Skip to contentSkip to conversation

Release method?

DISCUSSION Matthias Saou 4 participants 8 messages Oct 23, 2002

lore
Share & tools
lore
#0 lore raw

Hi,

Seeing how the 0.9.0rc4 release went, I was wondering if the ALSA people wouldn't mind re-thinking the release method?
Apparently, Jaroslav and Takashi are the ones doing most, so this is especially targetted to them.

One suggestion :

- Would it be possible to release some tarballs, supposedly identical to the final ones but a few hours/days before public release and announce? This is because :
1) Many people (including me) would more easily find time to test these then to test the current CVS tree
2) Some problems may show up not in the actual source, but in the way it is packaged (like the automake links)

One question :

- Would it be possible to have a different versioning scheme? For packagers, and I am one, versions like "0.9.0rc4" < "0.9.0" are horrible since they mean that some ugly hacking is needed (namely Epoch: tags for rpm packages). It would have been so much easier if the versions had been 0.8.1, 0.8.2 etc. Especially as nearly everyone uses 0.9.0x anyway, and that 0.5.0 is unmaintained...

Cheers,
Matthias

Signature · 10 lines
Clean custom Red Hat Linux rpm packages : http://freshrpms.net/
Red Hat Linux release 7.3 (Valhalla) running Linux kernel 2.4.18-10acpi
Load : 0.06 0.07 0.09


-------------------------------------------------------
This sf.net email is sponsored by: Influence the future 
of Java(TM) technology. Join the Java Community 
Process(SM) (JCP(SM)) program now. 
http://ads.sourceforge.net/cgi-bin/redirect.pl?sunm0002en
#1 lore raw

At Wed, 23 Oct 2002 17:05:53 +0200,

Quoted 10 lines · Matthias Saou

Hi,

Seeing how the 0.9.0rc4 release went, I was wondering if the ALSA people wouldn't mind re-thinking the release method?
Apparently, Jaroslav and Takashi are the ones doing most, so this is especially targetted to them.

One suggestion :

Quoted 7 lines

- Would it be possible to release some tarballs, supposedly identical to the final ones but a few hours/days before public release and announce? This is because :
1) Many people (including me) would more easily find time to test these then to test the current CVS tree
2) Some problems may show up not in the actual source, but in the way it is packaged (like the automake links)

well, this method would be nice once after 0.9.0-final is out. then we can release a pre-version tarball for tests before the final version comes out, as well as the linux kernel releases.

but, unfortuantely, we are in the stage of rc - that is, basically, each release could be the final one (what an ironical situation...) so, the test for the rc makes no sense from its definition.

anyway, i myself would love to release 0.9.0 final once if most of obvious bugs on rc5 are fixed. but, it's the decision by Jaroslav.

Quoted 8 lines

One question :

- Would it be possible to have a different versioning scheme? For packagers, and I am one, versions like "0.9.0rc4" < "0.9.0" are horrible since they mean that some ugly hacking is needed (namely Epoch: tags for rpm packages). It would have been so much easier if the versions had been 0.8.1, 0.8.2 etc. Especially as nearly everyone uses 0.9.0x anyway, and that 0.5.0 is unmaintained...

basically, this comes also from the same reason: simply we're too long in the rc stage.

yes, it's annyoing. but the similar situation could happen for the version numbers such like "1.0.2a" and "1.0beta5". rpm is not enough clever to check such a word :)

Takashi

-------------------------------------------------------
This sf.net email is sponsored by: Influence the future of Java(TM) technology. Join the Java Community Process(SM) (JCP(SM)) program now.
http://ads.sourceforge.net/cgi-bin/redirect.pl?sunm0002en

#2 lore raw
Quoted 3 lines · Takashi

but, unfortuantely, we are in the stage of rc - that is, basically, each release could be the final one (what an ironical situation...) so, the test for the rc makes no sense from its definition.

Ironical indeed since this has been lasting for some time now :-/ I don't mean to criticize your good work on ALSA or anything else, but "release candidates" represent for me something that all developers agree as what should already be the actual release, but that they're releasing to the public as a "candidate" just to iron out bugs they could have missed. This is not what the latest 0.9.0rc releases seem to me, as ALSA is still evolving much.

I think that these release candidates should have been versioned 0.8.90, 0.8.91 etc. as this is trivial for string comparison.
For developers and some users, all the strings "rc", "pl", "alpha", "beta" seem obvious, but for regular users they're not, nor are they for any type of automated version comparaison.

Quoted 2 lines

anyway, i myself would love to release 0.9.0 final once if most of obvious bugs on rc5 are fixed. but, it's the decision by Jaroslav.

When that will happen, what next? Will be have 1.0alphaX, then 1.0betaX, then 1.0rcX... until 1.0? That is what I dislike, as it's really not harder to use 0.9.X instead, as long as you mention that it's the development branch that will lead to 1.0 and that the current suggested release is 0.9.0.

I'm just suggesting something that I think would make life easier for everyone, and trying to argument my opinion, not trying to force you into deciding something you've probably already thought about. I think I'm just mostly wondering.

Matthias

Signature · 10 lines
Clean custom Red Hat Linux rpm packages : http://freshrpms.net/
Red Hat Linux release 7.3 (Valhalla) running Linux kernel 2.4.18-10acpi
Load : 0.02 0.21 0.27


-------------------------------------------------------
This sf.net email is sponsored by: Influence the future 
of Java(TM) technology. Join the Java Community 
Process(SM) (JCP(SM)) program now. 
http://ads.sourceforge.net/cgi-bin/redirect.pl?sunm0002en
#3 lore raw

I disagree for the moment here Matthias.
In my case I have waited to see what the response was from several people b4 thinking about installing RC4 ... the 0.9 series is after all the 'development release'. I noticed that there were problems and so I tried it to simply help out - to see if I had different errors with my OS / hardware.

One of the more beautiful things about ALSA is the fact that it is not ready to settle ... it is progressive ... that in my opinion is why the API changes, along with other things like module names, etc. I'm sure though that it will and is mature enough to start settling naturally ....

Matt

Quoted 52 lines · Matthias Saou
Quoted 3 lines · Takashi

but, unfortuantely, we are in the stage of rc - that is, basically, each release could be the final one (what an ironical situation...) so, the test for the rc makes no sense from its definition.

Ironical indeed since this has been lasting for some time now :-/ I don't mean to criticize your good work on ALSA or anything else, but "release candidates" represent for me something that all developers agree as what should already be the actual release, but that they're releasing to the public as a "candidate" just to iron out bugs they could have missed. This is not what the latest 0.9.0rc releases seem to me, as ALSA is still evolving much.

I think that these release candidates should have been versioned 0.8.90, 0.8.91 etc. as this is trivial for string comparison.
For developers and some users, all the strings "rc", "pl", "alpha", "beta" seem obvious, but for regular users they're not, nor are they for any type of automated version comparaison.

Quoted 2 lines

anyway, i myself would love to release 0.9.0 final once if most of obvious bugs on rc5 are fixed. but, it's the decision by Jaroslav.

When that will happen, what next? Will be have 1.0alphaX, then 1.0betaX, then 1.0rcX... until 1.0? That is what I dislike, as it's really not harder to use 0.9.X instead, as long as you mention that it's the development branch that will lead to 1.0 and that the current suggested release is 0.9.0.

I'm just suggesting something that I think would make life easier for everyone, and trying to argument my opinion, not trying to force you into deciding something you've probably already thought about. I think I'm just mostly wondering.

Matthias

Signature · 15 lines
Clean custom Red Hat Linux rpm packages : http://freshrpms.net/
Red Hat Linux release 7.3 (Valhalla) running Linux kernel 2.4.18-10acpi
Load : 0.02 0.21 0.27


-------------------------------------------------------
This sf.net email is sponsored by: Influence the future 
of Java(TM) technology. Join the Java Community 
Process(SM) (JCP(SM)) program now. 
http://ads.sourceforge.net/cgi-bin/redirect.pl?sunm0002en

_______________________________________________
Alsa-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/alsa-devel
Signature · 13 lines
http://mffm.darktech.org

WSOLA TimeScale Audio Mod  : http://mffmtimescale.sourceforge.net/
FFTw C++                   : http://mffmfftwrapper.sourceforge.net/
Vector Bass                : http://mffmvectorbass.sourceforge.net/
Multimedia Time Code       : http://mffmtimecode.sourceforge.net/


-------------------------------------------------------
This sf.net email is sponsored by: Influence the future 
of Java(TM) technology. Join the Java Community 
Process(SM) (JCP(SM)) program now. 
http://ads.sourceforge.net/cgi-bin/redirect.pl?sunm0002en
#4 lore raw
Quoted 5 lines · Matt Flax

One of the more beautiful things about ALSA is the fact that it is not ready to settle ... it is progressive ... that in my opinion is why the API changes, along with other things like module names, etc. I'm sure though that it will and is mature enough to start settling naturally ....

And, really, if you look at kernel development, this a very good way to do development - not just while closing in on 1.0.

If we continue comparing to the kernel, biggest problem with ALSA has been the 'app-developers <-> ALSA' interface. While glibc has been the stable wrapper for linux, alsa-lib's public API has been changing as much as the internals - if not even more. ;) But of course, this comparison isn't really fair as glibc mostly implements existing standard interfaces (POSIX, XOPEN), while ALSA's APIs are new ones.

Anyway, my main point is that these two sides of ALSA (the alsa-lib public API, and the rest), should be discussed separately. For the former, changes are highly undesirable, while for the latter, continuous progress is the only way to go.

PS Btw; nothing stops from developing new, better public interfaces
for alsa-lib after 0.9 and 1.0, but these should be developed as alternatives, not breaking the old API.

Signature · 10 lines
 http://www.eca.cx
 Audio software for Linux!



-------------------------------------------------------
This sf.net email is sponsored by: Influence the future 
of Java(TM) technology. Join the Java Community 
Process(SM) (JCP(SM)) program now. 
http://ads.sourceforge.net/cgi-bin/redirect.pl?sunm0002en
#5 lore raw
In reply to Matthias Saou

At Wed, 23 Oct 2002 17:54:46 +0200,

Quoted 21 lines · Matthias Saou
Quoted 3 lines · Takashi

but, unfortuantely, we are in the stage of rc - that is, basically, each release could be the final one (what an ironical situation...) so, the test for the rc makes no sense from its definition.

Ironical indeed since this has been lasting for some time now :-/ I don't mean to criticize your good work on ALSA or anything else, but "release candidates" represent for me something that all developers agree as what should already be the actual release, but that they're releasing to the public as a "candidate" just to iron out bugs they could have missed. This is not what the latest 0.9.0rc releases seem to me, as ALSA is still evolving much.

I think that these release candidates should have been versioned 0.8.90, 0.8.91 etc. as this is trivial for string comparison.
For developers and some users, all the strings "rc", "pl", "alpha", "beta" seem obvious, but for regular users they're not, nor are they for any type of automated version comparaison.

Quoted 8 lines
Quoted 2 lines

anyway, i myself would love to release 0.9.0 final once if most of obvious bugs on rc5 are fixed. but, it's the decision by Jaroslav.

When that will happen, what next? Will be have 1.0alphaX, then 1.0betaX, then 1.0rcX... until 1.0? That is what I dislike, as it's really not harder to use 0.9.X instead, as long as you mention that it's the development branch that will lead to 1.0 and that the current suggested release is 0.9.0.

no, as you can see in the 0.5.x series, the sub-version number will be (hopefully) increased at each release once after 0.9.0 is out. again, the current sitatuion is excpetional.

my understanding of rc-version is exactly what you suggested -- it's nothing but the test-release tarball before the official tarball. (please don't take the example of 0.9.0rcX here. i mean, ideally.)

so, in this matter, 1.0-rcX _should_ come later on, but they must not survive long like 0.9.0rcX, and they are not for normal (end-)users. the people who don't understand what the suffix means should use the version only with numbers.

the question is "beta" version.
here i'm not sure whether the beta suffix is better or not. but "rc" version should be no problem if it's used properly.

Takashi

-------------------------------------------------------
This sf.net email is sponsored by: Influence the future of Java(TM) technology. Join the Java Community Process(SM) (JCP(SM)) program now.
http://ads.sourceforge.net/cgi-bin/redirect.pl?sunm0002en

#6 lore raw
Quoted 4 lines · Takashi

no, as you can see in the 0.5.x series, the sub-version number will be (hopefully) increased at each release once after 0.9.0 is out. again, the current sitatuion is excpetional.

Quoted 8 lines

my understanding of rc-version is exactly what you suggested -- it's nothing but the test-release tarball before the official tarball. (please don't take the example of 0.9.0rcX here. i mean, ideally.)

so, in this matter, 1.0-rcX _should_ come later on, but they must not survive long like 0.9.0rcX, and they are not for normal (end-)users. the people who don't understand what the suffix means should use the version only with numbers.

I get your point. So we both seem to agree that the current situation is rather annoying, but it wasn't supposed to be :-)
I really don't mind the "rcX" or "preX" versions, but these should never be widely released, advertised nor packaged, and a good example would be the Linux kernel itself, where version numbers are clear, although "hidden" to the end users there are all those pre-patches underlying ;-)

I guess we just should hope that from now on branching and releases will be done at the right time to always provide users with a recent recommended version, comprehensively numbered with no "rc", "beta" or whatever tag in it.

Cheers,
Matthias

Signature · 10 lines
Clean custom Red Hat Linux rpm packages : http://freshrpms.net/
Red Hat Linux release 7.3 (Valhalla) running Linux kernel 2.4.18-10acpi
Load : 0.16 0.16 0.17


-------------------------------------------------------
This sf.net email is sponsored by: Influence the future 
of Java(TM) technology. Join the Java Community 
Process(SM) (JCP(SM)) program now. 
http://ads.sourceforge.net/cgi-bin/redirect.pl?sunm0002en
#7 lore raw
In reply to Matthias Saou
Quoted 2 lines · Matthias Saou

Seeing how the 0.9.0rc4 release went, I was wondering if the ALSA people wouldn't mind re-thinking the release method?

Yes, but...

Quoted 2 lines

- Would it be possible to release some tarballs, supposedly identical to the final ones but a few hours/days before public release and announce?

I'd suggest just the opposite. I'd say it's better to follow the Linux-style and release early and _often_. It's just a fact of OSS life, that any release that is a ...

1) CVS- or some other snapshot
2) pre-version
3) beta-version
4) equipped with "development!" warnings

... is only going to be tested by the same limited group of people who have been testing for the last x monts/years. And even those doing testing only report really annoying bugs, as there's "no sense to report little things when testing development releases"...

Only way around is to put out real releases (that are real and look real! :)) and put on the flame suit. Yes, sometimes embarrasing bugs will be found (as has happened numerous times with kernel releases), but this is unavoidable. If something like this happens, you fix it, "make dist" and put out a another release. As a result, development as a whole just goes faster and the developers feel mentally much healthies (no release-pressure).

I got caught into this trap with ecasound (eleven 2.1devX releases during one year, and judging from the download-logs, lots of downloads but done by a the same _very_ small group of people). I feel that ALSA has been caught in this trap, too (0.6 -> 0.9 development has been a bit painful).

And as part of ALSA is driver development, this issue is much more serious. It will never be possible to fully test an ALSA driver release. There's just too many hardware combinations out there. You have to release and wait for the results.

Quoted 2 lines

1) Many people (including me) would more easily find time to test these then to test the current CVS tree

Would there be so many and would it really make the procedure more efficient? For instance, I found and reported the libtool issue quite some time ago, but neither I or the ALSA guys thought it to be something really serious. Well now, there have been so many about this issue that it's clear it has to be handled in some way.

Quoted 2 lines

2) Some problems may show up not in the actual source, but in the way it is packaged (like the automake links)

Although, I admit, this is a very good point. But I'd still just prefer more regular releases. You could compare this to the kernel releases that from time to time have the wrong version number. ;)

Quoted 2 lines

- Would it be possible to have a different versioning scheme? For packagers, and I am one, versions like "0.9.0rc4" < "0.9.0" are horrible

I've tried to get around this by using x.ydevt style versioning (and other experiments), but unfortunately people are used to the x.y.z-{rc,pre,beta}t version numbers. If you use something else, then people won't test the almost stable versions or you confuse users about what is stable and what is development.

Quoted 4 lines

since they mean that some ugly hacking is needed (namely Epoch: tags for rpm packages). It would have been so much easier if the versions had been 0.8.1, 0.8.2 etc. Especially as nearly everyone uses 0.9.0x anyway, and that 0.5.0 is unmaintained...

Yup, I know it's annoying.

Signature · 10 lines
 http://www.eca.cx
 Audio software for Linux!



-------------------------------------------------------
This sf.net email is sponsored by: Influence the future 
of Java(TM) technology. Join the Java Community 
Process(SM) (JCP(SM)) program now. 
http://ads.sourceforge.net/cgi-bin/redirect.pl?sunm0002en