via limits.nas). It is something like a default aircraft for the sim, isn't it? Which aircraft are considered the most "finished"?" [87]
- " I'm a bit confused by all the aircraft models in various stages of completion. Even the install package comes with some below-par and alpha stage models. Is there a compiled list of aircraft that are considered "well done"?" [88]
- The default set of airplanes in FG should be the absolute best of the best, simply because that's what a new user is going to be exposed to for their first time. [89]
- Language files should be synced between English and other languages, so translators can work on them before the release ;-)
- the nasal_api_doc.py script in $FG_SRC/scripts/python should be run as part of the release process, to create an updated doc file for $FG_ROOT/Docs and ship it with each release [90]
- IMPROVE RELEASE PLAN: New/updated Nasal scripts contributed to the base package should be checked to properly support important features like simulator reset, this also applies to Nasal scripts used by aircraft, Nasal scripts that fail these criteria, end up breaking existing features! [91] (also see Release:Aircraft Selection Criteria)
- regarding aircraft included in the release: "I must stress usefulness of the Autostart feature, present in most aircraft not running at startup. It keeps frustration away from those who just want to enjoy the flight . (Please note that I actually agree with aircraft being shut down at startup, as long as autostart is present, or the starting procedure is trivially doable by just trying what you see in the cockpit.) " [92] (also see Release:Aircraft Selection Criteria)
- also, it would apparently make sense to provide tutorials for the default aircraft: "At first startup, I noticed the "Need help? use help->tutorials" message, and because I had no idea how to start up the plane (it would be just plain try and fail, than try something else), I did just that and started some basic tutorials. I wouldn't say going through the tutorials was frustrating, but they were quite boring and I was eager to get in the air as soon as possible." [93] (also see Release:Aircraft Selection Criteria)
- "I discovered however, that there can be some problems on Linux about the planes (eg. some versions of the L39 Albatros undergoing several improvements lately). The problems can be caused by Linux being case sensitive about file paths (Windows is not), and I suspect, more models could suffer from some developers not knowing that. It's easy to fix if you know about the problem, but it would better be done on the developer side, as you never know if the smoke is just not implemented or missing due to this. Not to mention how lengthy it would be to go through more aircraft..." [94]
- Docs: Relevant FlightGear paths should ideally not be "hard-coded" in the manual, but rather also configured via the build system, i.e. using cmake, so that the FG/SG cmake configuration can be shared, to automatically update the correct paths without requiring manual maintenance [95]
- accessibility of README files in $FG_ROOT/Docs should be improved for our Win/Mac users [84]
- Usability:
- the huge number of ads placed on the official website, and the non-intuitive layout of the website caused quite some irritation, not only among new users, but also among seasoned long-time contributors - flightgear.org has been repeatedly described as leaving the impression of even being a "scam" [96] [97]
- A little downside is how the FGcom is done as a standalone program just cooperating with FG itself. It took me some fiddling with the settings for about two hours to get it working, but again installation was simply done from repos (FGcom and than FGcomGui as well). [98] (this is planned [99]) Fixed since 2.99+
- Most likely because of the Intel graphics, I suffered for a long time from a problem with aircraft models (and some ground textures too) being black or missing some parts (see my post in an older thread complaining about similar problem). I solved it by adding a command line option turning off texture compression. [100] (also see Release:Aircraft Selection Criteria)
- We should probably add a new menu to the menubar for our online resources (wiki, forum, issue tracker, FAQ) so that people more easily find important resources just by selecting them from a menu.
- Release Candidates:
- a number of users reported crashes, for better debugging support, consider linking in Google BreakPad (cross platform stack traces) [101]
- Release Candidates should probably have a higher default logging level while writing everything to a log file that can be easily shared via the issue tracker/forums, so that end users can provide better bug reports.
- some users reported "OpenGL out of memory" and "out of space" errors when testing the RCs, we may want to link in a leak detector library or simply add BoehmGC - which is used by Mozilla to track leaking subsystems (a runtime log is created and dumped at process termination), that way non-developers could provide better leak reports. [102] [103] [104]
- How about having a test run a week or two in advance, just to make sure we can indeed produce release installers for Win+Mac - and then release the first RC on December 17th/18th or 19th [105]
- We've already got a fairly extensive lead-in time for the release. More testers on more platforms would seem to be the answer. Perhaps we should advertize for testers of those platforms that aren't adequately covered by developers running git? Making a complete package available, not just the binaries would help, as testers wouldn't need to be git-aware. [106]
- IMPROVE RELEASE PLAN: The main area to improve is to distribute release candidates for all platforms earlier - preferably starting immediately after the freeze. That would already give us more time for testing - without extending the actual freeze period. [107]
- IMPROVE RELEASE PLAN: aircraft packages should be prepared prior to the official release date: "For the 2.8 release I didn't start making aircraft download packages (or uploading them to the ftp servers) until after the official release date which was a mistake" [108]
- IMPROVE RELEASE PLAN: RC's should probably be built with Built-in Profiler support enabled [109].
- IMPROVE RELEASE PLAN: When releasing RC's do not limit them to Win/Mac binaries, but also create source snapshots so that distros can already work on the next package versions.
