I was wondering, is there a beta test/update quarantine phase for calibre releases?

I think every release should first be made available to people who agree to participate in the beta tests. That way all these regressions would stop making simple folk mad (here i mean my sister, as i am the one who introduced her to calibre so I have to provide her 'customer support' as well :)

Anyway, back to the point, there should be a checkbox in settings that enables beta program participation. if one checks that box then they receive notification of the new release immediately and those who dont check it, get notified after a predefined time period.. say a couple of weeks?

It would also release some of the pain of entirely too frequent updates for the general users and still keep people like me happy, i.e those of us who love to see bugs :P and report on them would still get a healthy dose of freq buggy releases. By Choice. :)

Now back to Singularity Sky :book2:

You can run from source and get stuff hot off the press. That is what many people do, if for no other reason than to avoid downloading the release.

Maybe its just me but the main updates channel feels more like a beta test channel due to such high frequency of updates and almost every other release fixing regressions. I am happy with cutting edge releases but I am sure a large percentage of users are non-tech and would like more stable releases as well as less frequent updates.

I wait a few days before deciding to download a new release to see if any reports of bugs shows up (assuming the release contains a new feature and/or bug fix I'm interested in). I'm watching 0.7.38 right now since it has a bug fix I requested. The only complaint I've seen so far was immediately fixed so it's looking good for me.

Keep in mind this program is still under development and I expect it will be for a long time (which is a good thing since it keeps getting better) so there will be occasional bugs slipping by so those who download new releases are essentially beta testers. Unlike commercial programs that do the same thing (the company that puts out my music notation program is especially notorious for this), fixes for reported bugs are rarely more than a week away instead of only once or twice a year (if that often). The only time I have had to revert to an earlier version because of a bug was with 0.7.35 and I didn't catch it right away (I was probably one of six people affected by it). Fortunately, I keep the previous half dozen or so installation files just in case.

If you don't want to be a "beta tester" for each update, just watch the boards for a few days after a release for any complaints about bugs before updating. Also check the change logs before updating. If there are no fixes or changes of interest to you, there is no point in updating. I usually skip a few updates just for that reason and I don't feel I'm missing out on anything.

With updates coming out once a week or more there is not really time to run a beta phase of the sort you mention. Also with development spread across multiple developers the co-ordination would get much harder. Therefore Kovid acts a a simple sanity check on submitted changes but does not carry out rigorous testing.

As was mentioned, if a bug is found it can be expected to be fixed within a week so that is nor much hardship. The side effect of this "perpetual beta" is that development of new functionality and its availability to end user happens extremely fast.

For those who come across a bug and do not want to revert to a previous release, calibre makes it extremely easy to run from source so that you can pick up the fix as soon as it hits the reposity.

I always tell people that complain about this: If you dislike frequent updates, simply don't update frequently. No one is forcing you to update. If your current calibre installation does all you want, use it happily, turn off the update notifications. There's a checkbox right on the notification itself to turn it off.

Not a complaint. Just a suggestion with a solution in my opinion. You don't have to accept it obviously.

Perhaps you misunderstand me. I don't mind frequent updates. I actually like to, as chaley says, 'get stuff hot off the press'. :)

But I was talking from a point of view of non tech users. Calibre is a very popular software and the targeted users are not technically proficient in general. They are not concerned with the fact that calibre is mostly in beta stage. They get a software that is highly recommended everywhere and expect it to work flawlessly.

It is also a fact that even the general users like to update as soon as an update is available, specially when the software itself tells them so, regardless of whether they need the fixes/features in the update or not.

My suggestion is to add two options to update notifications. One as it is right now and the other after a small time period. Say 3 days or 7 days. There is no need of a rigorous testing phase or beta testers. It would be obvious if any big flaws creep in an update from the immediate updater users.

I think this would greatly improve the perceived stability for such 'general users'.

maybe have the previous release that needed no regression listed first on the download page,as a stable version or something, and have it check only for a new "stable" version, for those end user types, who tend to rant when something breaks.
This would only be updated to the latest version a week or so after a release, as long as no new functionality breaking bugs are found. Most people who know about this forum, or know how to submit tickets etc, would stay with the latest version, updating asap to see new functions etc. Non-Tech types who can barely use the program, and install updates, could be pointed to the "safe" version.

