)]}'
{"README.rst":[{"author":{"_account_id":2472,"name":"Doug Hellmann","email":"dhellmann@redhat.com","username":"doug-hellmann"},"change_message_id":"ebe441bd4ebb218b8df92a0f5ad57139ba1599b2","unresolved":false,"context_lines":[{"line_number":243,"context_line":"  testing."},{"line_number":244,"context_line":""},{"line_number":245,"context_line":"  Thus, `!\u003d`, `\u003c` and `\u003c\u003d` specifiers must also be applied to the previous"},{"line_number":246,"context_line":"  release (and transitively backwards to the oldest supported release)."},{"line_number":247,"context_line":""},{"line_number":248,"context_line":"- Libraries should contain a sensible known working minimum version"},{"line_number":249,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"da20952f_19566c06","line":246,"updated":"2015-09-10 20:50:43.000000000","message":"Does that imply that the cap needs to be applied in the stable branch before master? Or just that we would want folks to submit the full set of changes to all branches so they could be approved in a short time frame?","commit_id":"a8b9d916157d30a68c94d5a73635857a37d358b2"},{"author":{"_account_id":4190,"name":"lifeless","email":"robertc@robertcollins.net","username":"lifeless"},"change_message_id":"0e1a26d85f9e2b8690e2ba63091456ad1ee95b1a","unresolved":false,"context_lines":[{"line_number":243,"context_line":"  testing."},{"line_number":244,"context_line":""},{"line_number":245,"context_line":"  Thus, `!\u003d`, `\u003c` and `\u003c\u003d` specifiers must also be applied to the previous"},{"line_number":246,"context_line":"  release (and transitively backwards to the oldest supported release)."},{"line_number":247,"context_line":""},{"line_number":248,"context_line":"- Libraries should contain a sensible known working minimum version"},{"line_number":249,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"da20952f_3d0d578f","line":246,"in_reply_to":"da20952f_19566c06","updated":"2015-09-11 00:03:02.000000000","message":"I think full set would be fine.","commit_id":"a8b9d916157d30a68c94d5a73635857a37d358b2"},{"author":{"_account_id":2750,"name":"Sean Dague","email":"sean@dague.net","username":"sdague"},"change_message_id":"b81a0097aef5644b87c6d87ef1d80bb1a62ca482","unresolved":false,"context_lines":[{"line_number":243,"context_line":"  testing."},{"line_number":244,"context_line":""},{"line_number":245,"context_line":"  Thus, `!\u003d`, `\u003c` and `\u003c\u003d` specifiers must also be applied to the previous"},{"line_number":246,"context_line":"  release (and transitively backwards to the oldest supported release)."},{"line_number":247,"context_line":""},{"line_number":248,"context_line":"- Libraries should contain a sensible known working minimum version"},{"line_number":249,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"da20952f_d925540a","line":246,"in_reply_to":"da20952f_19566c06","updated":"2015-09-10 21:05:40.000000000","message":"honestly, I disagree. stable/kilo has shipped, it\u0027s in the field. Deciding that we cap stable/kilo later only helps for people that upgrade stable/kilo to that point, which might, itself, trigger a downgrade.\n\nAll this rule does is create an unrealistic sandbox environment and stick our fingers in our ears about how people\u0027s environments are going to end up in the field.","commit_id":"a8b9d916157d30a68c94d5a73635857a37d358b2"},{"author":{"_account_id":4190,"name":"lifeless","email":"robertc@robertcollins.net","username":"lifeless"},"change_message_id":"0e1a26d85f9e2b8690e2ba63091456ad1ee95b1a","unresolved":false,"context_lines":[{"line_number":243,"context_line":"  testing."},{"line_number":244,"context_line":""},{"line_number":245,"context_line":"  Thus, `!\u003d`, `\u003c` and `\u003c\u003d` specifiers must also be applied to the previous"},{"line_number":246,"context_line":"  release (and transitively backwards to the oldest supported release)."},{"line_number":247,"context_line":""},{"line_number":248,"context_line":"- Libraries should contain a sensible known working minimum version"},{"line_number":249,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"da20952f_3d6eb700","line":246,"in_reply_to":"da20952f_d925540a","updated":"2015-09-11 00:03:02.000000000","message":"@Sean - certainly, thats one way to argue it. However lets look a bit more closely - consider the pattern from liberty on, where we have published known-good for everything (upper-constraints.txt).