Hi Taher, all,we already had several discussions about this pattern and the bad effects we have to suffer from it. As an example, see [1], [2], there are others.
Appealing and discussing does not seem to work for a a few people so I am more and more in favor of a general review-then-commit (RTC) process for the following types of changes:
- architectural changes - new features, modules, plugins - API changes of bigger impact, deprecations- changes to core functionalities like security, request handling, login, etc.
Most people by choice follow the process to propose/discuss a change, provide a patch as an example or base for discussions, work together on this base until there is consensus among the ones interested and only then commit their work to the codebase. This works effectively, reduces noise, the need for reverts or endless discussions about them and so on.
For those who cannot (or don't want to) follow this natural process, we might need to have a more strict policy. An easy policy is that a commit in the above mentioned areas of changes needs a proposed patch and at least 2 independent committers who agree to the changes.
Another policy should be that a commit has to be reverted immediately when at least 2 independent committers ask the committer to do so. The arguments can be discussed after the revert. This prevents delays and effects like we have with e.g. [3] and is easy to revert again if the original commit was correct.
Let's see what others think about it. Regards, Michael Brohl ecomify GmbH www.ecomify.de [1] https://lists.apache.org/thread.html/0b62f4912bc4a340b0526012bb5e1c120dd7dac496ca9b1395d7e13e@%3Cdev.ofbiz.apache.org%3E[2] https://lists.apache.org/thread.html/1cc84ded035247ec439199ae0694d699fe1bcffb5d067c3ae732a0e9@%3Cdev.ofbiz.apache.org%3E
[3] https://issues.apache.org/jira/browse/OFBIZ-9833 Am 21.02.18 um 11:43 schrieb Taher Alkhateeb:
So here is what happened from my view: - you committed something which impacts design without taking it to the community - we objected to the design and suggested to revert and cooperate on a new design - instead you went ahead and immediately committed a new design without asking anyone for reviews Is this the spirit of how community works? Do we just shove in code and then tell people if you don't like it change it? I am assuming that we don't want to turn to strict rules, but rather support the concept of community and team work. Maybe others have differing views so I leave it to everyone to express their views here. On Feb 20, 2018 12:00 AM, "Jacques Le Roux" <[email protected]> wrote:I proposed a 1st patch, we discussed it, it was not OK, I agreed, I implemented the way Taher proposed, we use the RTC, please now review Thanks Jacques Le 19/02/2018 à 20:57, Michael Brohl a écrit :+1 We are discussing this over and over again. I wonder what's so difficult to stick to some basic rules of collaboration. Am 19.02.18 um 20:48 schrieb Taher Alkhateeb:Thank you for the work Jacques. I was hoping as stated earlier that you share the work before committing it since it is an architectural decision that requires community consensus. On Feb 19, 2018 10:29 PM, "Jacques Le Roux" < [email protected]> wrote: Done Jacques Le 18/02/2018 à 20:33, Jacques Le Roux a écrit : Taher,I agree using a property is hackish. I'll try to implement what you suggest using a keep-autologin-cookie webapp attribute which will be false by default and true for the applications mentioned below. I'll check it make sense for webpos before using true there. Thanks Jacques Le 18/02/2018 à 11:43, Taher Alkhateeb a écrit : Hi Jacques,I don't think your proposed solution works either. It seems you might be missing the underlying problem. There are patterns that I see over and over and I wish we can eliminate them, I will explain the general patterns and then a suggestion for reasonable solution: 1- Wrong dependency direction: Framework should never know about core apps. Core apps should never know about plugins. The arrow is always from the outside to the inside. So when you create a solution, it should always be generic and allow "plugging in" values from the outside to the inside, not pulling or coding by hand 2- Global Shared Mutable State: The two keywords here are "shared" and "mutable". The less we have of both, the better. So no, I don't have a problem with having properties, I have a problem with global mutable variables. They make the system much harder to reason about and debug. So it is always better for a global variable to be immutable, and even better, to not