I agree with both Kovid and kevinrs. Please keep the quick series of incremental upgrades coming. I love being on the bleeding edge (I used to run debian sid, back when I actually had time to mess with stuff like that). However there should be a "stable" or "long-term" release that would allow folks to download a known-good version, that could be packaged into linux distros without being immediately out of date, and ideally that would have bug fixes available (for serious bugs only).

On the other hand, release management is costly and time consuming. Regressions in calibre are rarely so bad that you can't live with them for a day or two while the dev team gets out a fix (and they're surprisingly quick at getting out fixes). Naming a "stable" release without a mechanism to maintain that release would do more harm than good, so if it doesn't come along with full release management, maintenance trees, bug fixes on the stable release, etc, then I'd rather it not be done.

Addendum: Calibre is still pre-1.0. Until it hits the big 1.0, maintaining a stable release is probably not a good idea. However as part of getting to 1.0, I think release management and maintenance of the stable branch would have to be seriously considered. I hope development continues apace once 1.0 finally drops, but at the same time that implies the software is mature and thus needs a mechanism of maintaining the 1.0 codebase separate from the subsequent 2.0 work.

Many of my friends are using older versions of calibre as one time or another an upgrade broke something for them then they went back to the version that was working and are unwilling to update as they think the new updates will give them problems.

So I do think a waiting period for general users of a week or so before release should decrease this I am not talking about actually releasing different updates instead just the notice that is shown inside in calibre should be delayed by a week and the latest version should be released maybe on the forum and if any thing major breaks general users are not given the notice that an update is available

Well here's a simple solution -- turn the notification off by default. That way people who don't want to upgrade don't feel like they're obliged to do so just because a new version came out. Those that do can go and turn on the update notifications.

Maybe add the ability to put a priority on releases. That is, if there's a major bug like a data loss issue, the fix for it should override anybody who turned off/left off update notifications and show them the update. For the rest of us who live on the edge, we want all update notifications without having to wait a week or two (contrary to popular belief, most of us don't check the calibre homepage daily looking for a new version).

Well here's a simple solution -- turn the notification off by default.

or define the notification as an integer: "How many releases to wait before notifying", so

0=Do Not Notify,
1=Notify every release
2=Wait for 2 releases
:
5=Wait for 5 releases

etc.

or define the notification as an integer: "How many releases to wait before notifying", so

0=Do Not Notify,
1=Notify every release
2=Wait for 2 releases
:
5=Wait for 5 releases

etc.How is waiting N releases going to help anything? For example, 0.7.36 had several regressions, fixed by 0.7.37. Someone who is waiting two releases would pick up the broken one when 0.7.38 is released. This would not be good.

What this thread is really asking for is for 'someone' to declare a given release stable and suitable for general public. 'Someone' must use intelligence. 'Someone' would need to decide on a release-by-release basis whether that release is suitable for the masses, then somehow propagate that information (how?). The same 'someone' would probably need to redo change logs to redact information about regressions and their repair. And 'someone' would get into the 'known problems' arena, listing new releases with the problem they fix and the problems they create.

Calibre is updated weekly. Things get added, and things get (usually unintentionally) broken. If people don't want to be exposed to that process, then turn off update notification. Personally, I have no objection to following toddos' suggestion: turning it off by default, but that is probably too radical a solution.

Note that we tried a beta program with the 0.6 to 0.7 transition. We had some testers and their input was invaluable, but it was nowhere near sufficient. When 0.7.0 came out there was a firestorm of complaint, mostly justified, sometimes not. A couple of times since I have tried to do betas, with almost no take up. I have found that the only way to get more than a small group of usual suspects to test something is to release it.

You are right, it would be a useful to declare stable releases.

The idea of notifying after N releases would be to reduce notifications for users. They would of course still need to check the release details but it would reduce the frequency of checking.