- IMPROVE RELEASE PLAN: For RC's it might make sense to distribute binaries with debugging symbols included and profiling support enabled, so that people can more easily provide useful bug reports, or even backtraces.
- Also, many end users still prefer using the forum for making bug reports and don't use the issue tracker - it might help to add a link (button) to the issue tracker to the about dialog or maybe even directly to the help menu ("Report an issue") (same for wiki/troubleshooting/faq ?)
- IMPROVE RELEASE PLAN: it might make sense to give wider exposure to our RCs, i.e. via the newsletter - possibly by adjusting the release schedule
- actually, it would even seem better to use our Release promotion checklist to send out an "Upcoming Release" announcement 4-6 weeks prior to the actual release, so that all the flightsim websites can notify their users to participate in RC beta testing.
- a number of forum users reported that the RC/release mirrors were a real bottleneck, and that downloading the 800MB RC installer would often take 2-3 hours (using torrents instead was suggested)
- it also seemed that a number of users had issues related to their browser corrupting the huge image download [110] (website should suggest to use a download manager instead!)
- so reducing the size of the installer (i.e. base package) would seem like another good idea to give our RCs wider exposure, for example by focusing only on 2-3 aircraft
- certain reported issues were really tricky to reproduce, we may want to provide an option to export crucial runtime settings to an XML file that can be easily shared with other testers/developers, or even extend the new flight recorder/replay tape system accordingly [111]
- it might be good if the forum release-candidates announcement mentions that tests and bug-reports should be done with a clean install of the release-candidate, with no third-party data used in tests.
- Build related:
- A normal Linux user has practically no chance to get last stable on his box running if it isn't in his distro - a normal Windows user gets everything nice and streamlined. [112]
- According to the issue tracker there were 3-5 different contributors who provided C++ patches that didn't end up reviewed/merged, which caused some irritation.
2.8
- Lack of stress-testing: A number of users reported severe memory growth issues (with fgfs consuming as much as 14gb of RAM), many directly related to new features, such as random buildings: [113] [114] [115] [116] These could have probably been identified earlier by running FG for extended periods of time, and testing the shipped aircraft with the default KSFO scenery, and new features such as random buildings enabled.
- Lack of graceful feature scaling: Several users with old graphics cards reported not being able to run FG 2.8 without crashing during startup, because the FG defaults didn't scale for old hardware [117]
- According to noaa.gov it seems that the NOAA metar source got phased out in 04/2012 and moved to a new URL[118], some users reported issues related to this[119]. However, the metar URL is currently hard-coded in the fgfs/metar source code - in addition, the default format is no longer a plain text dump [120]. It would make sense to make the URL a string property that can be put into preferences.xml and then use a Nasal listener to parse the resulting XML/HTML and set a plain text property instead, that can be processed by the existing metar code.
- Broken OSX downloads
- the OSX 10.8 release and code signing caused some irritation
- After the 2.8 release a number of users on the forums reported seeing GLSL related errors, because some of the 2.8 shaders used GLSL features only supported by more recent GLSL compilers/drivers - it would probably make sense to test all shader settings on all 3 platforms and check if they cause any errors (and "backport" shaders as necessary). Apple/Mac OSX seems to be more problematic
- Microsoft Redistributables were apparently not shipped with the Windows installer ?
- The changelog should be written as early as possible
- The code freeze could probably be lifted for patches that are not normally enabled/used by any default code paths (or shipped aircraft) in a FlightGear release. This probably involves Nasal extension functions, fgcommands, telnet commands, but also custom hard coded instruments or instrumentation-related APIs (think Canvas). Basically, whenever there's no chance to break a release by committing a certain patch, because the code path will not be executed by default without explicitly enabling it. For 2.8, this also meant that the Nasal Canvas API could not be included due to the code freeze, which however wasn't used by any systems or aircraft - so that there would have been zero chance for breakage [121] [122].
- The wiki contains a number of resources to help new users with hardware decisions, such as Hardware Recommendations Notebooks known to run FlightGear and Supported Video Cards - these should probably be updated for each release several weeks in advance.
2.6
- feature freeze in general
- helped a lot during release management. Kept the commit traffic low and thus helped identifying those commits required to pick into the release.
- feature freeze for aircraft
- Technically, a feature freeze for aircraft is not necessary as long as this aircraft is not part of the base distribution and no common parts are affected. If it's guaranteed that the changes remain in FGDATA/Aircraft/MyAircraft and no other files are touched, these updates should be OK up to shortly before the release.
- switching to a new version of supporting libraries like OSG.
- The move to OSG 3.x introduced some major issues. If at all possible, switch to a new library early in the development cycle.
- manual creation of release candidates and the release binaries
- It's preferable to have equal numbers for release candidates for all O/S and probably a git-tag for each candidate.
- release date/time frame
- It took several days to release all the subparts. Might be better to upload all files and pages to hidden folders and publish them all at once (or at least within a couple of hours). That'll have several advantages:
- no big difference between releases for the various OS.
- the website will switch to the new release state quickly. With 2.6.0, the aircraft page was published before the setup. The release announcement was published even later.
- It took several days to release all the subparts. Might be better to upload all files and pages to hidden folders and publish them all at once (or at least within a couple of hours). That'll have several advantages: