Re: (lack of) couchdb windows binaries
On Thu, Apr 1, 2010 at 12:54 PM, Mark Hammond skippy.hamm...@gmail.com wrote: On the other hand though, I *do* see that being able to specify the data directory would be a nice feature - but not a critical one that will impact couchdb adoption on Windows. Are there other directories you are concerned about? The log directory. -juhani
Re: (lack of) couchdb windows binaries
On Wed, Mar 31, 2010 at 3:03 PM, Jan Lehnardt j...@apache.org wrote: 2) Although we support Windows as a target, the current state of both CouchDB and Erlang make it impossible for certain operations to succeed. Most notably compaction (a feature that can be faked with local replication) and the quick succession of deleting and creating databases (which the test suite does, but is a rare production use). Because the two main issues are easy to work around, Windows users are happy to use CouchDB in their environment. I disagree on the easiness of the compaction workaround. It's not useable on a busy db. Please, review and commit COUCHDB-86 asap. Cheers, -juhani
Re: (lack of) couchdb windows binaries
Hello, I have been trying to use the Windows binaries of CouchDB but find that the installer creates issues that never get mentioned. The first concern and one that should be fixed is the distribution and linking of the Erlang binaries in the install. There should always be an option to use the Erlang installation already on the machine. Hard linking the install to the packaged Erlang binaries will almost guarantee non-use of the installer and a subsequent hunt for a way to compile CouchDB seperately. My second concern is the lack of user defined paths for the installation. This also will cause many to uninstall and wait. What should be remembered is that Windows users that are installing CouchDB want the same options that they would get when installing an RDMS. If these are not available then they will move on and never give any input so quality assurance is lost. Hope I did not step on any toes here :) Carl McDade On Thu, Apr 1, 2010 at 12:49 AM, Noah Slater nsla...@me.com wrote: On 31 Mar 2010, at 22:36, Mark Hammond wrote: *sigh* - only a few messages ago in this thread you said: If you're happy preparing the binary, I am happy to call a vote on it. I did my end of the bargain, so would it be possible for you to do yours? No. I know nothing about them, have no way of testing them. Putting my name on them and putting them up for a vote is something that doesn't interest me much. I am struggling enough with my free time at the moment, the last thing I want to do is add yet more work to my plate. * If and only if they are prepared from the source tarball released as 0.11.0 They were - the only complication was that the installer build scripts (ie, the etc/windows directory) is not in the source archive for some reason I don't understand - however, this only impacts the generation of the installer itself, *not* the generation of the couchdb binaries. Is the binary artefact prepared directly from the tarball that is distributed to our users? Or are you preparing it from the source checked out from the repository? -- Carl McDade Webmaster - Drupal developer - PHP programmer Stockholm Sweden Drupal.se Linkedin.com/drupalse
Re: (lack of) couchdb windows binaries
Hi Carl, excellent points. Do you mind opening issues for these points on https://issues.apache.org/jira/browse/COUCHDB so we don't lose track? If you have patches to fix said issues, that's even better :) Cheers Jan -- On 1 Apr 2010, at 11:17, Carl McDade wrote: Hello, I have been trying to use the Windows binaries of CouchDB but find that the installer creates issues that never get mentioned. The first concern and one that should be fixed is the distribution and linking of the Erlang binaries in the install. There should always be an option to use the Erlang installation already on the machine. Hard linking the install to the packaged Erlang binaries will almost guarantee non-use of the installer and a subsequent hunt for a way to compile CouchDB seperately. My second concern is the lack of user defined paths for the installation. This also will cause many to uninstall and wait. What should be remembered is that Windows users that are installing CouchDB want the same options that they would get when installing an RDMS. If these are not available then they will move on and never give any input so quality assurance is lost. Hope I did not step on any toes here :) Carl McDade On Thu, Apr 1, 2010 at 12:49 AM, Noah Slater nsla...@me.com wrote: On 31 Mar 2010, at 22:36, Mark Hammond wrote: *sigh* - only a few messages ago in this thread you said: If you're happy preparing the binary, I am happy to call a vote on it. I did my end of the bargain, so would it be possible for you to do yours? No. I know nothing about them, have no way of testing them. Putting my name on them and putting them up for a vote is something that doesn't interest me much. I am struggling enough with my free time at the moment, the last thing I want to do is add yet more work to my plate. * If and only if they are prepared from the source tarball released as 0.11.0 They were - the only complication was that the installer build scripts (ie, the etc/windows directory) is not in the source archive for some reason I don't understand - however, this only impacts the generation of the installer itself, *not* the generation of the couchdb binaries. Is the binary artefact prepared directly from the tarball that is distributed to our users? Or are you preparing it from the source checked out from the repository? -- Carl McDade Webmaster - Drupal developer - PHP programmer Stockholm Sweden Drupal.se Linkedin.com/drupalse
Re: (lack of) couchdb windows binaries
Sorry, I forgot to mention that I don't think there should be any Official Release of the installer until these issues are fixed. Seperating out some of the parts like the erlang binaries would also remove some of the legal headache involved in the distribution for Windows. In other words while the installer is nice to have, it's just not ready for primetime. On Thu, Apr 1, 2010 at 11:17 AM, Carl McDade carlmcd...@gmail.com wrote: Hello, I have been trying to use the Windows binaries of CouchDB but find that the installer creates issues that never get mentioned. The first concern and one that should be fixed is the distribution and linking of the Erlang binaries in the install. There should always be an option to use the Erlang installation already on the machine. Hard linking the install to the packaged Erlang binaries will almost guarantee non-use of the installer and a subsequent hunt for a way to compile CouchDB seperately. My second concern is the lack of user defined paths for the installation. This also will cause many to uninstall and wait. What should be remembered is that Windows users that are installing CouchDB want the same options that they would get when installing an RDMS. If these are not available then they will move on and never give any input so quality assurance is lost. Hope I did not step on any toes here :) Carl McDade On Thu, Apr 1, 2010 at 12:49 AM, Noah Slater nsla...@me.com wrote: On 31 Mar 2010, at 22:36, Mark Hammond wrote: *sigh* - only a few messages ago in this thread you said: If you're happy preparing the binary, I am happy to call a vote on it. I did my end of the bargain, so would it be possible for you to do yours? No. I know nothing about them, have no way of testing them. Putting my name on them and putting them up for a vote is something that doesn't interest me much. I am struggling enough with my free time at the moment, the last thing I want to do is add yet more work to my plate. * If and only if they are prepared from the source tarball released as 0.11.0 They were - the only complication was that the installer build scripts (ie, the etc/windows directory) is not in the source archive for some reason I don't understand - however, this only impacts the generation of the installer itself, *not* the generation of the couchdb binaries. Is the binary artefact prepared directly from the tarball that is distributed to our users? Or are you preparing it from the source checked out from the repository? -- Carl McDade Webmaster - Drupal developer - PHP programmer Stockholm Sweden Drupal.se Linkedin.com/drupalse -- Carl McDade Webmaster - Drupal developer - PHP programmer Stockholm Sweden Drupal.se Linkedin.com/drupalse
Re: (lack of) couchdb windows binaries
On 1/04/2010 8:17 PM, Carl McDade wrote: Hello, I have been trying to use the Windows binaries of CouchDB but find that the installer creates issues that never get mentioned. The first concern and one that should be fixed is the distribution and linking of the Erlang binaries in the install. There should always be an option to use the Erlang installation already on the machine. Hard linking the install to the packaged Erlang binaries will almost guarantee non-use of the installer and a subsequent hunt for a way to compile CouchDB seperately. I'm afraid I need to disagree here - this issue has been raised so infrequently that I simply can't accept it as fact, especially given the number of happy user reports we have seen. There is no evidence that the way we are packaging will guarantee non-use of the installer - do you have references to anyone else suggesting this is true for them or anyone else? Indeed, I've seen so few reports of Windows users building from source that IMO it is patently false. I personally think our strategy is perfectly reasonable. My experience with many Python based binary releases backs this up - eg, tools such as mercurial, bit-torrent, miro, spambayes, etc are distributed as a binary distribution on Windows and includes the full Python runtime - I'm not aware of any requests for such tools to allow for an already installed Python to be used with a binary. Some people do choose to run from source for various reasons, but the vast majority - even those with Python already installed - are completely happy with the way the binaries work. Finally, providing an all-in-one installer significantly reduces the support burden for the project - there is no chance that user-installed bits and pieces will conflict with the install and cause erroneous error/support requests to be raised. I'm curious - why is this important to you? My second concern is the lack of user defined paths for the installation. This also will cause many to uninstall and wait. The installer allows you to select the path you want to install into, and the normal couchdb mechanisms for overriding individual directories such as the data directory (ie, modifying the .ini files) works perfectly. In this regard I don't see Windows as being at all different than other platforms. On the other hand though, I *do* see that being able to specify the data directory would be a nice feature - but not a critical one that will impact couchdb adoption on Windows. Are there other directories you are concerned about? Maybe you just want this spelt out better in the installer readme? What should be remembered is that Windows users that are installing CouchDB want the same options that they would get when installing an RDMS. If these are not available then they will move on and never give any input so quality assurance is lost. As above, I see no evidence this is true for anyone other than yourself. I understand some of these things might be nice to have, but I would need some evidence before I could accept they are a general concern shared by a significant number of potential users. Maybe you could take this to the -user list and see how many people agree this is critical rather than merely a nice optional feature? Hope I did not step on any toes here :) Not at all - although I simply can't agree with your conclusions :) Cheers, Mark
Re: (lack of) couchdb windows binaries