How about 'Delay n days' after released.
the release folder has a timestamp

n days would give a 'settle' period
n range 0 to 6
0 = Get it now

Strangely enough, the only people that ever request that calibre be released less frequently are the relatively technically proficient users. Most "ordinary" users that I hear from love the fact that calibre is released so frequently. Apparently they feel that someone cares. It seems that we geeks have a slightly distorted picture of what the ordinary user wants. I often get notes from people who send donations and many of those notes mention how appreciative people are of the frequent updates.

My philosophy with updates is to never release an update that could cause data loss without a beta first. Regressions in features that affect limited functionality on the other hand are acceptable, and quickly fixed.

About 60% of downloads (about 80000) of every calibre release are people installing calibre for the first time, given that there are currently 3 million calibre installs, a million of whom have started the calibre GUI at least once in the last 3 months. So I'm not convinced that ordinary users download new releases all the time.

As toddos pointed out, calibre is *vast*. Don't be fooled by the regression lists. It's relatively rare that a regression is very serious or affects a large fraction of users. And in such cases an update is usually released the next day. And that means that about 5k people are affected by the regression.

And I have to agree with chaley. In my experience, betas never have enough coverage to catch all regressions. So I don't see how to run a successful beta program.

Finally, to those people that suggest having updates advertised only every N releases. IMO that would be counter productive. Imagine you are on a release that has regressions, you will now have to wait N weeks instead of 1 week or 1 day for the fix. The point is that for regressions to be found and identified, large uptake is needed, and if update notifications are delayed, the fixes will be delayed and finding the regressions will also be delayed.

...It seems that we geeks have a slightly distorted picture of what the ordinary user wants...

Wow, Kovid! Most geeks (not an insult here) have difficulty understanding that they do not always understand the needs of us "ordinary" users (in my case, computer idiot). My hat's off to you (wait, I'm not wearing one)!

...So I'm not convinced that ordinary users download new releases all the time...

I'm one of the ordinary users who doesn't always download new releases. First, I like to wait a few days to make sure there aren't any bugs before I download. Second, if the new release doesn't have anything that would be useful to me, what would be the point in downloading? With weekly updates, one doesn't have long to wait for an update that will be of interest. I leave the update notification on since I run calibre 24/7 and the notification pops up only when I reboot. It's nice to know a new one is out and it doesn't take all that long to delete the pop-up (1 second; big deal). At least the pop-up isn't an obnoxious ad.

...In my experience, betas never have enough coverage to catch all regressions. So I don't see how to run a successful beta program...

I so totally agree. I see professional programs that do run betas and still don't "get it right." Typically, their mistakes don't get fixed for at least a year, if then. It's impossible to duplicate all hardware combinations nor anticipate how many boneheaded manuevers users may make or how they will use features. Calibre, on the other hand, has a large user base that will report any problems which are, typically, fixed in a week.

Wow, Kovid! Most geeks (not an insult here) have difficulty understanding that they do not always understand the needs of us "ordinary" users (in my case, computer idiot). My hat's off to you (wait, I'm not wearing one)!

Over the years if there's one thing that calibre has taught me, it's that other people (whether geeks or not) think about things in ways that I never would have conceived. Some of them are crazy, some of them are awesome, but they're all different :)

Finally, to those people that suggest having updates advertised only every N releases. IMO that would be counter productive. Imagine you are on a release that has regressions, you will now have to wait N weeks instead of 1 week or 1 day for the fix.

I'm happy with the way it is now. My only suggestion might be that the default be a 12 or 24 hour delay in notification of a new release and provide a tweak to allow for a 0 hour notification.

This might also help the mirrors to handle the load.

Finally, to those people that suggest having updates advertised only every N releases. IMO that would be counter productive. Imagine you are on a release that has regressions, you will now have to wait N weeks instead of 1 week or 1 day for the fix. The point is that for regressions to be found and identified, large uptake is needed, and if update notifications are delayed, the fixes will be delayed and finding the regressions will also be delayed.

Agreed! Scratch my suggestion, it was not a good idea. I generally download each new calibre straight after release and have not yet had to roll one back due to a significant problem. Thanks for all the hard work.