exist. However all systems require some level of global variables but we should use them carefully and try to isolate their impact. Essential things like database connections, tomcat sessions, etc ... are reasonable, but going too far might be a problematic architectural design, not a necessity. So in the example we are discussing here, one idea for a solution would be some kind of configuration (perhaps in ofbiz-component.xml) where you declare whatever settings you want, and then the login worker would simply loop over all components to decide. This is a clean (outside-to-inside) solution instead of this hack job. My recommendation is to revert all of this work, discuss a proper solution in the ML, open a JIRA and provide a patch. I also recommend _not_ committing any architectural changes without discussing them in the community first. On Sun, Feb 18, 2018 at 11:36 AM, Jacques Le Roux <[email protected]> wrote: Thanks for the review Taher,Sorry I completely forgot this thread. When I wrote the autoLogoutFromAllBackendSessions() method I let a TODO there // remove all the autoLoginCookies but if in ecommerce/ecomseo and webpos (it's done manually there, not sure for webpos TODO: check) and I remember I thought about harcoding some applications was not good. And especially, how to let know users, who need this feature, to keep the autologin cookie in their custom plugins applications. I know you don't like having too much properties. But to avoid hard wiring these applications in the framework, Isuggest to have a new keep-autologin-cookie security properties, with default OOTB plugins applications # -- Names of the component where the autologin cookie should be kept one year keep-autologin-cookie=ecommerce,ecomseo,webpos This way, users who need to keep the autologin cookie just have to add the concerned application/s to this list. And autoLogoutFromAllBackendSessions() can then use this list to avoid removing autologin cookies from these applications. Would this work for you? As I wrote in the TODO, I'm not sure why webpos have a the autoLogout request-map in its controller. I'm not even sure the webpos really implements it, like ecommerce does. So we could remove webpos from the keep-autologin-cookie list. I'll check that. If it's really needed, as the autologin-cookie feature just remembers you but does not sign you automatically (you need to know the password) I don't think it's a security flaw. Anyway, if it was the only problem with the webpos :/ Jacques Le 12/02/2018 à 19:48, Taher Alkhateeb a écrit : I just checked this code and it looks really worrying to me. You havehard wired the ecommerce component with logic into the heart of the framework, I think we need to review the entire body of work and maybe revert it. On Sat, Feb 10, 2018 at 2:38 PM, Jacques Le Roux <[email protected]> wrote: Le 10/02/2018 à 12:33, Jacques Le Roux a écrit :Hi,Almost 6 years ago OFBIZ-4959 "Logout do not remove autoLogin" was created and I closed as incomplete. Recently while working on OFBIZ-10206 "Security issue in Token Based Authentication" which followed my work in OFBIZ-9833 "Token Based Authentication" I needed a way to get the userLoginId (or userLogin) from the session. But, as explained in OFBIZ-10206, at this stage it was unavailable. So I decided to go with autoLoginCookies. I then " remembered" OFBIZ-4959. So I'd like to commit the patch I provided at OFBIZ-4959. But before that I want to discuss about autoLoginCookies and the feature to be sure we are all on the same field. The auto login feature is used in ecommerce applications (ie OOTB ecommerce and ecomseo) to welcome an user when s/he gets back. It does not really log the user in but eases the login process. From the code, the same feature exists in the webpos, I did not check. AutoLoginCookies are also generated for all applications, but are not used for the auto login feature like in ecommerce applications. It can be nevertheless useful as proves OFBIZ-10206 "Security issue in Token Based Authentication". But for OFBIZ-10206 and security in general it's better to remove the autoLoginCookies of the other applications (ie no ecommerce and webpos) when the user logout. Of course if the user quits the session w/o login out the autoLoginCookies remains so it's best to start with a clean state and remove the autoLoginCookies at start. Without negative opinions I'll commit the OFBIZ-4959.patch in 1 week. Jacques Forgot to say that the autoLoginCookies have a time to live of 1 year.Jacques
smime.p7s
Description: S/MIME Cryptographic Signature