Jan, Can you point me to the correct source code repository for the installer? Carl On Thu, Apr 1, 2010 at 11:32 AM, Jan Lehnardt j...@apache.org wrote: On 1 Apr 2010, at 11:28, Carl McDade wrote: Sorry, I forgot to mention that I don't think there should be any Official Release of the installer until these issues are fixed. Seperating out some of the parts like the erlang binaries would also remove some of the legal headache involved in the distribution for Windows. We already resolved that there is no legal headache. In other words while the installer is nice to have, it's just not ready for primetime. While I agree this is a nice feature, it's a feature, not a blocker. There are many things in CouchDB that fall in the same category (no auto-compaction comes to mind) and don't stop us from making releases. That said, if someone offers patches to fixes these issues, I don't see a reason why we shouldn't include them. Cheers Jan -- On Thu, Apr 1, 2010 at 11:17 AM, Carl McDade carlmcd...@gmail.com wrote: Hello, I have been trying to use the Windows binaries of CouchDB but find that the installer creates issues that never get mentioned. The first concern and one that should be fixed is the distribution and linking of the Erlang binaries in the install. There should always be an option to use the Erlang installation already on the machine. Hard linking the install to the packaged Erlang binaries will almost guarantee non-use of the installer and a subsequent hunt for a way to compile CouchDB seperately. My second concern is the lack of user defined paths for the installation. This also will cause many to uninstall and wait. What should be remembered is that Windows users that are installing CouchDB want the same options that they would get when installing an RDMS. If these are not available then they will move on and never give any input so quality assurance is lost. Hope I did not step on any toes here :) Carl McDade On Thu, Apr 1, 2010 at 12:49 AM, Noah Slater nsla...@me.com wrote: On 31 Mar 2010, at 22:36, Mark Hammond wrote: *sigh* - only a few messages ago in this thread you said: If you're happy preparing the binary, I am happy to call a vote on it. I did my end of the bargain, so would it be possible for you to do yours? No. I know nothing about them, have no way of testing them. Putting my name on them and putting them up for a vote is something that doesn't interest me much. I am struggling enough with my free time at the moment, the last thing I want to do is add yet more work to my plate. * If and only if they are prepared from the source tarball released as 0.11.0 They were - the only complication was that the installer build scripts (ie, the etc/windows directory) is not in the source archive for some reason I don't understand - however, this only impacts the generation of the installer itself, *not* the generation of the couchdb binaries. Is the binary artefact prepared directly from the tarball that is distributed to our users? Or are you preparing it from the source checked out from the repository? -- Carl McDade Webmaster - Drupal developer - PHP programmer Stockholm Sweden Drupal.se Linkedin.com/drupalse -- Carl McDade Webmaster - Drupal developer - PHP programmer Stockholm Sweden Drupal.se Linkedin.com/drupalse -- Carl McDade Webmaster - Drupal developer - PHP programmer Stockholm Sweden Drupal.se Linkedin.com/drupalse
Re: (lack of) couchdb windows binaries
On Thu, Apr 1, 2010 at 11:54 AM, Mark Hammond skippy.hamm...@gmail.comwrote: On 1/04/2010 8:17 PM, Carl McDade wrote: Hello, I have been trying to use the Windows binaries of CouchDB but find that the installer creates issues that never get mentioned. The first concern and one that should be fixed is the distribution and linking of the Erlang binaries in the install. There should always be an option to use the Erlang installation already on the machine. Hard linking the install to the packaged Erlang binaries will almost guarantee non-use of the installer and a subsequent hunt for a way to compile CouchDB seperately. I'm afraid I need to disagree here - this issue has been raised so infrequently that I simply can't accept it as fact, especially given the number of happy user reports we have seen. There is no evidence that the way we are packaging will guarantee non-use of the installer - do you have references to anyone else suggesting this is true for them or anyone else? Indeed, I've seen so few reports of Windows users building from source that IMO it is patently false. I am going from my experiences in working with Windows 2003 Server as a dev platform. Maybe for a hobbyist or first time user there would be no problem. But most Windows users frequently install Java, Ruby, PHP and Python in seperate directories under Program Files or X:\[SOURCE] and use pointers in configuration of the software that needs the sources. If the program in question does not accept the sources already in place then it becomes a point of frustration because multiple installs of the same source create confusion. I guess there are two camps on this one. One being that the CouchDB windows installer should provide a complete stack One-click install for development environment for Erlang, CouchDB and a Webserver, similar to Ruby One-Click, Wampserver, Xammp etc. But it does not appear to do this yet. So while YAWS, Ejabbard and other software would be running on single instance of Erlang. guaranteeing use of single version, CouchDB might be running on a different version. In my experiments there are collisions in running multiple versions of the Erlang VM, Inets, Mochiweb and some binaries. YAWS for example simply refuses to start while CouchDB (windows installation) is running. It becomes chaotic and inconvenient to fix all the different instances. Windows users love convenience and convention any disturbance in that and they typically drop the software without blogging, bug reporting or emailing. There are exceptions but not many. That being said I would think that it's okay if the Erlang binaries in the CouchDB environment can be used when installing YAWS or something similar. This is the other camp where everything is encapsulated in the CouchDB One-click and there is no need for anything else. I personally think our strategy is perfectly reasonable. My experience with many Python based binary releases backs this up - eg, tools such as mercurial, bit-torrent, miro, spambayes, etc are distributed as a binary distribution on Windows and includes the full Python runtime - I'm not aware of any requests for such tools to allow for an already installed Python to be used with a binary. Some people do choose to run from source for various reasons, but the vast majority - even those with Python already installed - are completely happy with the way the binaries work. Finally, providing an all-in-one installer significantly reduces the support burden for the project - there is no chance that user-installed bits and pieces will conflict with the install and cause erroneous error/support requests to be raised. I'm curious - why is this important to you? Well you know that PHP is popular and the greater number of web developers use Windows as their platform of choice. Since I see that Erlang and CouchDB have a potentially bright futures in web work I would hate to see this ruined with a reputation for not being Windows friendly. PostgreSQL suffered from this for many years. But once a good Windows installer was released it has seen significant improvement in popularity over the last three years. I guess I want to see CouchDB grow without going through the same trials. My second concern is the lack of user defined paths for the installation. This also will cause many to uninstall and wait. The installer allows you to select the path you want to install into, and the normal couchdb mechanisms for overriding individual directories such as the data directory (ie, modifying the .ini files) works perfectly. In this regard I don't see Windows as being at all different than other platforms. On my last try the installer refused to accept any changes to the x:\Program Files\ install directory. I will give it another try and report it as a bug if confirmed. On the other hand though, I *do* see that being able to specify the data directory would be a nice feature - but not a
Re: (lack of) couchdb windows binaries
I tend to agree with you on this. If, for example, you look at Eclipse you can see it's capable of using multiple versions of Java that might be installed on the same box. Many installers bundle their own JRE precisely to insure they get things right. However, the current level of Erlang and CouchDB I'd go with the latter approach for now. It sounds like you have larger fish to fry still. Best, Bob On Apr 1, 2010, at 7:51 AM, Mark Hammond wrote: Just to follow up on a bit of this: On 1/04/2010 10:09 PM, Carl McDade wrote: Wampserver, Xammp etc. But it does not appear to do this yet. So while YAWS, Ejabbard and other software would be running on single instance of Erlang. guaranteeing use of single version, CouchDB might be running on a different version. It appears the ejabberd installer for windows takes the same approach as us - the webpage says The installers contain all the libraries and dependencies needed to run ejabberd and indeed, a copy of the erlang runtime and binaries are installed directly in the ejabberd directory - ie, it appears to not offer installing into an already installed erlang binary distribution either. Cheers, Mark
Re: (lack of) couchdb windows binaries
I could not disagree more about packaging Erlang. I do not know any windows developers who already have Erlang installed! Most users of CouchDB, certainly on Windows, will not have used Erlang before coming to CouchDB. Suggesting that packaging the Erlang binaries with the installer will decrease adoption sounds ludicrous. On Thu, Apr 1, 2010 at 6:57 AM, Robert Dionne dio...@dionne-associates.comwrote: I tend to agree with you on this. If, for example, you look at Eclipse you can see it's capable of using multiple versions of Java that might be installed on the same box. Many installers bundle their own JRE precisely to insure they get things right. However, the current level of Erlang and CouchDB I'd go with the latter approach for now. It sounds like you have larger fish to fry still. Best, Bob On Apr 1, 2010, at 7:51 AM, Mark Hammond wrote: Just to follow up on a bit of this: On 1/04/2010 10:09 PM, Carl McDade wrote: Wampserver, Xammp etc. But it does not appear to do this yet. So while YAWS, Ejabbard and other software would be running on single instance of Erlang. guaranteeing use of single version, CouchDB might be running on a different version. It appears the ejabberd installer for windows takes the same approach as us - the webpage says The installers contain all the libraries and dependencies needed to run ejabberd and indeed, a copy of the erlang runtime and binaries are installed directly in the ejabberd directory - ie, it appears to not offer installing into an already installed erlang binary distribution either. Cheers, Mark
Re: (lack of) couchdb windows binaries
Thank's Mark! Much appreciated. /Per 31 mar 2010 kl. 03.03 skrev Mark Hammond: Great - it can be found at http://people.apache.org/~mhammond/dist/0.11.0/
Re: (lack of) couchdb windows binaries
On 31 Mar 2010, at 02:03, Mark Hammond wrote: Great - it can be found at http://people.apache.org/~mhammond/dist/0.11.0/ Would you like to call the vote on this yourself? If you prepared the Windows artefacts, and called the vote, that should remove the dependancy chain between me and you - as well as speeding things up quite a bit, and taking a little bit of the load of my back. Can I just double check that you prepared this from THE source artefact? Sorry, but this needs to be undertaken by someone who actually believes there is an issue and can articulate it. Not true, it just needs to be done by someone who understands how the package is built. The purpose of legal-discuss is for developers who generally don't know or care about the legal things to get a yea or a nay from people who do. It would be enough simply for you to tell them what you've put into the artefact, and how it's built, and then just ask them for a thumbs up before the vote. To re-itterate, you don't have to think there is a problem, or describe any legal issue. All you have to do is provide a description of how you packaged CouchDB for Windows, and ask them for approval. They may ask you a few technical questions (ones which I could not answer, for example - and I don't fancy playing chinese whispers for people) to get clarification on a few points - but it shouldn't be anything you can't answer. If you're going to be part of the release process here, it would make sense for you to get involved with legal-discuss.