\n\nFolk installing can be bucketed:\n - devstack\n - custom PyPI/git based CD deployment\n - binary RPM/dpkg etc deployment\n - other non-PyPI\n\ndevstack will be insulated - if we prevent downgrades, it won\u0027t get into constraints, and devstack in liberty on uses constraints.\n\na custom set of tooling can utilise constraints if they wish - and one of the themes around our dependencies is signalling what works - so I certainly hope folk do use it\n\nbinary RPM etc deployments are fundamentally more curated than our experience w/devstack, but again - the packages look to our requirements files and constraints files for guidance\n\nSo I see this more as avoiding handing out bad advice, not making an unrealistic sandbox.","commit_id":"a8b9d916157d30a68c94d5a73635857a37d358b2"},{"author":{"_account_id":6873,"name":"Matt Riedemann","email":"mriedem.os@gmail.com","username":"mriedem"},"change_message_id":"b6849700cc604ea022105d675240e55b9ca656c7","unresolved":false,"context_lines":[{"line_number":232,"context_line":""},{"line_number":233,"context_line":"  In short, only use version caps to capture known breakage, not to avoid"},{"line_number":234,"context_line":"  possible breakage. The constraints system protects us from accidental"},{"line_number":235,"context_line":"  temporary breaks and gives us early warning of incompatible upstream"},{"line_number":236,"context_line":"  releases."},{"line_number":237,"context_line":""},{"line_number":238,"context_line":"- Version upgrades must never downgrade libraries"}],"source_content_type":"text/x-rst","patch_set":4,"id":"7a2fa921_92b3cd97","line":235,"updated":"2015-10-09 20:50:26.000000000","message":"I guess I\u0027m not familiar with the early warning system, but from a packager perspective (when I was doing that anyway), we\u0027d rely on requirements.txt in a project to determine the versions of dependencies we should be shipping, which meant raising minimums when needed and not going past caps.\n\nWith upper-constraints, those caps are enforced in the openstack CI system but not in requirementst.txt in the projects themselves which is what we\u0027re ultimately installing (not in a venv in a lot of cases).  So if we know that a newer version of a thing will break an openstack project, it seems wrong to not cap it in g-r even if we\u0027re isolated with capping it in upper-constraints.  The recent requests 2.8.0 breakage comes to mind.\n\nAnd I don\u0027t think it\u0027s really fair to packagers/deployers to say, you have to only install in venvs and you have to install using not only requirements.txt but also upper-constraints.txt - which is a thing defined in a separate repo from the actual project you\u0027re installing.\n\nMaybe none of this mini-rant makes sense here though in what you\u0027re trying to describe. :)  I\u0027m having a hard time describing what I think are going to be painful situations that packagers and deployers are going to have to find out the hard way which we\u0027re isolating ourselves from.","commit_id":"c906b812ad5841067db041f1beb3a6b27975c16b"},{"author":{"_account_id":4190,"name":"lifeless","email":"robertc@robertcollins.net","username":"lifeless"},"change_message_id":"69ea8e728003e674b9009a36859de28c974dc513","unresolved":false,"context_lines":[{"line_number":232,"context_line":""},{"line_number":233,"context_line":"  In short, only use version caps to capture known breakage, not to avoid"},{"line_number":234,"context_line":"  possible breakage. The constraints system protects us from accidental"},{"line_number":235,"context_line":"  temporary breaks and gives us early warning of incompatible upstream"},{"line_number":236,"context_line":"  releases."},{"line_number":237,"context_line":""},{"line_number":238,"context_line":"- Version upgrades must never downgrade libraries"}],"source_content_type":"text/x-rst","patch_set":4,"id":"7a2fa921_5ce69517","line":235,"in_reply_to":"7a2fa921_92b3cd97","updated":"2015-10-11 22:00:16.000000000","message":"http://specs.openstack.org/openstack/openstack-specs/specs/requirements-management.html digs into this in more detail, but fundamentally we *can\u0027t* use requirements.txt to list \u0027might be broken\u0027. We can use it to list \u0027known broken\u0027.\n\n !\u003d is a \u0027known broken\u0027 situation.\n \u003c and \u003c\u003d are \u0027might be broken\u0027 statements.\n\nSo when you say \"So if we know that a newer version of a thing will break an openstack project, it seems wrong to not cap it in g-r even if we\u0027re isolated with capping it in upper-constraints. The recent requests 2.8.0 breakage comes to mind.\"  - thats \"known breakage\" not \"possible breakage\" - and we should totally !