Re: (lack of) couchdb windows binaries
On 31/03/2010 10:09 PM, Jan Lehnardt wrote: I'm on it. Mark, aside from Erlang, Spidermonkey and ICU, are there any other pieces in the installer, like Windows dev-tools/libs? libcurl is still in the build instructions even though I'm not sure it is currently used. The MS C runtime libraries are also included, but they are a requirement for practically all binaries built with the MS compiler, so is almost certainly already distributed with the Apache web server etc... Thanks, Mark Cheers Jan -- On 31 Mar 2010, at 13:00, Noah Slater wrote: On 31 Mar 2010, at 02:03, Mark Hammond wrote: Great - it can be found at http://people.apache.org/~mhammond/dist/0.11.0/ Would you like to call the vote on this yourself? If you prepared the Windows artefacts, and called the vote, that should remove the dependancy chain between me and you - as well as speeding things up quite a bit, and taking a little bit of the load of my back. Can I just double check that you prepared this from THE source artefact? Sorry, but this needs to be undertaken by someone who actually believes there is an issue and can articulate it. Not true, it just needs to be done by someone who understands how the package is built. The purpose of legal-discuss is for developers who generally don't know or care about the legal things to get a yea or a nay from people who do. It would be enough simply for you to tell them what you've put into the artefact, and how it's built, and then just ask them for a thumbs up before the vote. To re-itterate, you don't have to think there is a problem, or describe any legal issue. All you have to do is provide a description of how you packaged CouchDB for Windows, and ask them for approval. They may ask you a few technical questions (ones which I could not answer, for example - and I don't fancy playing chinese whispers for people) to get clarification on a few points - but it shouldn't be anything you can't answer. If you're going to be part of the release process here, it would make sense for you to get involved with legal-discuss.
Re: (lack of) couchdb windows binaries
(Sending again to keep the thread alive, sorry for the mis-reply) -- Okay, I did some research and came across http://www.apache.org/legal/resolved.html#category-b Our dependencies are: - Erlang, http://erlang.org/ License EPL: http://ftp.sunet.se/pub/lang/erlang/EPLICENSE - Spidermonkey, Mozilla's JavaScript engine, http://www.mozilla.org/js/spidermonkey/ License MPL/GPL/LGPL tri-license - ICU, IBM Components for Unicode, http://icu-project.org/ License ICU (MIT like: http://source.icu-project.org/repos/icu/icu/trunk/license.html) And the URL above clearly states that we can ship binaries with these included if we label the binary accordingly. So I think there is no legal issue and we can proceed as planned (see my next mail). I mentioned a potential legal issue to Mark on IRC just because I wasn't sure about the situation but I *was* sure that I wanted to be rather safe than sorry. Cheers Jan -- On 31 Mar 2010, at 13:09, Jan Lehnardt wrote: I'm on it. Mark, aside from Erlang, Spidermonkey and ICU, are there any other pieces in the installer, like Windows dev-tools/libs? Cheers Jan -- On 31 Mar 2010, at 13:00, Noah Slater wrote: On 31 Mar 2010, at 02:03, Mark Hammond wrote: Great - it can be found at http://people.apache.org/~mhammond/dist/0.11.0/ Would you like to call the vote on this yourself? If you prepared the Windows artefacts, and called the vote, that should remove the dependancy chain between me and you - as well as speeding things up quite a bit, and taking a little bit of the load of my back. Can I just double check that you prepared this from THE source artefact? Sorry, but this needs to be undertaken by someone who actually believes there is an issue and can articulate it. Not true, it just needs to be done by someone who understands how the package is built. The purpose of legal-discuss is for developers who generally don't know or care about the legal things to get a yea or a nay from people who do. It would be enough simply for you to tell them what you've put into the artefact, and how it's built, and then just ask them for a thumbs up before the vote. To re-itterate, you don't have to think there is a problem, or describe any legal issue. All you have to do is provide a description of how you packaged CouchDB for Windows, and ask them for approval. They may ask you a few technical questions (ones which I could not answer, for example - and I don't fancy playing chinese whispers for people) to get clarification on a few points - but it shouldn't be anything you can't answer. If you're going to be part of the release process here, it would make sense for you to get involved with legal-discuss.
Re: (lack of) couchdb windows binaries
Hi, I think we have three separate issues that are all entangled in a little mess: 1) Supporting Windows. 2) Making all tests pass on Windows. 3) Making an official CouchDB binary release for Windows. 3.1) Legal issues with said binary release. 1) The 0.11.x tree and the 0.11.0 treat Windows as a first-class target for CouchDB. We have active users on that platform and I don't think it is any question whether we want to keep doing that. Mark has done significant work to resolve any technical issues as well as making the installation a snap (yay Mark!) 2) Although we support Windows as a target, the current state of both CouchDB and Erlang make it impossible for certain operations to succeed. Most notably compaction (a feature that can be faked with local replication) and the quick succession of deleting and creating databases (which the test suite does, but is a rare production use). Because the two main issues are easy to work around, Windows users are happy to use CouchDB in their environment. So far, Mark provided an unofficial installer for CouchDB. Unofficial meaning that there was no vote on dev@ and the installer does not constitute an official Apache release. For the source release, we treat the test-suite as a contract between developers and the release master (and by proxy our users) to ensure all code is fine. In case of Windows *at this point*, that is technically not feasible without patching both CouchDB and Erlang. Patches for both systems exist and the combination is subject to be tested this week. 3) Making an official release for the Windows installer calls for a dev@ vote. Nothing stops us from doing that. 3.1) resolved, see previous emails. -- Action plan. Here's what I say we should do: - Vote on Mark's installer for 0.11.0; on success, release it. - Add big fat warning about the limitations to the downloads page. - Commit COUCHDB-86 to trunk and 0.11.x (after review). - Later, when 0.11.1 is released, we can remove the big fat warning. Alternative action plan: - Review COUCHDB-86 commit it to trunk and 0.11.x. - Bundle 0.11.1 as both a source and Windows binary release and vote on it. - Releaseparty. The first plan gives us an official Windows installer earlier. The second plan would mean we wouldn't get a binary Windows release for 0.11.0 and only 0.11.1. I'm supporting the first. Cheers Jan -- On 31 Mar 2010, at 03:03, Mark Hammond wrote: On 30/03/2010 5:00 PM, Noah Slater wrote: On 30 Mar 2010, at 01:58, Mark Hammond wrote: I understand that - however, the Windows issues are well known, have existed forever and has never before been raised as a blocker for a windows binary. If I knew it would be considered as such I would not have invested any further efforts in Windows binaries until the windows issues were resolved. If you're happy preparing the binary, I am happy to call a vote on it. Great - it can be found at http://people.apache.org/~mhammond/dist/0.11.0/ ... As above, I have no interest in, or knowledge of the issues here, so I will leave the honours to someone who does (or at least someone who has enough grasp of this to consider it a problem.) Actually, I think you're the only person with enough knowledge to handle this. You're not expected to know anything about the law. The purpose of the list is for people with the technical knowledge to ask the people with the legal knowledge what the best way forward is. If you start a thread on legal telling them how the Windows binary is constructed, and asking them if that is okay, that should be all that you have to do. Sorry, but this needs to be undertaken by someone who actually believes there is an issue and can articulate it. This person also needs to understand the couchdb dependencies on any platform (Windows is no different in this regard) and understands the concept of a binary release. While I meet the last 2 criteria, I don't meet the first. So please let me be completely clear and explicitly decline for the 3rd time :) Cheers, Mark
Re: (lack of) couchdb windows binaries
On 31 Mar 2010, at 13:03, Jan Lehnardt wrote: Action plan. Here's what I say we should do: - Vote on Mark's installer for 0.11.0; on success, release it. - Add big fat warning about the limitations to the downloads page. - Commit COUCHDB-86 to trunk and 0.11.x (after review). - Later, when 0.11.1 is released, we can remove the big fat warning. Agreed. Mark, could you call a vote on your 0.11.0 binaries: * If and only if they are prepared from the source tarball released as 0.11.0 * Using my vote calling template The release procedure is: http://wiki.apache.org/couchdb/Release_procedure Please take the time to familiarise yourself with it, even though you will not be doing most of it. Please then create a new section called: Making a Binary Release This should be a top level item, the same as Making a Source Release, at the bottom of the document. Please then document the process you go through, like I have done for the source release. Once you've done that, we can talk about how to announce it and update the site. Thanks, N
Re: (lack of) couchdb windows binaries