\u003d 2.8.0.\n\nI am in now way am proposing that packagers or deployers only use venvs.\n\n*however* - requirements.txt with upper caps is pathologically bad: it requires a *release* of the world to lift any cap. That is a tonne of work that has nothing to do with whether those projects needed changing to be compatible. Its reflecting the need to be coinstallable - and thats a centralised project, thus upper-constraints.txt.\n\nPerhaps we should talk on IRC or at the summit - upper-constraints.txt actually is aimed at communicating known-good much much better for packagers - so they shoud have one and only one place to look for what is known to work. If they want to ship older or newer versions of course, then they need their own CI to test it - because requirements.txt is *not* tested across the full set of releases of dependencies (and upper-constraints.txt has in no way changed that testing).","commit_id":"c906b812ad5841067db041f1beb3a6b27975c16b"},{"author":{"_account_id":14288,"name":"Matthew Thode","display_name":"prometheanfire","email":"mthode@mthode.org","username":"prometheanfire"},"change_message_id":"a2b51fa66f63dae67cdc3f558786ae6c006be1b0","unresolved":false,"context_lines":[{"line_number":232,"context_line":""},{"line_number":233,"context_line":"  In short, only use version caps to capture known breakage, not to avoid"},{"line_number":234,"context_line":"  possible breakage. The constraints system protects us from accidental"},{"line_number":235,"context_line":"  temporary breaks and gives us early warning of incompatible upstream"},{"line_number":236,"context_line":"  releases."},{"line_number":237,"context_line":""},{"line_number":238,"context_line":"- Version upgrades must never downgrade libraries"}],"source_content_type":"text/x-rst","patch_set":5,"id":"1a122d0e_555c3f4c","line":235,"updated":"2016-04-30 20:54:39.000000000","message":"I\u0027m not sure what the conclusion was, but library deprication was something that we discussed in a couple of sessions at summit.\n\nhttps://etherpad.openstack.org/p/newton-backwards-compat-libs\n\nI walked in late to another session that was also about this, I think the decision was that libraries should support N-1 release, not sure about implimentation details, but could be based on the review in that etherpad.  https://review.openstack.org/#/c/226157/","commit_id":"02fe101679b7c3721196c95e4d0300c09bd5008f"},{"author":{"_account_id":6873,"name":"Matt Riedemann","email":"mriedem.os@gmail.com","username":"mriedem"},"change_message_id":"8de20ce2d7ba9bee739036ecae62c7ef31e4197a","unresolved":false,"context_lines":[{"line_number":232,"context_line":""},{"line_number":233,"context_line":"  In short, only use version caps to capture known breakage, not to avoid"},{"line_number":234,"context_line":"  possible breakage. The constraints system protects us from accidental"},{"line_number":235,"context_line":"  temporary breaks and gives us early warning of incompatible upstream"},{"line_number":236,"context_line":"  releases."},{"line_number":237,"context_line":""},{"line_number":238,"context_line":"- Version upgrades must never downgrade libraries"}],"source_content_type":"text/x-rst","patch_set":5,"id":"9a8ffd7b_3b6ede96","line":235,"updated":"2015-12-02 03:04:06.000000000","message":"The early warning system is only as good as the test coverage we have, which failed when oslo.messaging 2.6.0 removed some internal modules that neutron was referencing in it\u0027s setup.cfg.  That was fixed in neutron master by changing the neutron setup.cfg to point at the new module names and raising the minimum required oslo.messaging, but that wouldn\u0027t work on stable/liberty because (1) we don\u0027t want to raise the minimum oslo.messaging in stable to fix that and (2) we couldn\u0027t change the neutron setup.cfg in liberty since it wouldn\u0027t work with the minimum version of oslo.messaging for liberty.\n\nSo to get around it, we added a deprecated alias module in oslo.messaging to fix stable/liberty neutron and released that, and then blocked all intermediate versions of oslo.messaging in stable/liberty.  But without capping the version of oslo.messaging in stable/liberty, we can\u0027t remove that deprecated alias in oslo.messaging, event on major version boundaries - which we should be able to do.\n\nOr I should say we can\u0027t do that until liberty-eol, which is a long time considering how often the oslo libraries are released.","commit_id":"02fe101679b7c3721196c95e4d0300c09bd5008f"}]}