On 31/03/2010 11:29 PM, Noah Slater wrote: On 31 Mar 2010, at 13:03, Jan Lehnardt wrote: Action plan. Here's what I say we should do: - Vote on Mark's installer for 0.11.0; on success, release it. - Add big fat warning about the limitations to the downloads page. - Commit COUCHDB-86 to trunk and 0.11.x (after review). - Later, when 0.11.1 is released, we can remove the big fat warning. Agreed. Mark, could you call a vote on your 0.11.0 binaries: *sigh* - only a few messages ago in this thread you said: If you're happy preparing the binary, I am happy to call a vote on it. I did my end of the bargain, so would it be possible for you to do yours? * If and only if they are prepared from the source tarball released as 0.11.0 They were - the only complication was that the installer build scripts (ie, the etc/windows directory) is not in the source archive for some reason I don't understand - however, this only impacts the generation of the installer itself, *not* the generation of the couchdb binaries. Cheers, Mark
Re: (lack of) couchdb windows binaries
On Wed, Mar 31, 2010 at 3:03 PM, Jan Lehnardt j...@apache.org wrote: Hi, I think we have three separate issues that are all entangled in a little mess: 1) Supporting Windows. 2) Making all tests pass on Windows. 3) Making an official CouchDB binary release for Windows. 3.1) Legal issues with said binary release. Agreed. Action plan. Here's what I say we should do: - Vote on Mark's installer for 0.11.0; on success, release it. - Add big fat warning about the limitations to the downloads page. - Commit COUCHDB-86 to trunk and 0.11.x (after review). - Later, when 0.11.1 is released, we can remove the big fat warning. Alternative action plan: - Review COUCHDB-86 commit it to trunk and 0.11.x. - Bundle 0.11.1 as both a source and Windows binary release and vote on it. - Releaseparty. The first plan gives us an official Windows installer earlier. The second plan would mean we wouldn't get a binary Windows release for 0.11.0 and only 0.11.1. I'm supporting the first. I like the second plan more. Mark prepared the installer and it serves well, providing windows users with an easy way to play with 0.11.0. Adding 'official' stamp retrospectively doesn't add much value, IMO. I'd rather focus on squishing COUCHDB-86 and then getting the release right. Also, there's fewer steps before the party. Has it been confirmed that jl/windows-file-share-delete will be in the next erlang release? In the last 'What's cooking in erlang/otp' (2010-03-22) it hadn't yet moved from 'cooking' to 'graduated'. I played a little with 0.11.0 w/ COUCHDB-86 patch applied: - All futon tests, except uuids, pass. - Pull replication didn't work for a large db (2.4M docs, 4.5GB). Couch crashed after 100 000 and 1.7M docs - Etap tests didn't work. Should they? Cheers, -juhani
Re: (lack of) couchdb windows binaries
On 30/03/2010 5:00 PM, Noah Slater wrote: On 30 Mar 2010, at 01:58, Mark Hammond wrote: I understand that - however, the Windows issues are well known, have existed forever and has never before been raised as a blocker for a windows binary. If I knew it would be considered as such I would not have invested any further efforts in Windows binaries until the windows issues were resolved. If you're happy preparing the binary, I am happy to call a vote on it. Great - it can be found at http://people.apache.org/~mhammond/dist/0.11.0/ ... As above, I have no interest in, or knowledge of the issues here, so I will leave the honours to someone who does (or at least someone who has enough grasp of this to consider it a problem.) Actually, I think you're the only person with enough knowledge to handle this. You're not expected to know anything about the law. The purpose of the list is for people with the technical knowledge to ask the people with the legal knowledge what the best way forward is. If you start a thread on legal telling them how the Windows binary is constructed, and asking them if that is okay, that should be all that you have to do. Sorry, but this needs to be undertaken by someone who actually believes there is an issue and can articulate it. This person also needs to understand the couchdb dependencies on any platform (Windows is no different in this regard) and understands the concept of a binary release. While I meet the last 2 criteria, I don't meet the first. So please let me be completely clear and explicitly decline for the 3rd time :) Cheers, Mark
Re: (lack of) couchdb windows binaries
On Wed, Mar 31, 2010 at 4:03 AM, Mark Hammond skippy.hamm...@gmail.com wrote: On 30/03/2010 5:00 PM, Noah Slater wrote: On 30 Mar 2010, at 01:58, Mark Hammond wrote: I understand that - however, the Windows issues are well known, have existed forever and has never before been raised as a blocker for a windows binary. If I knew it would be considered as such I would not have invested any further efforts in Windows binaries until the windows issues were resolved. If you're happy preparing the binary, I am happy to call a vote on it. Great - it can be found at http://people.apache.org/~mhammond/dist/0.11.0/ This is a relief. Thank you. I was rather worried for a while. Providing binaries is a critical for CouchDB adoption on windows. And the build should definitely be part of the release process, IMO. Otherwise there's a risk of getting stuck in a chicken and egg situation: crippling bugs - don't release - not enough users complaining about the bugs - nobody fixes the bugs. Cheers! -juhani
(lack of) couchdb windows binaries
Hi all, Chatting on IRC with a few devs, the following issues were (belatedly) raised regarding Windows support for couchdb - specifically, the lack of a couchdb 0.11 release for Windows. * The test suite does not pass on Windows due to issues compacting and deleting databases and views. There was a suggestion we can't have an official release where the tests fail. While a fix for this is in the pipeline, it isn't in 0.11. While this has been true since 0.10 and all throughout the earlier discussions about an official Windows release, it appears to now be a show-stopper. * I was on 10 days vacation recently. Even though the 0.11 release process took many weeks, I was in email contact for the entire time and I returned from vacation the same day the vote results were announced, it was felt that this was enough to cause Windows to be dropped from the 0.11 voting procedure. Given the delays already encountered in this release, I'm disappointed this position was taken, but cest-la-vie. * The couchdb windows builds includes non ASF binaries - for example, binaries for erlang itself are included, and even though most are built locally as part of the release process, there was a concern about legal issues in providing such binaries. IANAL and have no desire to play on one TV, so I have no capability of either assessing nor resolving this. Jan asked me to send a note to the -dev list on these issues which I am dutifully doing. I don't think there is anything I can personally do to help resolve any these issues, so I hope someone here can help to get a Windows installer back on track. The last issue mentioned is of particular concern to me - if there are any legal issues around this, the same legal issues presumably apply to me personally, which implies it would be prudent for me to stop creating couchdb binary snapshots (and also to remove the old existing ones) until it is resolved. Cheers, Mark
Re: (lack of) couchdb windows binaries
On 30 Mar 2010, at 00:15, Mark Hammond wrote: * The test suite does not pass on Windows due to issues compacting and deleting databases and views. There was a suggestion we can't have an official release where the tests fail. While a fix for this is in the pipeline, it isn't in 0.11. While this has been true since 0.10 and all throughout the earlier discussions about an official Windows release, it appears to now be a show-stopper. The test suite forms a contract between ourselves, and our users. If they are failing, then they are largely useless for that purpose. I regard failing tests to be blockers for the release. If that means that we remove a test until we can get it to pass, then so be it. * I was on 10 days vacation recently. Even though the 0.11 release process took many weeks, I was in email contact for the entire time and I returned from vacation the same day the vote results were announced, it was felt that this was enough to cause Windows to be dropped from the 0.11 voting procedure. Given the delays already encountered in this release, I'm disappointed this position was taken, but cest-la-vie. Delays should not beget delays. I felt that given the circumstances, 0.11 could be the release that lets Windows support simmer in the wild. You will notice that the documentation included now mentions Windows as a first class platform for the software. There are a number of issues with the Windows build that ideally need fixing, and I created a ticket for one of them. It is my hope that by the time of the next release, we may be in the position to include the Windows binaries. * The couchdb windows builds includes non ASF binaries - for example, binaries for erlang itself are included, and even though most are built locally as part of the release process, there was a concern about legal issues in providing such binaries. IANAL and have no desire to play on one TV, so I have no capability of either assessing nor resolving this. This needs to be brought up on legal. Would you do the honours? I don't think there is anything I can personally do to help resolve any these issues, so I hope someone here can help to get a Windows installer back on track. It hasn't faltered! This release is the most Windows-y of them all. We just have to keep on keeping on! Please check JIRA for Windows related bugs, and see how many we can fix before the next release is proposed. I have no problem including the Windows binaries, as long as legal approves. The timing of your holiday was unfortunate, and as I already mentioned, I think a little bit of simmering might help. Feel free to advertise the availability in whatever medium you think will get the biggest audience. The more eyeballs the better. The last issue mentioned is of particular concern to me - if there are any legal issues around this, the same legal issues presumably apply to me personally, which implies it would be prudent for me to stop creating couchdb binary snapshots (and also to remove the old existing ones) until it is resolved. Bring up the issue on legal, and see what they say.
Re: (lack of) couchdb windows binaries
On 30 Mar 2010, at 00:42, Jan Lehnardt wrote: Now since windows support will require a patch to 0.11.0, I'd propose to call the windows release 0.11.1 that we ship along with a source release. What patch does it require?
Re: (lack of) couchdb windows binaries
On 30/03/2010 10:45 AM, Noah Slater wrote: On 30 Mar 2010, at 00:42, Jan Lehnardt wrote: Now since windows support will require a patch to 0.11.0, I'd propose to call the windows release 0.11.1 that we ship along with a source release. I'm not sure require is really the correct word here - people are already using couchdb on Windows with this issue still in place. From the POV of these users, I suspect that making this a requirement which stalls further releases is a step backwards rather than forwards. What patch does it require? The patch in https://issues.apache.org/jira/browse/COUCHDB-86, along with either waiting for the next erlang release, or a hand-patched erlang build. As mentioned though, we *already* have happy Windows users with this issue unresolved - it isn't ideal and should remain a high priority to fix, but it works for them today. Cheers, Mark